11_chapter6

97

CHAPTER 6

EXPERIMENTAL STUDIES

6.1 OVERVIEW

This chapter provides a detailed description of the both the studies

and their associated results. Initially, the pretests are discussed, followed by a

description of the experiments in both the studies. Finally the results of the

experiments are examined and a discussion of the results is presented.

6.2 PRETESTS

Two pretests were conducted. The first pretest was done to identify

the products that were familiar and of interest to the study population. The

second pretest was done to identify the country that was considered proximal

to the target population. The second pretest was conducted to develop stimuli

for the experiments in Study 1. Both the pretests were conducted on forty four

MBA (first year) students in a large South Indian University. The average age

of the students was 20 and 56% of them were male. The questionnaire Q1 is

shown in Appendix 2.

6.2.1 Product Category

This pretest was conducted to identify products that were relevant

to the target population. A set of 10 products were selected for the pretest.

The selection was based on products used in extant green advertising studies

and popular products that used green advertisements in India (chosen from

98

Table 4.4 in Chapter 4). The ten products were laundry detergent (Schuhwerk

& Lefkoff-Hagius 1995; Kong & Zhang 2013), shampoo (Chang 2011),

mobile phone (Paladino & Ng 2013), mineral water (Grimmer & Woolley

2012), jeans, laptop, skin whiteners, scooter, notebooks and wristwatch.

Consumer involvement scale (Traylor & Joseph 1984) - a six item seven-

point scale that is used to gauge consumers’ involvement across product

categories was used to measure consumer involvement with the selected

products (The scale Q1a is shown in Appendix 2). The results of the pretest

are shown in Table 6.1a. Mobile phones (M=17.45, S.D=7.949) and

wristwatches (M=15.93, S.D=5.699) were ranked high by the consumers.

Table 6.1a Results of pretest for product preferrences

Det

erge

nt

Sham

poo

Mob

ile

Jean

s

Lapt

op

Wat

erBo

ttle

Skin

whi

tene

r

Scoo

ter

Not

eboo

k

Wri

stw

atch

Mean 28.02 26.55 17.45 19.14 18.45 24.93 22.77 19.57 23.50 15.93

N 44 44 44 44 44 44 44 44 44 44

Std. Deviation

8.245 7.866 7.949 7.438 6.670 8.445 8.523 7.053 6.743 5.699

A t-test was also conducted to verify if there was any relationship between

gender and product preferences. The results of the t-tests are shown in Table

6.1b. The results show that there were no gender differences in the product

preferences. Based on this pretest, mobile phones and wristwatches were

chosen as the products for the experiments.

99

Table 6.1b Results of independent samples t-test to test gender

difference in product preferrences

Levene's Test for

Equality of Variances

t-test for Equality of Means

F Sig. t dfSig. *

(2-tailed)

MeanDifference

Std. Error Difference

95% Confidence Interval of

the Difference

Lower UpperDetergent Equal

variances assumed

.076 .784 .238 42 .813 .600 2.524 -4.494 5.694

Equalvariances not assumed

.241 41.889 .811 .600 2.493 -4.431 5.631

Shampoo Equalvariances assumed

1.420 .240 1.446 42 .156 3.400 2.352 -1.347 8.147


1.475 41.946 .148 3.400 2.305 -1.251 8.051

Mobile Equalvariances assumed

.725 .399 .072 42 .943 .175 2.435 -4.739 5.089


.074 40.911 .941 .175 2.359 -4.589 4.939

Jeans Equalvariances assumed

.408 .527 .333 42 .741 .758 2.276 -3.834 5.351


.330 38.783 .743 .758 2.298 -3.890 5.406

Laptop Equalvariances assumed

.444 .509 .333 42 .740 .633 1.899 -3.199 4.466


.326 35.889 .746 .633 1.940 -3.303 4.569

100

Table 6.1b (Continued)

Levene's Test for

Equality of Variances

t-test for Equality of Means

F Sig. t dfSig. *

(2-tailed)

MeanDifference

Std. Error Difference

95% Confidence Interval of

the Difference

Lower UpperWater_Bottle Equal

variances assumed

.250 .619 1.277 42 .209 3.242 2.538 -1.881 8.364


1.284 41.320 .206 3.242 2.525 -1.856 8.339

Skin_whitener Equalvariances assumed

.004 .952 -.616 42 .542 -1.600 2.599 -6.846 3.646


-.616 40.772 .541 -1.600 2.596 -6.844 3.644

Scooter Equalvariances assumed

4.590 .038 .324 42 .747 .700 2.158 -3.655 5.055


.339 38.753 .737 .700 2.068 -3.483 4.883

Notebook Equalvariances assumed

1.025 .317 -.669 42 .507 -1.375 2.055 -5.522 2.772


-.682 41.976 .499 -1.375 2.015 -5.442 2.692

Wristwatch Equal variances assumed

.044 .835 .975 42 .335 1.683 1.727 -1.801 5.168


.963 38.160 .342 1.683 1.748 -1.855 5.221

101

6.2.2 Geographical Proximity of Environmental Issues

Students rated the relevancy of environmental issues based on

geographical (spatial) proximity using seven point scales from ‘relevant to

me’ to ‘irrelevant to me’ (Chang 2012) (The scale Q1b is shown in

Appendix 2). Table 6.2 shows the results of this pretest.

It can be seen that issues related to South India (M=1.93

S.D=1.676) were considered highly relevant when compared to North India

and other countries (China, USA and Australia). Environmental issues in

Australia were considered least important (M=4.30 S.D=1.960).

Table 6.2 Pretest for geographical proximity

China North_India USA South_India AustraliaMean 3.75 2.57 3.98 1.93 4.30

N 44 44 44 44 44

Std. Deviation 1.754 1.485 1.886 1.676 1.960

6.3 STUDY 1: EXPERIMENT 1: TEMPORAL AND

GEOGRAPHICAL FRAMING OF THREAT

This study was conducted to evaluate the effect of temporal and

geographical framing of threat on the PMT variables. The effect of the PMT

variables on involvement and the subsequent influence of involvement on

attitudes and purchase intention was also evaluated. The stimuli were

developed based on the products chosen using the pretests i.e mobile phone

and wristwatch. The experiment was also used to assess the stimuli, content

and face validity of the instrument.

102

6.3.1 Experimental Design

A 2 (temporal proximity of health threat: day vs. year) x 2

(geographical proximity of the health threat: local vs. global) between

subjects experimental design was utilized to investigate the hypotheses. This

resulted in four possible combinations of the factors. Fifty nine valid

responses were obtained from MBA students from a large South Indian

University (39 % male, median age=22). The students were randomly

assigned to the four possible conditions for the mobile phone stimuli.

Similarly, forty one valid responses were obtained from MBA students from a

large South Indian University (62 % male, median age=23). The students

were randomly assigned to the four possible conditions for the watch stimuli.

Data collection was through a paper and pencil questionnaire (Q2

shown in Appendix 3). Students first filled the questionnaire containing the

major dependent variables. Next, they were asked to answer a filler

questionnaire which asked them to describe their favourite celebrity. This was

a filler task designed to distract the respondents from associating the

personality variables question with the next questionnaire. On completion,

they filled counterbalanced questionnaires containing the questions for the

variables related to the environment (environmental concern, environmental

knowledge) and the personality variable (consideration for future

consequences).

6.3.2 Stimuli

A total of four print advertisements were developed for the four

cells: temporally proximal threat and geographically proximal threat;

temporally proximal threat and geographically distant threat; temporally

distant threat and geographically proximal threat; temporally distant threat

103

and geographically distant threat. The advertisements also listed the

environment friendly features of the wristwatch or the mobile phone.

In the temporal threat conditions, the advertisement distinguished

between a day vs. year. Either day or year was used as the reference to specify

the number of people who suffer from respiratory diseases and cancers due to

toxins from either plastic waste (in case of watch) or electronic waste (in case

of mobiles) (Chandran & Menon 2004). In terms of geographical proximity

“India” was used to denote proximity and “world” was used to denote

geographically distant threats (Chang 2012). Although Australia was shown

as the location that is most geographically distant in the pretests, it was not

meaningful to represent an equivalent message that presented a threat in

Australia as part of the stimuli. Hence a more generic “world” was used to

denote a geographically distant threat. The ad was similar to those appearing

in the current Indian print media. The layout and format were not distinct and

contained basic information about the mobile phone or watch. The ad showed

a photograph of the mobile phone/watch and contained a description of the

product features and its environmental attributes. The mobile phone ad

contained a description of the display unit, talk time, standby time, OS and

memory. The mobile phone ad also specified that it contained recyclable

materials and avoided toxic components. The wristwatch ad contained details

about the materials and components used. The ad for the wristwatch also

highlighted its biodegradability.

6.3.3 Treatment Validity

The four print advertisements were analyzed by an expert panel to

assess if it contained the necessary variations in the temporal proximity and

the geographical proximity. This panel consisted of 3 marketing professors

and 2 Phd students who were familiar with marketing literature on fear

104

appeals, PMT and temporal framing. The panel suggested changes to the

presentation format and the final version of the advertisements are shown in

Appendix 4 (Figure A4.1, Figure A4.2, Figure A4.3, Figure A4.4, Figure

A4.5, Figure A4.6, Figure A4.7 and Figure A4.8)

6.3.4 Manipulation Checks

Manipulation checks were conducted by adding questions to verify

if the manipulations were successful. To this end, two questions were

included in the questionnaire. The temporal manipulation was checked asking

the question: “How long do you think it takes for plastic waste pollution to

cause respiratory diseases or cancer?”. The response to this item was

measured using seven point semantic scales anchored from 1 = the near future

and 7 = Distant future. Geographical manipulation was checked asking the

question: “Is the issue of plastic pollution relevant to your country?”. The

response to this item was measured using seven point semantic scales

anchored from 1 = Relevant to my country and 7 = Irrelevant to my country.

6.3.5 Dependent Variables

The study has mostly used previously validated instruments to

measure the constructs. The dependent variables include perceived severity,

perceived vulnerability, perceived self-efficacy, perceived response-efficacy,

fear, message involvement, attitude towards the advertisement, attitude

towards the brand and purchase intention. The other variables related to

individual characteristics included environmental concern, objective

environmental knowledge and consideration for future consequences. The

sources and scales are described in detail below and shown in Appendix 3 as

discussed previously.

105

6.3.5.1 Protection motivation theory variables

The PMT variables were adapted from Milne et al (2002). They are

described in detail below:

Perceived severity

Perceived severity was measured using a four item seven point

scale where 1 = Strongly Disagree and 7 = Strongly Agree. Participants were

asked to indicate their responses on the following statements: “I believe that

plastic waste / e-waste in the environment may cause severe health issues like

respiratory diseases and cancer”, “I believe that plastic waste / e-waste

pollution is a serious threat to human health”, “I do not think that plastic

waste /e-waste will affect our health”, “I believe plastic waste / e-waste

pollution is a significant problem”. In the case of wristwatch “plastic waste”

was used and “e-waste” was used with mobile phone stimuli.

Perceived vulnerability

Perceived vulnerability was measured using a three item seven

point scale where 1 = Strongly Disagree and 7 = Strongly Agree to determine

the participants’ susceptibility to the threat. Participants were asked to

indicate their responses to the following statements: “I am worried that I

might get respiratory illness or cancer because of plastic waste / e-waste”,

“Plastic waste /E-waste pollution is a big concern for me as it might affect my

health”, “It is possible that I am at risk of being affected by respiratory illness

or cancer because of plastic waste / e-waste”.

Response efficacy

Response efficacy was also measured using a three item seven


106

whether participants’ believed if purchasing recyclable / biodegradable

products averted the threat. Participants rated their responses on the following

statements: “Buying biodegradable / recyclable products is highly effective in

preventing diseases due to plastic/e-waste pollution”, “Buying biodegradable /

recyclable products will significantly lower the risk of being affected by

respiratory diseases and cancer”, “Buying biodegradable products is an

effective method of reducing respiratory illness and cancer in humans”. In the

case of mobile phone, the word “biodegradable” was changed to “recyclable”.

Self efficacy

Self efficacy was measured using a three item seven point scale

where 1 = Strongly Disagree and 7 = Strongly Agree to determine whether

participants’ believed if they were capable of averting the threat. Participants

rated their responses to the following statements: “It would be easy for me to

identify a biodegradable watch/ mobile made of recycled materials”, “It is not

difficult for me to check if the watch contains plastic or not / mobile is made

of recycled materials or not”, “I can easily identify a biodegradable watch /

mobile made of recycled materials”.

Fear

Fear is an affective response to the threat levels presented in the

stimuli. Participants rated their emotions (the extent to which they

experienced each of the emotions afraid, scared, fearful, anxious and worried)

while viewing the advertisement on a seven item seven point Likert Scale (1 =

Strongly Disagree and 7 = Strongly Agree). This measure is similar to fear

measures used in previous studies that employ the protection motivation

theory and the scale had a high internal reliability (Hartmann et al 2013).

107

6.3.5.2 Message involvement

Participants reported agreement with six statements (on a seven

point Likert scale) adapted from Cox & Cox (2001) : “I got involved in what

the advertisement had to say,” “The ad's message seemed relevant to me,”

“This ad really made me think”, “This ad was thought-provoking” , “The ad

was very interesting,” and “I felt strong emotions while reading this ad.” This

scale had a good internal reliability score in previous studies (Cox & Cox

2001; Cauberghe et al 2009).

6.3.5.3 Attitudes and Intentions

Attitude towards ad

Attitude towards the ad was measured by using three seven point

semantic differential scales: good/bad, pleasant/unpleasant, and

favorable/unfavorable ( =0.88) (Mackenzie & Lutz 1989).

Attitude towards the brand

Attitude towards the brand was measured by using three seven

point semantic differential scales: good/bad, pleasant/unpleasant, and

favorable/unfavorable ( =0.93) (Muehling & Laczniak 1988).

Purchase intention

Participants were asked to respond to three sets of bipolar

adjectives (unlikely-likely, definitely would-definitely would not, improbable-

probable) placed on seven point scales to indicate how likely they were to

purchase the advertised brand. This scale was also adapted from previous

research (MacKenzie et al 1986).

108

6.3.5.4 Variables related to the environment and personality

Environmental concern

Participants’ environmental concern was measured using the scale

proposed by Schultz (2001). The scale requires the participants to rank their

environmental concerns from one to seven on sub-categories namely

biospheric concerns (plants, marine life, birds, and animals), altruistic

concerns (humanity, children, people in the country, future generations) and

egoistic concerns (me, my future, my health, my lifestyle).

Objective environmental knowledge

Objective environmental knowledge was measured using a set of

fifteen questions similar to the MEAK subscale on environmental knowledge

(Maloney et al 1975). The questions were based on combination of general

questions about environmental awareness (for instance, impact of climate

change, pollutants in batteries and CFLs) and issues specific to India (for

example, Bhopal disaster, maximum greenhouse emissions in India). Some of

the questions were taken from an online quiz (http://edugreen.teri.res.in/

explore/quiz/quiz.htm). The scale is in a quiz format and the correct answers

are summed to form the objective environmental knowledge score. Higher

scores reveal a high degree of factual knowledge about the environment and

vice-versa.

Consideration for future consequences (CFC)

Individual’s temporal orientation was measured using the

consideration of future consequences fourteen item scale (Joireman et al

2012). The scale has two components measuring the concern with future

109

consequences and concern with immediate consequences. The consideration

for future consequences score was determined after recoding the immediate

items (3, 4, 5, 9, 10, 11, 12).

6.3.6 Results of Experiment 1

The experiment was conducted with the mobile phone and

wristwatch stimuli to evaluate the effect of temporal and geographical

framing of threat on the PMT variables. The effect of PMT variables on

involvement and the subsequent influence of involvement on attitudes and

intention was also evaluated. The experiment was also used to assess the

content and face validity of the instrument. The results are discussed below.

6.3.6.1 Manipulation check

Mobile phone stimuli

Temporal proximity manipulations were not successful as there was

no difference in the way participants evaluated the temporal proximity of the

threat. There was no significant main effect of the temporal proximity

manipulation (day vs. year) in both the proximal (M = 3.66) and distal (M =

3.87) conditions (F (1,57) = 0.224 ; p>0.5). The mean values and the ANOVA

tests are shown in Table 6.3a and Table 6.3b.

Table 6.3a shows that the mean values are very close in value. It

can also be seen from Table 6.3b that the temporal proximity did not have any

effect on the manipulation check variable. Geographical manipulations were

checked next.

110

Table 6.3a Experiment 1: Mean values of the manipulation check variable for temporal distance of threat (mobile phone stimuli)

MC_TIME

N Mean Std.Deviation

Std.Error

95%Confidence Interval for

Mean Minimum Maximum

Lower Bound

Upper Bound

Day 29 3.66 1.610 .299 3.04 4.27 1 6Year 30 3.87 1.814 .331 3.19 4.54 1 7Total 59 3.76 1.705 .222 3.32 4.21 1 7

Table 6.3b Experiment 1: One way ANOVA - manipulation check variable for temporal distance of threat (mobile phone stimuli)

ANOVAMC_TIME

Sum of Squares df Mean Square F Sig. Between Groups .660 1 .660 .224 .638Within Groups 168.018 57 2.948Total 168.678 58

Similarly geographical proximity manipulations were also not successful as

there was no significant main effect of the geographical proximity

manipulations as proximity manipulation (India vs. World) did not produce

any statistically significant effect in both the proximal (M = 2.37) and distal

(M = 1.90) conditions (F (1,57) = 2.552 ; p>0.05). The results of the

geographical manipulation checls are shown in Tables 6.4a and 6.4b.

111

Table 6.4a Experiment 1: Mean value of the manipulation check variable for geographical distance of threat (mobile phone stimuli)

Descriptives MC_GEOGRAPHY


Std. Error

95% Confidence Interval for Mean

Minimum MaximumLower Bound

Upper Bound

India 30 2.37 1.299 .237 1.88 2.85 1 6World 29 1.90 .939 .174 1.54 2.25 1 4Total 59 2.14 1.152 .150 1.84 2.44 1 6

Table 6.4b Experiment 1: One way ANOVA of the manipulation check variable for geographical distance of threat (mobile phone stimuli)

ANOVAMC_GEOGRAPHY

Sum of Squares df Mean

Square F Sig.

Between Groups 3.259 1 3.259 2.522 .118Within Groups 73.656 57 1.292Total 76.915 58

Watch stimuli

Temporal proximity manipulations were not successful for the

watch stimuli as there was no significant main effect of the temporal

proximity manipulation (day vs. year) in both the proximal (M = 4.05) and

distal (M = 3.55) conditions (F (1, 40) = 0.990; p>0.05). The results are

shown in Tables 6.5a, 6.5b.

112

Table 6.5a Experiment 1: Mean values of the manipulation check variable for temporal distance of threat (watch stimuli)

Descriptives MC_TIME


Std. Error



Upper Bound

Day 22 4.05 1.704 .363 3.29 4.80 1 7Year 20 3.55 1.504 .336 2.85 4.25 1 7Total 42 3.81 1.612 .249 3.31 4.31 1 7

Table 6.5b Experiment 1: One way ANOVA of the manipulation check variable for temporal distance of threat (watch stimuli)

ANOVA

MC_TIME

Sum of Squares

df Mean

Square F Sig.

Between Groups 2.572 1 2.572 .990 .326

Within Groups 103.905 40 2.598

Total 106.476 41

Similarly geographical proximity manipulations were also not successful as

there was no significant main effect of the geographical proximity

manipulations (India vs. World) in both the proximal (M = 1.96) and distal

(M = 2.24) conditions (F(1,40) = 0.414 ; p>0.05). Tables 6.5c and 6.5d show

these results.

113

Table 6.5c Experiment 1: Mean values of the manipulation check variable for geographical distance of threat (watch stimuli)

Descriptives

MC_GEOGRAPHY

N MeanStd.

DeviationStd.

Error



Upper Bound

India 25 1.96 1.306 .261 1.42 2.50 1 5

World 17 2.24 1.437 .349 1.50 2.97 1 5

Total 42 2.07 1.351 .208 1.65 2.49 1 5

Table 6.5d Experiment 1: One way ANOVA of the manipulation check variable for geographical distance of threat (watch stimuli)

ANOVA

MC_GEOGRAPHY

Sum of Squares

df Mean

Square F Sig.

Between Groups .767 1 .767 .414 .523

Within Groups 74.019 40 1.850

Total 74.786 41

A failed manipulation check in social psychology research is not of

great concern and does not indicate that the manipulation of the independent

variable failed (Sigall & Mills 1998). Therefore further analyses on the data

were conducted.

114

6.3.6.2 Scale reliability


The internal consistency of the scales was assessed using Cronbach

. Table 6.6 below shows the reliability scores. Almost all the constructs

meet and exceed 0.6 – the rule of thumb criteria suggested by Nunnally

(Nunnally 1970) indicating that the instrument is reasonably reliable.

Although self-efficacy has a lower reliability score, most PMT studies report

such low score. Since > 0.5 is acceptable, the same measure was used and

the reliabilities are deemed acceptable. As this study was also used to evaluate

the scales, the scale was retained. Environmental knowledge is treated as a

single formative indicator and therefore reliability score was not calculated

for this measure as it is illogical to check correlations between the indicators

for such a construct (Chin 1998).

Table 6.6 Experiment 1: Reliability scores using mobile phone stimulus

Construct Cronbach Perceived severity 0.64Perceived vulnerability 0.88Response Efficacy 0.82Self Efficacy 0.52Message involvement 0.82Fear 0.89Attitude towards ad 0.81Attitude towards brand 0.90Purchase intention 0.96Environmental Knowledge Environmental concern 0.86Consideration for future consequences 0.85

115

Watch stimuli

Table 6.7 below shows the reliability scores for the watch stimuli.

Table 6.7 Experiment 1: Reliability scores using watch stimulus

Construct Cronbach

Perceived severity 0.57

Perceived vulnerability 0.90

Response Efficacy 0.88

Self Efficacy 0.41

Message involvement 0.82

Fear 0.89

Attitude towards ad 0.82

Attitude towards brand 0.94

Purchase intention 0.93

Environmental concern 0.89

Consideration for future consequences 0.78

It can be seen from Table 6.7 that almost all the constructs meet

and exceed 0.6 Nunnally’s rule of thumb (1970). The instrument is therefore

reasonable reliable. Self-efficacy has a lower reliability score and < 0.5 is

unacceptable. Therefore, this measure was not used for further analysis with

the watch stimulus.

6.3.6.3 Hypotheses tests of the effect of manipulations on PMT

variables

The hypothesized effect of temporal and geographical

manipulations on the PMT variables was analyzed using MANOVA or

MANCOVA as appropriate. After checking for missing data and outliers, it

116

was found that most of the PMT variables were negatively skewed.

MANOVA is “robust to violations of multivariate normality and to violations

of homogeneity of variance/covariance matrices if groups are of nearly equal

size” (Leech et al 2011). The dependent variables (perceived severity,

perceived vulnerability, fear, response-efficacy and self-efficacy) were

moderately correlated (0.27 – 0.49) and therefore there was no risk of

multicollinearity to pose a hindrance to conductiong MANOVA.


Table 6.8a shows the distribution characteristics of the protection

motivation variables and Table 6.8b shows the group wise means. It can be

seen that most of the variables have a mean value that is closer to the highest

score on the scale i.e. 7. Perceived severity ranks high among the threat

appraisal variables with a mean value of 5.91.

Table 6.8a Experiment 1: Distribution characteristics of the protection motivation variables (mobile phone stimuli)

Variable Minimum Maximum Mean Std. DeviationPERC_SEV 4.00 7.00 5.91 0.73PERC_VUL 1.00 7.00 4.68 1.22RESP_EFFICACY 2.00 7.00 4.98 1.25SELF_EFFICACY 3.50 7.00 4.41 1.72FEAR 2.00 6.14 3.60 0.92

Table 6.8b below does not show much variation across the groups

either. The perceived severity and perceived vulnerability scores appear close

in almost all the conditions.

117

Table 6.8b Experiment 1: Group wise mean values of protection motivation variables for the mobile phone stimuli

Factor PerceivedSeverity

PerceivedVulnerability

Response Efficacy

Self Efficacy

Fear

Temporal proximity: Day

6.12 4.93 4.82 4.31 3.70

Temporal proximity: Year

5.72 4.45 5.13 4.52 3.49

Geographical proximity: India

5.83 4.50 5.13 4.74 3.53

Geographical proximity: World

6.00 4.88 4.82 4.08 3.67

Hypothesis 1 stated that consumers who viewed advertisements that

contained threats proximal in time would perceive higher severity and high

vulnerability when compared to consumers who viewed threats that were

distant in time. A one-way MANOVA was conducted to ascertain if there

were significant differences regarding perceived severity and perceived

vulnerability, among the groups in response to manipulation of the temporal

proximity of the threat. Cell sizes were approximately equal (29 and 30) and

the Box's Test indicated that the assumptions of normality were not violated

as there was no significant differences between the covariance matrices. The

one-way MANOVA results were: Pillai’s Trace=0.078; Wilks’ lambda =

0.992; Hotelling’s Trace and Roy’s Largest Root = 0.085, F(2,56)=2.379 as

shown in Table 6.9a. Since the results of the multivariate tests were not

significant, the dependent variables are not significantly dependent on the

temporal proximity of the threat. The results indicate that there was no

statistically significant difference in severity or vulnerability based on

temporal proximity. Therefore hypothesis 1 (H1) was not supported. Table

6.9a and 6.9b shows the results of the test. Although the multivariate tests

were not significant, Table 6.9b showed the possibility of an influence of the

time factor on perceived severity of the threat.

118

Table 6.9a Experiment1: Hypothesis 1: multivariate tests (mobile phone stimuli)

Multivariate Testsb

Effect Value F Hypothesis df

Error df Sig.

Intercept Pillai's Trace .986 1980.986a 2.000 56.000 .000Wilks' Lambda .014 1980.986a 2.000 56.000 .000Hotelling's Trace 70.749 1980.986a 2.000 56.000 .000Roy's Largest Root 70.749 1980.986a 2.000 56.000 .000

Multivariate Testsb


Error df Sig.

TIME_FACTORPillai's Trace .078 2.379a 2.000 56.000 .102Wilks' Lambda .922 2.379a 2.000 56.000 .102Hotelling's Trace .085 2.379a 2.000 56.000 .102Roy's Largest Root .085 2.379a 2.000 56.000 .102

a. Exact statistic b. Design: Intercept + TIME_FACTOR

Table 6.9b Experiment 1: Hypothesis 1: tests of between-subjects effects (mobile phone stimuli)

Tests of Between-Subjects Effects

Source Dependent Variable

Type III Sum of Squares

df MeanSquare F Sig.

Corrected Model PERC_SEV 2.309a 1 2.309 4.500 .038PERC_VUL 3.334b 1 3.334 2.257 .139

Intercept PERC_SEV 2069.131 1 2069.131 4032.657 .000PERC_VUL 1299.221 1 1299.221 879.607 .000

TIME_FACTOR PERC_SEV 2.309 1 2.309 4.500 .038PERC_VUL 3.334 1 3.334 2.257 .139

Error PERC_SEV 29.246 57 .513PERC_VUL 84.192 57 1.477

Total PERC_SEV 2098.938 59PERC_VUL 1384.889 59

Corrected Total PERC_SEV 31.555 58PERC_VUL 87.525 58

a. R Squared = .073 (Adjusted R Squared = .057) b. R Squared = .038 (Adjusted R Squared = .021)

119

A follow up one-way ANOVA was hence conducted to test if the

temporal proximity of the threat had an effect on the perceived severity.

Table 6.9c shows the results of the test. There was a statistically significant

difference between groups as determined by one-way ANOVA

(F(1,57) = 4.500, p <.05). It can be seen from Table 6.9c that temporal

proximity of the threat influenced the perceived severity of the threat. A plot

was producted to check the effect. Figure 6.1 shows that participants who

viewed threats that were closer in time perceived higher levels of severity.

This implies that the temporal proximity of the threat has an effect on the

perceived severity.

Table 6.9c Experiment 1: Tests of between-subjects effects of the temporal proximity of threat on perceived severity (mobile phone stimuli)


Dependent Variable:PERC_SEV

Source Type III Sum of Squares

dfMean

SquareF Sig

Partial Eta

Squared

Noncent Parameter

Observed Powerb

Corrected Model 2309a 1 2.309 4.500 .038 .073 4.500 .550

Intercept 2069.131 1 2069.131 4032.657 .000 .986 4032.657 1.000

TIME_FACTOR 2.309 1 2.309 4.500 .038 .073 4.500 .550

Error 29.246 57 .513

Total 2098.938 59

CorrectedTotal 31.555 58

a.RSquared=.073 (Adjusted R Squared = .057)

b. Computed using alpha = .05

120

Figure 6.1 Experiment 1 – effect of framing a temporally and geographically proximal threat on perceived severity

The main effect of geographical proximity of threat on perceived

severity and vulnerability to the threat was investigated using one-way

MANOVA. Hypothesis 2 (H2) was also not supported as there was no

statistically significant difference in severity and vulnerability based on the

geographical proximity of threat. The one-way MANOVA results were:

Pillai’s Trace=0.027; Wilks’ lambda = 0.876; Hotelling’s Trace and Roy’s

Largest Root = 0.028, F(2,56)=0.787. The results are shown in Table 6.10a

and 6.10b.

121

Table 6.10a Experiment 1: Hypothesis 2: multivariate tests (mobile phone stimuli)

Multivariate Testsb

Effect Value F Hypothesis df Error df Sig.

Intercept Pillai's Trace .985 1861.603a 2.000 56.000 .000

Wilks' Lambda .015 1861.603a 2.000 56.000 .000

Hotelling's Trace 66.486 1861.603a 2.000 56.000 .000

Roy's Largest Root 66.486 1861.603a 2.000 56.000 .000

GEOG_FACTOR Pillai's Trace .027 .787a 2.000 56.000 .460

Wilks' Lambda .973 .787a 2.000 56.000 .460

Hotelling's Trace .028 .787a 2.000 56.000 .460

Roy's Largest Root .028 .787a 2.000 56.000 .460

a. Exact statistic

b. Design: Intercept + GEOG_FACTOR





df Mean

Square F Sig.

Corrected Model PERC_SEV .453a 1 .453 .830 .366

PERC_VUL 2.186b 1 2.186 1.460 .232

Intercept PERC_SEV 2067.826 1 2067.826 3789.661 .000

PERC_VUL 1298.797 1 1298.797 867.497 .000

GEOG_FACTOR PERC_SEV .453 1 .453 .830 .366

PERC_VUL 2.186 1 2.186 1.460 .232

Error PERC_SEV 31.102 57 .546

PERC_VUL 85.339 57 1.497

122


Source Dependent

Variable

Type III

Sum of

Squares

df Mean

Square

F Sig.

Total PERC_SEV 2098.938 59

PERC_VUL 1384.889 59

Corrected Total PERC_SEV 31.555 58

PERC_VUL 87.525 58

a. R Squared = .014 (Adjusted R Squared = -.003)

b. R Squared = .025 (Adjusted R Squared = .008)

Hypothesis 3 predicted interaction effects and stated that an

interaction between temporal proximity and geographical proximity would

cause perception of higher levels of severity and vulnerability under proximal

conditions. A 2 (temporal proximity of threat:day vs.year) x 2 (geographical

proximity of threat: India vs. World) multivariate analysis of variance

(MANOVA) was conducted to examine this. The results also indicated that no

interaction effect exists between temporal proximity of threat and

geographical proximity of threat on both the PMT variables (Pillai’s

Trace=0.016; Wilks’ lambda = 0.984; Hotelling’s Trace and Roy’s Largest

Root = 0.017, F(2,54)=0.450). The results are shown in Table 6.11a and Table

6.11b. Therefore, Hypothesis 3 was not accepted.

123


Multivariate Tests b

Effect Value FHypothesis

df Error

df Sig.

Intercept Pillai's Trace .986 1972.206a 2.000 54.000 .000Wilks' Lambda

.014 1972.206a 2.000 54.000 .000

Hotelling's Trace

73.045 1972.206a 2.000 54.000 .000

Roy's Largest Root

73.045 1972.206a 2.000 54.000 .000

TIME_FACTOR Pillai's Trace .082 2.398a 2.000 54.000 .101Wilks' Lambda

.918 2.398a 2.000 54.000 .101

Hotelling's Trace

.089 2.398a 2.000 54.000 .101

Roy's Largest Root

.089 2.398a 2.000 54.000 .101

GEOG_FACTOR Pillai's Trace .030 .846a 2.000 54.000 .435Wilks' Lambda

.970 .846a 2.000 54.000 .435

Hotelling's Trace

.031 .846a 2.000 54.000 .435

Roy's Largest Root

.031 .846a 2.000 54.000 .435

TIME_FACTOR * GEOG_FACTOR

Pillai's Trace .016 .450a 2.000 54.000 .640Wilks' Lambda

.984 .450a 2.000 54.000 .640

Hotelling's Trace

.017 .450a 2.000 54.000 .640

Roy's Largest Root

.017 .450a 2.000 54.000 .640

a. Exact statistic b. Design: Intercept + TIME_FACTOR + GEOG_FACTOR + TIME_FACTOR * GEOG_FACTOR

124





df Mean

Square F Sig.

Corrected Model PERC_SEV 3.228a 3 1.076 2.089 .112PERC_VUL 6.267b 3 2.089 1.414 .249


TIME_FACTOR PERC_SEV 2.309 1 2.309 4.483 .039PERC_VUL 3.375 1 3.375 2.284 .136

GEOG_FACTOR PERC_SEV .473 1 .473 .919 .342PERC_VUL 2.238 1 2.238 1.515 .224


PERC_SEV .430 1 .430 .835 .365PERC_VUL .652 1 .652 .441 .509





To test H4, a one way MANCOVA with perceived severity and

perceived vulnerability as dependent variable and CFC as the covariate was

conducted. The temporal proximity of the threat was the independent variable.

The assumptions for MANCOVA were met. In particular, the homogeneity of

the regression effect was evident for the covariate, and the covariate was

linearly related to the dependent measure. The one-way MANCOVA results

were as follows: (Pillai’s Trace=0.064; Wilks’ lambda = 0.936; Hotelling’s

Trace and Roy’s Largest Root = 0.068, F(2,54)=1.846). There were no

125

interaction effects and therefore H4 was not supported. The results are shown

in Table 6.12a and 6.12b below. It can be seen from Table 6.12a that the

temporal proximity of the threat did not have a significant effect on the

hypothesized variables. Hence Table 6.12b was not interpreted.


Multivariate Testsb


Errordf Sig.

Intercept Pillai's Trace .764 87.364a 2.000 54.000 .000Wilks' Lambda

.236 87.364a 2.000 54.000 .000

Hotelling's Trace

3.236 87.364a 2.000 54.000 .000

Roy's Largest Root

3.236 87.364a 2.000 54.000 .000

TIME_FACTOR Pillai's Trace .087 2.585a 2.000 54.000 .085Wilks' Lambda

.913 2.585a 2.000 54.000 .085

Hotelling's Trace

.096 2.585a 2.000 54.000 .085

Roy's Largest Root

.096 2.585a 2.000 54.000 .085

TIME_FACTOR * CFC_TOTAL

Pillai's Trace .064 1.846a 2.000 54.000 .168Wilks' Lambda

.936 1.846a 2.000 54.000 .168

Hotelling's Trace

.068 1.846a 2.000 54.000 .168

Roy's Largest Root

.068 1.846a 2.000 54.000 .168

CFC_TOTAL Pillai's Trace .097 2.902a 2.000 54.000 .063Wilks' Lambda

.903 2.902a 2.000 54.000 .063

Hotelling's Trace

.107 2.902a 2.000 54.000 .063

Roy's Largest Root

.107 2.902a 2.000 54.000 .063

a. Exact statistic b. Design: Intercept + TIME_FACTOR + TIME_FACTOR * CFC_TOTAL + CFC_TOTAL

126

Table 6.12b Experiment 1: Hypothesis 4: tests of between-subjects effects




df Mean

Square F Sig.

Corrected Model PERC_SEV 5.611a 3 1.870 3.965 .012

PERC_VUL 15.982b 3 5.327 4.096 .011


PERC_VUL 34.893 1 34.893 26.824 .000

TIME_FACTOR PERC_SEV 1.585 1 1.585 3.360 .072

PERC_VUL 5.090 1 5.090 3.913 .053


PERC_SEV .984 1 .984 2.086 .154

PERC_VUL 3.981 1 3.981 3.060 .086

CFC_TOTAL PERC_SEV 1.640 1 1.640 3.476 .068

PERC_VUL 6.052 1 6.052 4.653 .035


PERC_VUL 71.543 55 1.301


PERC_VUL 1384.889 59


PERC_VUL 87.525 58


Since the hypothesis testing did not yield any specific results, a

three way ANOVA was conducted to examine the interactions among the

factors and covariates on the individual dependent variables (perceived

severity and perceived vulnerability). Tables 6.13a and 6.13b show the result.

The results showed significant interactions of the factors and the covariate

CFC on perceived severity and perceived vulnerability of the threat.

127

Table 6.13a Experiment 1: Tests of between-subjects effects (effect of temporal, geographical proximity of threat and CFC on perceived severity)




dfMean

SquareF Sig.

Partial Eta

Squared

Noncent. Parameter

Observed Powerb

Corrected Model 9.538a 7 1.363 3.156 .008 .302 22.093 .918

Intercept 76.799 1 76.799 177.895 .000 .777 177.895 1.000

TIME_FACTOR 3.449 1 3.449 7.988 .007 .135 7.988 .792

GEOG_FACTOR 2.536 1 2.536 5.875 .019 .103 5.875 .662

CFC_TOTAL .209 1 .209 .485 .489 .009 .485 .105


2.492 1 2.492 5.773 .020 .102 5.773 .654

GEOG_FACTOR * CFC_TOTAL

2.093 1 2.093 4.849 .032 .087 4.849 .579

TIME_FACTOR * GEOG_FACTOR * CFC_TOTAL

1.875 1 1.875 4.343 .042 .078 4.343 .534


1.545 1 1.545 3.579 .064 .066 3.579 .459

Error 22.017 51 .432

Total 2098.938 59

Corrected Total 31.555 58

a. R Squared = .302 (Adjusted R Squared = .206)


128

Table 6.13b Experiment 1: Tests of between-subjects effects (effect of temporal, geographical proximity of threat and cfc on perceived vulnerability)


Dependent Variable:PERC_VUL


dfMean

SquareF Sig.

Partial Eta

Squared

Noncent. Parameter

Observed Powerb


Intercept 43.236 1 43.236 37.162 .000 .422 37.162 1.000

TIME_FACTOR 10.796 1 10.796 9.279 .004 .154 9.279 .848

GEOG_FACTOR 9.775 1 9.775 8.402 .006 .141 8.402 .812

CFC_TOTAL .783 1 .783 .673 .416 .013 .673 .127


8.870 1 8.870 7.624 .008 .130 7.624 .773


8.071 1 8.071 6.937 .011 .120 6.937 .734


3.782 1 3.782 3.251 .077 .060 3.251 .424


3.343 1 3.343 2.874 .096 .053 2.874 .384

Error 59.336 51 1.163

Total 1384.889 59


a. R Squared = .322 (Adjusted R Squared = .229)


The following Table (Table 6.14) summarizes the results of the

hypotheses testing for the effect of the stimuli on the PMT variables (H1-H4).

129

Table 6.14 Experiment 1: Summary of the hypotheses (H1 – H4) –mobile phone stimulus

Hypothesis Factor Perceivedseverity

Perceivedvulnerability

H1 Temporal proximity (H1) X X

H2 Geographical proximity (H2) X X

H3 Temporal proximity * Geographical proximity (H3)

X X

H4 CFC (H4) X X

X – no effect - Effect present

Watch stimuli

Table 6.15 shows the distribution characteristics of the protection

motivation variables. Similar to the mobile phone stimulus, most of the values

were negatively skewed. Group averages for the variables are shown in Table

6.16. Self efficacy was not included in the analysis as the scale reliability was

very low.

Table 6.15 Experiment 1: Distribution characteristics of the protection motivation variables for the watch stimuli

Variable Minimum Maximum Mean Std. Deviation

Perceived Severity 2.00 7.00 6.39 0.92

Perceived Vulnerability 1.00 7.00 4.73 1.43

Response Efficacy 2.33 7.00 5.15 1.24

Fear 1.83 6.50 3.86 0.99

130

Table 6.16 Experiment 1: Group wise mean values of protection motivation variables for the watch stimuli



Response Efficacy

SelfEfficacy

Fear

Temporal proximity: Day

6.27 4.60 5.13 3.87

Temporal proximity: Year 6.52 4.86 5.16 3.84

Geographical proximity: India 6.60 4.64 5.18 3.80

Geographical proximity: World 6.08 4.86 5.09 3.94

A one-way MANOVA was conducted to verify H1 which stated

that there will be significant differences in perceived severity and perceived

vulnerability among the groups in response to manipulation of the temporal

proximity of the threat. H1 was not accepted (Pillai’s Trace=0.021; Wilks’

lambda = 0.979; Hotelling’s Trace and Roy’s Largest Root = 0.021,

F(2,39)=0.415. Table 6.17a and 6.17b show the results.

Table 6.17a Experiment 1: Hypothesis 1: multivariate tests (watch stimuli)

Multivariate Testsb

Effect Value F Hypothesis df Error df Sig.Intercept Pillai's Trace .981 989.360a 2.000 39.000 .000

Wilks' Lambda .019 989.360a 2.000 39.000 .000Hotelling's Trace 50.736 989.360a 2.000 39.000 .000Roy's Largest Root 50.736 989.360a 2.000 39.000 .000

TIME_FACTOR Pillai's Trace .021 .415a 2.000 39.000 .664Wilks' Lambda .979 .415a 2.000 39.000 .664Hotelling's Trace .021 .415a 2.000 39.000 .664Roy's Largest Root .021 .415a 2.000 39.000 .664

a. Exact statistic b. Design: Intercept + TIME_FACTOR

131

Table 6.17b Experiment 1: Hypothesis 1: tests of between-subjects effects (watch stimuli)




df Mean

Square F Sig.

Corrected Model PERC_SEV .667a 1 .667 .782 .382

PERC_VUL .711b 1 .711 .343 .561


PERC_VUL 940.055 1 940.055 453.602 .000

TIME_FACTOR PERC_SEV .667 1 .667 .782 .382

PERC_VUL .711 1 .711 .343 .561


PERC_VUL 82.897 40 2.072


PERC_VUL 1023.333 42


PERC_VUL 83.608 41


b. R Squared = .009 (Adjusted R Squared = -.016)

Hypothesis 2 was not supported as geographical proximity had no

effect on perceived severity and vulnerability (Pillai’s Trace=0.118; Wilks’

lambda = 0.882; Hotelling’s Trace and Roy’s Largest Root = 0.134,

F(2,39)=2.60). The results are shown in Tables 6.18a and 6.18b.

132


Multivariate Testsb



GEOG_FACTOR Pillai's Trace .118 2.607a 2.000 39.000 .087Wilks' Lambda .882 2.607a 2.000 39.000 .087Hotelling's Trace .134 2.607a 2.000 39.000 .087Roy's Largest Root .134 2.607a 2.000 39.000 .087

a. Exact statistic b. Design: Intercept + GEOG_FACTOR





dfMean

Square F Sig.

Corrected Model PERC_SEV 2.650a 1 2.650 3.301 .077PERC_VUL .502b 1 .502 .242 .626


PERC_VUL 913.772 1 913.772 439.808 .000GEOG_FACTOR PERC_SEV 2.650 1 2.650 3.301 .077

PERC_VUL .502 1 .502 .242 .626Error PERC_SEV 32.118 40 .803

PERC_VUL 83.106 40 2.078



a. R Squared = .076 (Adjusted R Squared = .053) b. R Squared = .006 (Adjusted R Squared = -.019)

133

A 2 (temporal proximity of threat: day vs.year) x 2 (geographical

proximity of threat: India vs. World) multivariate analysis of variance

(MANOVA) was conducted to examine the interaction effects between

temporal proximity and geographical proximity to verify H3. It can be seen

from Table 6.19a and Table 6.19b that no interaction effect exists between

temporal proximity of threat and geographical proximity of threat on

perceived severity and vulnerability (Pillai’s Trace=0.044; Wilks’ lambda =

0.956; Hotelling’s Trace and Roy’s Largest Root = 0.046, F(2,37)=0.858).

Therefore, H3 was not accepted and the results are shown in


Multivariate Testsb


dfError

df Sig.

Intercept Pillai's Trace .982 1003.743a 2.000 37.000 .000Wilks' Lambda .018 1003.743a 2.000 37.000 .000Hotelling's Trace 54.256 1003.743a 2.000 37.000 .000Roy's Largest Root 54.256 1003.743a 2.000 37.000 .000

GEOG_FACTOR Pillai's Trace .129 2.735a 2.000 37.000 .078Wilks' Lambda .871 2.735a 2.000 37.000 .078Hotelling's Trace .148 2.735a 2.000 37.000 .078Roy's Largest Root .148 2.735a 2.000 37.000 .078


GEOG_FACTOR * TIME_FACTOR

Pillai's Trace .052 1.017a 2.000 37.000 .372Wilks' Lambda .948 1.017a 2.000 37.000 .372Hotelling's Trace .055 1.017a 2.000 37.000 .372Roy's Largest Root .055 1.017a 2.000 37.000 .372

a. Exact statistic b. Design: Intercept + GEOG_FACTOR + TIME_FACTOR + GEOG_FACTOR * TIME_FACTOR

134





df Mean

Square F Sig.

Corrected Model PERC_SEV 5.079a 3 1.693 2.167 .108

PERC_VUL 1.233b 3 .411 .190 .903


PERC_VUL 906.243 1 906.243 418.050 .000

GEOG_FACTOR PERC_SEV 2.853 1 2.853 3.651 .064

PERC_VUL .412 1 .412 .190 .665

TIME_FACTOR PERC_SEV 1.377 1 1.377 1.762 .192

PERC_VUL .695 1 .695 .321 .575

GEOG_FACTOR * TIME_FACTOR

PERC_SEV 1.500 1 1.500 1.920 .174

PERC_VUL .116 1 .116 .053 .819


PERC_VUL 82.376 38 2.168


PERC_VUL 1023.333 42


PERC_VUL 83.608 41


To test H4 a one-way MANCOVA was conducted with perceived

severity and perceived vulnerability as dependent variables and CFC as the

covariate. It can be inferred from Tables 6.20a and 6.20b that there was no

statistically significant difference between the groups and hence H4 was also

not supported.

135


Multivariate Testsb



CFC_TOTAL Pillai's Trace .086 1.791a 2.000 38.000 .181Wilks' Lambda .914 1.791a 2.000 38.000 .181Hotelling's Trace .094 1.791a 2.000 38.000 .181Roy's Largest Root .094 1.791a 2.000 38.000 .181


a. Exact statistic b. Design: Intercept + CFC_TOTAL + TIME_FACTOR




Type III Sum of Squares df Mean

Square F Sig.

Corrected Model PERC_SEV 1.441a 2 .720 .843 .438PERC_VUL 7.717b 2 3.859 1.983 .151


CFC_TOTAL PERC_SEV .774 1 .774 .906 .347PERC_VUL 7.006 1 7.006 3.600 .065

TIME_FACTOR PERC_SEV .712 1 .712 .834 .367PERC_VUL .859 1 .859 .441 .510




a. R Squared = .041 (Adjusted R Squared = -.008) b. R Squared = .092 (Adjusted R Squared = .046)

136

Similar to the previous experiment, a three way ANOVA was

conducted to analyse the effect of the independent variables and covariate on

perceived severity and perceived vulnerability. The results are shown in

Tables 6.21a and 6.21b. Unlike the mobile phone stimuli, the factors and the

covariate did not influence perceived severity or perceived vulnerability.

Table 6.21a Experiment 1: Tests of between-subjects effects (effect of temporal, geographical proximity of threat and cfc on perceived severity)




dfMean

SquareF Sig.

Partial Eta

Squared

Noncent. Parameter

Observed Powerb


Intercept 15.534 1 15.534 19.097 .000 .353 19.097 .989

TIME_FACTOR .084 1 .084 .103 .750 .003 .103 .061

GEOG_FACTOR .608 1 .608 .747 .393 .021 .747 .134

CFC_TOTAL 1.295 1 1.295 1.591 .215 .043 1.591 .233


.266 1 .266 .328 .571 .009 .328 .086


.321 1 .321 .395 .534 .011 .395 .094

TIME_FACTOR* GEOG_FACTOR * CFC_TOTAL

1.520 1 1.520 1.869 .180 .051 1.869 .265

Error 28.470 35 .813

Total 1751.250 42


a. R Squared = .181 (Adjusted R Squared = .041) b. Computed using alpha = .05

137

Table 6.21b Experiment 1: Tests of between-subjects effects (effect of temporal, geographical proximity of threat and cfc on perceived vulnerability)


Dependent Variable: PERC_VUL


dfMean

SquareF Sig.

Partial Eta

Squared

Noncent. Parameter

Observed Powerb

Corrected Model 11.303a 6 1.884 .912 .498 .135 5.471 .310

Intercept 5.637 1 5.637 2.729 .108 .072 2.729 .362

TIME_FACTOR 3.258 1 3.258 1.577 .218 .043 1.577 .231

GEOG_FACTOR .017 1 .017 .008 .928 .000 .008 .051

CFC_TOTAL 2.046 1 2.046 .990 .327 .028 .990 .162


2.738 1 2.738 1.326 .257 .036 1.326 .201


.040 1 .040 .019 .890 .001 .019 .052


.304 1 .304 .147 .703 .004 .147 .066

Error 72.305 35 2.066

Total 1023.333 42




The following table (Table 6.22) summarizes the effects of the

manipulations on the PMT variables.

138

Table 6.22 Experiment 1: Summary of the hypotheses (H1 – H4) -watch stimulus

Hypothesis Factor Perceived severity


H1 Temporal proximity X X

H2 Geographical proximity X X

H3 Temporal proximity * Geographical proximity

X X

H4 CFC X X

X – no effect - Effect present

6.3.6.4 Hypotheses tests of the relationship among PMT variables,

involvement, attitudes and intentions

Standard simple or multiple linear regression (ordinary least

squares (OLS)) analysis was run to ascertain the relationship between the

PMT variables, involvement, attitudes and intention variables. Only one or

two predictors were considered at a time. Similarly assumptions regarding

multicollinearity, homoscedascity, linearity and normality of residuals were

met in all the scenarios


A multiple regression analysis was conducted to evaluate how well

the perceived severity and perceived vulnerability predicted fear and the

efficacy variables. Neither perceived severity nor perceived vulnerability

predicted the dependent variables. Therefore H9a, H9b and H9c were not

supported. The results are shown in Table 6.23a, 6.23b and 6.23c. It can

therefore be inferred that the perceived threat levels did not affect fear.

139

Table 6.23a Experiment 1: Hypothesis 9a: effect of threat appraisal components on fear (mobile phone stimuli)

Model Summary Model R R Square Adjusted R Square Std. Error of the Estimate

1 .280a .079 .046 .90543a. Predictors: (Constant), PERC_VUL, PERC_SEV

ANOVAb

Model Sum of Squares df Mean

Square F Sig.

1 Regression 3.919 2 1.960 2.390 .101a

Residual 45.909 56 .820Total 49.828 58

a. Predictors: (Constant), PERC_VUL, PERC_SEV b. Dependent Variable: FEAR

Table 6.23b Experiment 1: Hypothesis 9b: effect of threat appraisal components on response efficacy (mobile phone stimuli)


1 .247a .061 .027 1.23681a. Predictors: (Constant), PERC_VUL, PERC_SEV

ANOVAb


Square F Sig.

1 Regression 5.543 2 2.771 1.812 .173a

Residual 85.663 56 1.530Total 91.205 58

a. Predictors: (Constant), PERC_VUL, PERC_SEV b. Dependent Variable: RESP_EFFICACY

140

Table 6.23c Experiment 1: Hypothesis 9c: effect of threat appraisal components on self efficacy (mobile phone stimuli)

Model Summary

Model R R Square Adjusted R Square Std. Error of the Estimate

1 .074a 0.005 -0.030 1.16446

a. Predictors: (Constant), PERC_VUL, PERC_SEV

ANOVAb

Model Sum of Squares

df Mean

Square F Sig.

1 Regression .419 2 .210 .155 .857a

Residual 75.935 56 1.356

Total 76.354 58


b. Dependent Variable: SELF_EFFICACY

Simple linear regression analyses was conducted to evaluate

how well fear and the other PMT variables (perceived severity, vulnerability,

response efficacy and self efficacy) predict message involvement. The results

are shown in the following tables (Tables 6.24a, 6.24b, 6.24c, 6.24d and

6.24e). It can be seen from Table 6.24b that perceived vulnerability predicts

message involvement ( =0.30 t(58)=2.43, p<0.05, R2=0.09). From Table

6.24d, it can be seen that response efficacy predicted message involvement

=0.30 t(58)=2.41, p<0.05, R2=0.09). Therefore only H10b and H10d were

supported. The other PMT variables did not have any effect on message

involvement. Hence only perceived vulnerability and response efficacy have

an effect on message involvement.

141

Table 6.24a Experiment 1: Hypothesis 10a: effect of perceived severity on message involvement (mobile phone stimuli)


1 .197a 0.039 0.022 1.06258a. Predictors: (Constant), PERC_SEV

ANOVAb

Model Sum of Squares df

MeanSquare F Sig.

1 Regression 2.596 1 2.596 2.299 .135a

Residual 64.357 57 1.129Total 66.953 58

a. Predictors: (Constant), PERC_SEV b. Dependent Variable: MESSAGE_INVOLVEMENT

Table 6.24b Experiment 1: Hypothesis 10b: effect of perceived vulnerability on message involvement (mobile phone stimuli)

Model Summary

Model R R Square Adjusted R

Square Std. Error of the

Estimate 1 .307a 0.094 0.078 1.03138

a. Predictors: (Constant), PERC_VUL

ANOVAb


df Mean

Square F Sig.

1 Regression 6.319 1 6.319 5.941 .018a

Residual 60.634 57 1.064Total 66.953 58

a. Predictors: (Constant), PERC_VUL b. Dependent Variable: MESSAGE_INVOLVEMENT

142


Coefficientsa

Model

Unstandardized Coefficients

Standardized Coefficients t Sig.

B Std. Error Beta

1 (Constant) 3.602 .534 6.743 .000

PERC_VUL .269 .110 .307 2.437 .018

a. Dependent Variable: MESSAGE_INVOLVEMENT

Table 6.24c Experiment 1: Hypothesis 10c: effect of fear on message involvement (mobile phone stimuli)

Model Summary

Model R R Square Adjusted R SquareStd. Error of the

Estimate

1 .193a 0.037 0.020 1.06344

a. Predictors: (Constant), FEAR

ANOVAb


dfMean

Square F Sig.

1 Regression 2.491 1 2.491 2.203 .143a

Residual 64.461 57 1.131

Total 66.953 58


b. Dependent Variable: MESSAGE_INVOLVEMENT

143

Table 6.24d Experiment 1: Hypothesis 10d: effect of response efficacy on message involvement (mobile phone stimuli)

Model Summary


1 .305a 0.093 0.077 1.032228

a. Predictors: (Constant), RESP_EFFICACY

ANOVAb


df Mean

Square F Sig.

1 Regression 6.214 1 6.214 5.831 .019a

Residual 60.739 57 1.066

Total 66.953 58


b. Dependent Variable: MESSAGE_INVOLVEMENT

Coefficientsa

Model



B Std. Error Beta

1 (Constant) 3.561 .555 6.415 .000

RESP_EFFICACY .261 .108 .305 2.415 .019


Table 6.24e Experiment 1: Hypothesis 10e: effect of self efficacy on message involvement (mobile phone stimuli)

Model Summary


1 .167a 0.028 0.011 1.06864

a. Predictors: (Constant), SELF_EFFICACY

144

Table 6.24e (Continued)

ANOVAb


Square F Sig.

1 Regression 1.859 1 1.859 1.628 .207a

Residual 65.094 57 1.142Total 66.953 58

a. Predictors: (Constant), SELF_EFFICACY b. Dependent Variable: MESSAGE_INVOLVEMENT

Simple regression analysis was conducted to evaluate how well coping variables predicted attitude towards the ad and purchase intention. It

can be seen from Table 6.25a and Table 6.25c that response efficacy significantly predicted attitude towards the ad ( =0.32 t(58)=2.550, p<0.05) and purchase intention ( =0.297 t(58)=2.352, p<0.05 R2=0.08) while self-efficacy did not predict both the variables (Tables 6.25b and Table 6.25d).

Therefore H11a and H11c were supported and H11b and H11d were not supported.

Table 6.25a Experiment 1: Hypothesis 11a: effect of response efficacy on attitude towards ad (mobile phone stimuli)


1 .320a 0.102 0.087 1.04128a. Predictors: (Constant), RESP_EFFICACY

ANOVAb


Square F Sig.

1 Regression 7.048 1 7.048 6.500 .013a

Residual 61.803 57 1.084Total 68.851 58

a. Predictors: (Constant), RESP_EFFICACY b. Dependent Variable: ATTITUDE_AD

145

Table 6.25a (Continued)

Coefficientsa

Model



B Std. Error Beta

1 (Constant) 3.858 .560 6.889 .000

RESP_EFFICACY .278 .109 .320 2.550 .013

a. Dependent Variable: ATTITUDE_AD

Table 6.25b Experiment 1: Hypothesis 11b: effect of self efficacy on attitude towards ad (mobile phone stimuli)

Model Summary


1 .126a 0.016 -0.001 1.09033


ANOVAb


df Mean

Square F Sig.

1 Regression 1.089 1 1.089 .916 .343a

Residual 67.762 57 1.189

Total 68.851 58

a. Predictors: (Constant), SELF_EFFICACY b. Dependent Variable: ATTITUDE_AD

146

Table 6.25c Experiment 1: Hypothesis 11c: effect of response efficacy on purchase intention (mobile phone stimuli)

Model Summary


1 .297a 0.088 0.072 1.66704


ANOVAb


df Mean

Square F Sig.

1 Regression 15.373 1 15.373 5.532 .022a

Residual 158.405 57 2.779

Total 173.778 58

a. Predictors: (Constant), RESP_EFFICACY b. Dependent Variable: PURCHASE_INTENTION

Coefficientsa

Model


Standardized Coefficients

t Sig. B

Std. Error

Beta

1 (Constant) 1.954 .896 2.180 .033

RESP_EFFICACY .411 .175 .297 2.352 .022

a. Dependent Variable: PURCHASE_INTENTION

Table 6.25d Hypothesis 11d: effect of self efficacy on purchase intention (mobile phone stimuli)

Model Summary

Model R R Square Adjusted R SquareStd. Error of the

Estimate

1 .0.215a 0.46 0.030 1.70519


147

Table 6.25d (Continued)

ANOVAb


Square F Sig.

1 Regression 8.041 1 8.041 2.765 .102a

Residual 165.737 57 2.908Total 173.778 58

a. Predictors: (Constant), SELF_EFFICACY b. Dependent Variable: PURCHASE_INTENTION

H12a was not supported as environmental knowledge predicted perceived severity in the direction opposite to the one hypothesized. H12b and H12c were not supported as simple regression analyses revealed that environmental knowledge did not predict perceived vulnerability and fear. Environmental knowledge did not predict message involvement and therefore H12d was not supported. Tables 6.26a, 6.26b 6.26c, 6.26d illustrate the results.

Table 6.26a Experiment 1: Hypothesis 12a: effect of environmental knowledge on perceived severity (mobile phone stimuli)

Model Summary

Model R R Square Adjusted R Square Std. Error of the

Estimate

1 .0.333a 0.111 0.095 0.70162

a. Predictors: (Constant), ENV_KNOW

ANOVAb


df Mean

Square F Sig.

1 Regression 3.496 1 3.496 7.102 .010a

Residual 28.059 57 .492

Total 31.555 58

a. Predictors: (Constant), ENV_KNOW b. Dependent Variable: PERC_SEV

148


Coefficientsa

Model



t Sig. B

Std.Error

Beta

1 (Constant) 5.053 .338 14.954 .000

ENV_KNOW .108 .041 .333 2.665 .010

a. Dependent Variable: PERC_SEV

Table 6.26b Experiment 1: Hypothesis 12b: effect of environmental knowledge on perceived vulnerability (mobile phone stimuli)

Model Summary


Estimate

1 .0.033a 0.001 -0.016 1.23849


ANOVAb


df Mean

Square F Sig.

1 Regression .096 1 .096 .063 .803a

Residual 87.429 57 1.534

Total 87.525 58

a. Predictors: (Constant), ENV_KNOW b. Dependent Variable: PERC_VUL

The results show that environmental knowledge played no role in

influencing threat perception.

149

Table 6.26c Experiment 1: Hypothesis 12c: effect of environmental knowledge on fear (mobile phone stimuli)

Model Summary


Estimate

1 .0.045a 0.002 -0.015 0.93401


ANOVAb


MeanSquare F Sig.

1 Regression .103 1 .103 .118 .733a

Residual 49.725 57 .872Total 49.828 58

a. Predictors: (Constant), ENV_KNOW b. Dependent Variable: FEAR

Table 6.26d Experiment 1: Hypothesis 12d: effect of environmental knowledge on message involvement (mobile phone stimuli)

Model Summary


Estimate

1 .0.065a 0.004 -0.013 1.08150


ANOVAb


df Mean

Square F Sig.

1 Regression .283 1 .283 .242 .625a

Residual 66.670 57 1.170Total 66.953 58

a. Predictors: (Constant), ENV_KNOW b. Dependent Variable: MESSAGE_INVOLVEMENT

150

Similar regression analysis was conducted to evaluate the

relationship among involvement variables and attitude towards ad.

Environmental concern was positively related to message involvement and

therefore H13a was supported ( =0.267, t(58)=2.055, p<0.05 R2=0.07).

However it did not predict attitude towards the ad and purchase intention.

Therefore H13b and H13c were not supported. The following tables (Tables

6.27a, 6.27b, 6.27c) show the regression results which clearly show that

environmental concern was related only to message involvement.

Table 6.27a Experiment 1: Hypothesis 13a: effect of environmental concern (enduring involvement with the environment) on message involvement (mobile phone stimuli)


1 0.267a 0.071 0.054 1.01307a. Predictors: (Constant), TOTAL_ENV_CONCERN

ANOVAb


Square F Sig.

1 Regression 4.336 1 4.336 4.225 .045a

Residual 56.447 55 1.026Total 60.784 56

a. Predictors: (Constant), TOTAL_ENV_CONCERN b. Dependent Variable: MESSAGE_INVOLVEMENT

Coefficientsa

Model



t Sig. B

Std.Error

Beta

1 (Constant) 3.021 .934 3.234 .002

TOTAL_ENV_CONCERN .324 .158 .267 2.055 .045a. Dependent Variable: MESSAGE_INVOLVEMENT

151

Table 6.27b Experiment 1: Hypothesis 13b: effect of environmental concern (enduring involvement with the environment) on attitude towards the ad (mobile phone stimuli)

Model Summary


1 .0.119a 0.014 -0.004 1.09816

a. Predictors: (Constant), TOTAL_ENV_CONCERN

ANOVAb


df Mean

Square F Sig.

1 Regression .947 1 .947 .785 .379a

Residual 66.328 55 1.206

Total 67.275 56

a. Predictors: (Constant), TOTAL_ENV_CONCERN b. Dependent Variable: ATTITUDE_AD

Table 6.27c Experiment 1: Hypothesis 13c: effect of environmental concern (enduring involvement with the environment) on purchase intention (mobile phone stimuli)

Model Summary


1 0.012a 0.000 -0.018 1.70730


ANOVAb


df Mean

Square F Sig.

1 Regression .022 1 .022 .007 .932a

Residual 160.318 55 2.915

Total 160.339 56

a. Predictors: (Constant), TOTAL_ENV_CONCERN b. Dependent Variable: PURCHASE_INTENTION

152

H14 was supported as message involvement predicted attitude

towards ad ( =0.616, t(58)=5.902, p<0.001). Overall model fit was R2=0.359.

Table 6.28 shows the results.

Table 6.28 Experiment 1: Hypothesis 14: effect of message involvement on attitude towards ad (mobile phone stimuli)

Model Summary


1 0.616a 0.359 0.368 0.86585

a. Predictors: (Constant), MESSAGE_INVOLVEMENT

ANOVAb


df Mean

Square F Sig.

1 Regression 26.118 1 26.118 34.838 .000a

Residual 42.733 57 .750

Total 68.851 58

a. Predictors: (Constant), MESSAGE_INVOLVEMENT b. Dependent Variable: ATTITUDE_AD

Coefficientsa

Model Unstandardized

Coefficients Standardized Coefficients t Sig.

B Std. Error Beta

1 (Constant) 2.207 .527 4.190 .000

MESSAGE_INVOLVEMENT .625 .106 .616 5.902 .000


It can be seen from Tables 6.29 and 6.30 that H15 and H16 were

also supported as attitude towards the ad significantly predicted the attitude

towards the brand ( =0.58 t(58)=5.412, p<0.001, R2=0.33) and attitude

153

towards the brand significantly predicted the purchase intention ( =0.517,

t(58)=4.557, p<0.001, R2=0.267).

Table 6.29 Experiment 1: Hypothesis 15: effect of attitude towards adon attitude towards brand (mobile phone stimuli)

Model Summary


1 0.583a 0.339 0.328 0.91829

a. Predictors: (Constant), ATTITUDE_AD

ANOVAb


df Mean

Square F Sig.

1 Regression 24.703 1 24.703 29.295 .000a

Residual 48.065 57 .843

Total 72.768 58


b. Dependent Variable: ATTITUDE_BRAND

Coefficientsa

Model



B Std. Error Beta

1 (Constant) 1.905 .592 3.215 .002

ATTITUDE_AD .599 .111 .583 5.412 .000

a. Dependent Variable: ATTITUDE_BRAND

154

Table 6.30 Experiment 1: Hypothesis 16: effect of attitude towards brand on purchase intention (mobile phone stimuli)

Model Summary


1 0.517 0.267 0.254 1.49485

a. Predictors: (Constant), ATTITUDE_BRAND

ANOVAb


df Mean

Square F Sig.

1 Regression 46.406 1 46.406 20.767 .000a

Residuals 127.372 57 2.235

Total 173.778 58


b. Dependent Variable: PURCHASE_INTENTION

Coefficientsa

Model Unstandardized Coefficients


t Sig.

B Std. Error Beta

1 (Constant) -.029 .905 -.032 .975

ATTITUDE_BRAND .799 .175 .517 4.557 .000


The results of the hypothesis tests regarding the relationship

between PMT variables, involvement, attitudes and intentions for the mobile

phone stimuli are summarised in Table 6.31.

155

Table 6.31 Experiment 1: Summary of hypotheses tests regarding relationship between PMT variables, involvement, attitudes and intentions (H9 – H16) (mobile phone stimuli)

Hypothesis Predictor Dependent

variable R2 Adjusted

R2Unstandardised

coefficient B Standardised coefficient

H9a Perceived Severity and Perceived vulnerability

Fear Not Significant

H9b Perceived Severity and Perceived vulnerability

Response Efficacy

Not Significant

H9c Perceived Severity and Perceived vulnerability

Self Efficacy Not Significant

H10a Perceived Severity

Message Involvement

Not Significant

H10b Perceived Vulnerability

Message Involvement

0.09 0.07 0.26 0.30*

H10c Fear Message Involvement

Not Significant

H10d Response Efficacy

Message Involvement

0.09 0.07 0.26 0.30*

H10e Self Efficacy Message Involvement

Not Significant

H11a Response efficacy

Attitude towards ad

0.10 0.08 0.27 0.32*

H11b Self Efficacy Attitude towards ad

Not Significant

H11c Response efficacy

Purchase Intention

0.08 0.07 0.41 0.29*

H11d Self Efficacy Purchase Intention

Not Significant

H12a EnvironmentalKnowledge

Perceived Severity

0.11 0.09 0.10 0.33*

156

Table 6.31 (Continued)

H12b EnvironmentalKnowledge

Perceived Vulnerability

Not Significant

H12c EnvironmentalKnowledge


H12d EnvironmentalKnowledge

Message Involvement

Not Significant

H13a Environmental concern

Message Involvement

0.07 0.05 0.32 0.26*

H13b Environmental concern

Attitude towards ad

Not significant

H13c Environmental concern

Purchase Intention

Not significant

H14 Message Involvement

Attitude towards ad

0.37 0.36 0.625 0.616***

H15 Attitude towards ad

Attitude towards brand

0.33 0.32 0.59 0.58***

H16 Attitude towards brand

Purchase Intention

0.26 0.25 0.79 0.51***

***p <.001 **p <.01 *p <.05; n= 58

Watch stimuli


perceived severity and perceived vulnerability predicted fear and response

efficacy. Table 6.32a shows that H9a was not supported as the model was not

significant. H9b was partially supported as perceived vulnerability positively

influenced response efficacy ( =0.595 t(39)=4.099, p<0.001, R2 =0.309). H9c

was not tested because of the low reliability values for self efficacy. The

results of the regression analysis are shown below in Table 6.32b.

157

Table 6.32a Experiment 1: Hypothesis 9a: effect of threat appraisal components on fear (watch stimuli)

Model Summary


1 .153a 0.023 -0.27 1.00674


ANOVAb


df Mean

Square F Sig.

1 Regression .949 2 .474 .468 .630a

Residual 39.527 39 1.014

Total 40.476 41

a. Predictors: (Constant), PERC_VUL, PERC_SEV b. Dependent Variable: FEAR

Table 6.32b Experiment 1: Hypothesis 9b: effect of threat appraisal components on response efficacy (watch stimuli)

Model Summary


1 .556a 0.309 0.273 1.05815


ANOVAb


df Mean

Square F Sig.

1 Regression 19.489 2 9.744 8.703 .001a

Residual 43.667 39 1.120

Total 63.156 41


158


Coefficientsa



B Std. Error Beta

1 (Constant) 3.854 1.163 3.313 .002

PERC_SEV -.180 .196 -.133 -.919 .364

PERC_VUL .517 .126 .595 4.099 .000

a. Dependent Variable: RESP_EFFICACY

H10a was supported as perceived severity predicted message

involvement ( =0.37 t(39)=2.496, p<0.05, R2 = 0.367). H10b and H10d were

supported, as simple linear regression analyses revealed that among the PMT

variables only perceived vulnerability ( =0.54 t(39)=4.108, p<0.001, R2 =

0.29) and response efficacy ( =0.39 t(39)=2.717, p<0.05,R2 = 0.156)

significantly predicted message involvement. The results are shown in the

following tables (Tables 6.33a, 6.33b, 6.33c and 6.33d). H10e was not tested

because it involved self efficacy.

Table 6.33a Experiment 1: Hypothesis 10a: effect of perceived severity on message involvement (watch stimuli)


1 .367a 0.135 0.113 1.02271a. Predictors: (Constant), PERC_SEV

ANOVAb


df Mean

Square F Sig.

1 Regression 6.518 1 6.518 6.232 .017a

Residual 41.837 40 1.046

159



df Mean

Square F Sig.

Total 48.355 41a. Predictors: (Constant), PERC_SEV b. Dependent Variable: MESSAGE_INVOLVEMENT

Coefficientsa



B Std. Error Beta

1 (Constant) 2.077 1.120 1.855 .071

PERC_SEV .433 .173 .367 2.496 .017


Table 6.33b Experiment 1: Hypothesis 10b: effect of perceived vulnerability on message involvement (watch stimuli)


1 .545a 0.297 0.279 0.92206a. Predictors: (Constant), PERC_VUL

ANOVAb


MeanSquare F Sig.

1 Regression 14.348 1 14.348 16.876 .000a

Residual 34.008 40 .850Total 48.355 41


160


Coefficientsa



B Std. Error Beta 1 (Constant) 2.886 .498 5.798 .000

PERC_VUL .414 .101 .545 4.108 .000a. Dependent Variable: MESSAGE_INVOLVEMENT

Table 6.33c Experiment 1: Hypothesis 10c: effect of fear on message involvement (watch stimuli)

Model Summary


1 .248a 0.061 0.038 1.06524


ANOVAb


df Mean

Square F Sig.

1 Regression 2.966 1 2.966 2.614 .114a

Residual 45.389 40 1.135

Total 48.355 41

a. Predictors: (Constant), FEAR b. Dependent Variable: MESSAGE_INVOLVEMENT

Table 6.33d Experiment 1: Hypothesis 10d: effect of response efficacy on message involvement (watch stimuli)



161


ANOVAb


df Mean

Square F Sig.

1 Regression 7.534 1 7.534 7.383 .010a

Residual 40.821 40 1.021

Total 48.355 41

a. Predictors: (Constant), RESP_EFFICACY b. Dependent Variable: MESSAGE_INVOLVEMENT

Coefficientsa



B Std. Error Beta

1 (Constant) 3.066 .673 4.556 .000

RESP_EFFICACY .345 .127 .395 2.717 .010


Results of the simple regression analysis revealed the effect of

response efficacy on attitude towards the ad and purchase intention. The

model was not significant. Table 6.34a and 6.34b show that H11a and H11c

were not supported as response efficacy did not predict attitude towards the ad

and purchase intention.

Table 6.34a Experiment 1: Hypothesis 11a: effect of response efficacy on attitude towards ad (watch stimuli)

Model Summary


1 .200a 0.040 0.016 1.01291


162


ANOVAb


df Mean

Square F Sig.

1 Regression 1.714 1 1.714 1.671 .204a

Residual 41.040 40 1.026

Total 42.754 41


Table 6.34b Experiment 1: Hypothesis 11c: effect of response efficacy on purchase intention (watch stimuli)

Model Summary


1 .266a 0.071 0.047 1.50427


ANOVAb


df Mean

Square F Sig.

1 Regression 6.865 1 6.865 3.034 .089a

Residual 90.513 40 2.263Total 97.378 41


H12a, H12b, H12c and H12d were not supported environmental knowledge did not predict the hypothesized dependent variables. Tables 6.35a, 6.35b 6.35c, 6.35d illustrate the results.

163

Table 6.35a Experiment 1: Hypothesis 12a: effect of environmental knowledge on perceived severity (watch stimuli)


1 .0.247a 0.061 0.038 0.90330a. Predictors: (Constant), ENV_KNOW

ANOVAb


Square F Sig.

1 Regression 2.130 1 2.130 2.610 .114a

Residual 32.638 40 .816Total 34.768 41


Table 6.35b Experiment 1: Hypothesis 12a: effect of environmental knowledge on perceived vulnerability (watch stimuli)

Model Summary Model R R Square Adjusted R Square Std. Error of the Estimate1 .0.219a 0.048 0.024 1.41063a. Predictors: (Constant), ENV_KNOW

ANOVAb


Square F Sig.

1 Regression 4.014 1 4.014 2.017 .163a

Residual 79.595 40 1.990Total 83.608 41


164

Table 6.35c Experiment 1: Hypothesis 12a: effect of environmental knowledge on fear (watch stimuli)

Model Summary


1 0.123a 0.015 -0.10 0.99834


ANOVAb


df Mean

Square F Sig.

1 Regression .609 1 .609 .611 .439a

Residual 39.867 40 .997

Total 40.476 41


Table 6.35d Experiment 1: Hypothesis 12b: effect of environmental knowledge on message involvement (watch stimuli)

Model Summary


1 .0.010a 0.000 -0.025 1.09944


ANOVAb


df Mean

Square F Sig.

1 Regression .005 1 .005 .004 .950a

Residual 48.350 40 1.209

Total 48.355 41


165

Simple regression analysis was conducted to evaluate if

involvement predicted attitude towards ad. It can be seen from Tables 6.36a,

6.36b, 6.36c that the regression model was not significant. Unlike the

previous experiment with mobile stimuli, environmental concern was not

related to message involvement, attitude towards ad or purchases intentions.

Therefore H13a, H13b and H13c were not supported.

Table 6.36a Experiment 1: Hypothesis 13a: effect of environmental concern (enduring involvement with the environment) on message involvement (watch stimuli)

Model Summary


1 0.227a 0.051 0.028 1.07086


ANOVAb


df Mean

Square F Sig.

1 Regression 2.485 1 2.485 2.167 .149a

Residual 45.870 40 1.147

Total 48.355 41


Table 6.36b Experiment 1: Hypothesis 13b: effect of environmental concern (enduring involvement with the environment) on attitude towards the ad (watch stimuli)

Model Summary


1 .0.157a 0.025 0.000 1.02015


166


ANOVAb


df Mean

Square F Sig.

1 Regression 1.052 1 1.052 1.009 .321a

Residual 41.702 40 1.043

Total 42.754 41


Table 6.36c Experiment 1: Hypothesis 13c: effect of environmental concern (enduring involvement with the environment) on purchase intention (watch stimuli)

Model Summary


1 0.233a 0.054 0.031 1.51735


ANOVAb


df Mean

Square F Sig.

1 Regression 5.284 1 5.284 2.295 .138a

Residual 92.094 40 2.302

Total 97.378 41


H14 was supported as message involvement significantly predicted

attitude towards ad ( =0.716, t(39)=6.482, p<0.001). Model fit was good as

R2=0.51. Table 6.37 shows the results.

167

Table 6.37 Experiment 1: Hypothesis 14: effect of message involvement on attitude towards ad (watch stimuli)

Model Summary


1 0.716a 0.512 0.500 0.72201


ANOVAb


df Mean

Square F Sig.

1 Regression 21.902 1 21.902 42.015 .000a

Residual 20.852 40 .521

Total 42.754 41


Coefficientsa



B Std. Error Beta

1 (Constant) 2.096 .515 4.068 .000



H15 and H16 were also supported as attitude towards the ad

significantly predicted the attitude towards the brand ( =0.709, t(39)=6.367,

p<0.001, R2=0.50) and attitude towards the brand significantly predicted the

purchase intention ( =0.744, t(39)=7.037, p<0.001, R2=0.55). Tables 6.38 and

6.39 show the regression results for H15 and H16.

168

Table 6.38 Experiment 1: Hypothesis 15: effect of attitude towards adon attitude towards brand (watch stimuli)

Model Summary


1 0.709a 0.503 0.491 0.90473


ANOVAb


df Mean

Square F Sig.

1 Regression 33.185 1 33.185 40.542 .000a

Residual 32.741 40 .819

Total 65.926 41

a. Predictors: (Constant), ATTITUDE_AD b. Dependent Variable: ATTITUDE_BRAND

Coefficientsa



B Std. Error Beta

1 (Constant) .503 .754 .666 .509

ATTITUDE_AD .881 .138 .709 6.367 .000

a. Dependent Variable: ATTITUDE_BRAND

Table 6.39 Experiment 1: Hypothesis 16: effect of attitude towards brand on purchase intention (watch stimuli)

Model Summary


1 0.744a 0.553 0.542 1.04296


169


ANOVAb


dfMean

Square F Sig.

1 Regression 53.868 1 53.868 49.521 .000a

Residual 43.511 40 1.088

Total 97.378 41

a. Predictors: (Constant), ATTITUDE_BRAND b. Dependent Variable: PURCHASE_INTENTION

Coefficientsa



B Std. Error Beta

1 (Constant) .097 .690 .141 .889

ATTITUDE_BRAND .904 .128 .744 7.037 .000


Table 6.40 shows the summary of the results regarding the relationship between the PMT variable, involvement, attitudes and purchase intention for the watch stimuli.

Table 6.40 Experiment 1: Summary of hypotheses tests regarding relationship between PMT variables, involvement, attitudes and intentions (H9 – H16) (watch stimuli)



R2Unstandardised





Response Efficacy

Not Significant

170


Hypothesis Predictor Dependent variable R2 Adjusted

R2Unstandardised


H9c Perceived Severity and Perceived vulnerability

Self Efficacy Not Tested


Message Involvement

0.135 0.113 0.433 0.367*

H10b PerceivedVulnerability

MessageInvolvement

0.29 0.27 0.41 0.54***


Not Significant


Message Involvement

0.15 0.14 0.345 0.39*


Not tested


Attitude towards ad

NotSignificant


Not tested


Purchase Intention

Not Significant


Not tested


Perceived Severity

Not Significant


Perceived Vulnerability

Not Significant




Message Involvement

Not Significant


Message Involvement

Not Significant


Attitude towards ad

Not significant


Purchase Intention

Not significant


Attitude towards ad

0.51 0.50 0.673 0.716***



0.50 0.49 0.88 0.70***


Purchase Intention

0.55 0.54 0.90 0.74***

***p <.001 **p <.01 *p <.05; n= 41

171

6.3.7 Conclusions from Experiment 1

This experiment examined the effects of temporal and geographical

framing of threat on PMT variables and the subsequent effects of the PMT

variables on message involvement, attitudes and purchase intention.

Although the hypothesized relationship regarding the main effects were not

significant, it was seen that temporal proximity of the threat had a significant

effect on perceived severity in the case of mobile phone stimuli. This finding

supports current research that argue that threats in “day” terms are considered

more closer that those that are presented in “year” terms (Chandran & Menon

2004). It also shows that the self-positivity bias is reduced (Gilovich et al

1993; Raghubir & Menon 1998) as people find a temporally closer threat

relevant. However this effect was not observed with the watch stimuli. This

could be because plastic waste pollution was viewed as a severe threat by

most participants as the mean value of perceived severity was very high

(6.39) when compared to the problem of e-waste (5.9). Plastic waste can be a

more familiar issue in India as the consumer encounters regular mandatory

governmental instructions and news articles on this issue. Hence for a familiar

issue, perceived severity is rated high when compared to an unfamiliar issue

like e-waste. This difference in the arousal of fear based on issue familiarity

has been discussed by Pelsmacker et al (2011). However, only perceived

severity was viewed differently in this experiment. In a similar vein,

Obermiller (1995) also found that different appeals worked for familiar and

unfamiliar issues. The reported perceived vulnerability and fear were almost

similar in both the cases. This could be again because of issue familiarity as

the watch stimulus highlights the threat of plastic waste. Therefore consumers

are more aware of the issue of plastic waste when compared to e-waste. Of

late, the government of India mandates the pricing of plastic bags that are

used to packing the goods sold by a number of retail stores. The stores also

prominently display statutory messages advocating the reduction of plastic.

172

Therefore, for a familiar issue, perceived severity and perceived vulnerability

could have had a greater impact when compared to an unfamiliar issue like e-

waste disposal.

The hypothesized effect of CFC was not supported. However, in

the case of mobile phone stimulus, CFC interacted with the factors to produce

significant effects on perceived severity and perceived vulnerability. This

finding supports existing literature that consumers perception of risk varied

based on their temporal orientation (Orbell et al 2004 & Orbell & Hagger

2006; Morison et al 2010). There were no statistically significant interaction

effects of temporal proximity and geographical proximity of the threat on the

PMT variables for both the watch and mobile phone stimuli. Most participants

exhibited high levels of perceived severity, vulnerability and fear towards

environmental threats.

In both the cases, perceived vulnerability and response efficacy

significantly predicted message involvement. This is similar to the finding by

Cauberghe et al (2009). Response efficacy is a variable that is linked to

“Perceived consumer effectiveness” (PCE) and is shown to be related to

consumer’s environmental behaviour (Gilg et al 2005).

Unlike previous studies environmental knowledge did not predict

severity or fear as hypothesized. In the case of watch stimuli, environmental

knowledge predicted perceived severity in the positive direction contrary to

the hypothesized nature of the relationship, but in the case of mobile phone

stimuli environmental knowledge did not have any effect.

Most significantly, environmental concern did not have any effect

on attitudes, message involvement or purchase intention. This is in direct

contrast to the propositions put forward by the ELM (Petty & Cacioppo 1986).

Issue involvement is supposed to activate message elaboration and therefore

173

increase involvement with the message (Maheswaran & Meyers-Levy 1990).

However consumers who reported high environmental concern (issue

involvement) did not exhibit this behaviour. This highlights the fact that better

measures are needed to assess environmental concern or enduring

involvement with the environment. Most respondents probably report

environmental concern as a socially desirable response (Bord et al 1998;

Ewert & Baker 2001; Ewert & Galloway 2009) and hence the results did not

support the related hypothesis.

Attitude towards the advertisement significantly predicted attitude

towards the brand and attitude towards the brand significantly predicted

purchase intentions in both the scenarios. This is in congruence with the dual

mediation hypothesis that states that attitude related cognitions affect the

attitude towards the ad, brand and in turn behaviour related to purchase

intentions (MacKenzie et al 1986; Teng et al 2007). The results support the

findings from most advertising studies that show that attitude towards

advertising has a strong influence on attitude towards the brand under high

involvement conditions (Gardner 1985; Park & Young 1986; Muehling &

Laczniak 1988).

This experiment showed that response efficacy and perceived

vulnerability greatly increased message involvement and message

involvement subsequently influenced attitudes and intentions. However

perceived severity and vulnerability did not cause fear arousal in both the

scenarios. Therefore it is necessary to investigate if fear arousal can be

obtained using a different message frame. Hence, study 2 was conducted

using goal frames and threat level as factors to increase the success of the

manipulations. Although it is not essential to conduct the two studies (study1

and study2) sequentially, they were conducted one after another in this

research to examine the effect of the framing manipulations.

174

6.4 STUDY 2: EXPERIMENT 2: THREAT LEVELS AND GOAL

FRAMING (WATCH STIMULI)

Experiment 2 was designed to use threat levels and goal frames to

influence the PMT variables. Plastic waste seemed to generate higher levels

of scores for the threat appraisal variables in the previous study. Therefore the

experiment was conducted with wristwatch as the chosen product. The

advertisements were designed for this product (Ad3). The results of this

experiment were used to design the stimuli for the next experiment.


A 2 (threat level: high vs. low) x 2 (goal frame: loss vs. gain)

between subjects experimental design was utilized to investigate the

hypotheses. This resulted in four possible combinations of the stimuli. Sixty

nine postgraduate M.E. students from a large South Indian University (95.7 %

male, median age=22) were randomly assigned to the four possible conditions

for the watch stimuli.

The experimental procedure was the same as Experiment 1. Data

collection was through a paper and pencil questionnaire. Students first filled

the questionnaire (Q3) containing dependent variables (Appendix 5). Next,

they were asked to answer the filler task which asked them to list the reason

why they liked their favourite celebrity similar to Experiment 1. On

completion, they filled counterbalanced questionnaires on environmental

concern and objective environmental knowledge. This was similar to the

questions in Experiment 1. The personality variable CFC was not included in

this questionnaire as it was related only to temporal framing.

175

6.4.2 Stimuli


cells: high threat level and loss frame, high threat level and gain frame, low

threat level and loss frame and low threat level and gain frame. The

advertisements also listed the environment friendly features of the watch.

In the low threat conditions, the advertisement highlighted the fact

that burning plastic waste may cause various health problems. In the high

threat condition, the threat was specific and vivid language was used to

indicate that toxins from burning waste may cause cancer or respiratory

problems. The loss frame mentioned that choosing plastic products will

accelerate air pollution and increase the chances of health hazards caused by

pollution. The gain frame emphasized that by choosing a green product one

can slow down air pollution and reduce the chances of health hazards caused

by pollution. The watch advertisement contained further details about its

biodegradability.

6.4.3 Treatment Validity

The four print advertisements were analyzed by an expert panel to

assess if it contained the necessary variations in the threat level and message

frames. This panel consisted of 3 marketing professors. The changes

suggested by the panel were made and the final versions of the advertisements

are shown in Appendix 6 (Figure A6.1, Figure A6.2, Figure A6.3, and Figure

A6.4).


Two questions were included in the questionnaire to check the

manipulations. Threat level manipulations were checked by including a

176

multiple response question: “What health problems did the advertisement

highlight? Choose only one answer”. The options were (1) Diseases (2)

Cancer (3) Respiratory problems (4) Cancer and respiratory problems. Frame

manipulations were checked asking the respondents to rate the following

questions: “I can gain health benefits by buying biodegradable products”, “I

can lose important health benefits if I don’t buy biodegradable products”. The

response to these items was measured using seven point Likert scales

anchored from 1 = Strongly Disagree and 7 = Strongly Agree.

6.4.5 Modified Dependent Variables

The study used the same dependent variables as Experiment 1.

However some of the PMT variables were modified to improve their

reliabilities. Perceived severity and self-efficacy variables were modified as

the reliabilities were low in Experiment1. Response efficacy and perceived

vulnerability were also changed to reflect the changes in the independent

factors. The variables were now adapted from the risk behaviour diagnosis

scale (Witte et al 1996), as this scale is widely used in measuring risks

associated with health messages. Apart from these changes, since the threat

levels varied in the advertisements, efficacy variables were changed to

measure generic diseases rather than specifying particular diseases like

respiratory diseases and cancer. The changed variables are described below

and the entire questionnaire (Q3) is shown in Appendix 5.

Perceived severity

Perceived severity was measured using a three item seven point


asked to indicate their responses on the following statements: “I believe that

plastic waste pollution is a serious threat to human health”, “I believe that

177

plastic waste disposal may cause severe health issues.”, “I believe that plastic

waste pollution is extremely harmful”.




the participants’ perceived susceptibility to the threat. Participants were asked

to indicate their responses on the following statements. “It is possible that I

might get affected by diseases caused by plastic waste pollution.”, “It is

probable that I will suffer from various diseases caused by plastic waste

pollution.”, “I am at risk for getting health problems caused by plastic waste

pollution.” These items were collapsed into a single perceived vulnerability

score.

Response efficacy

Response efficacy was measured using a three item seven point

scale where 1 = Strongly Disagree and 7 = Strongly Agree to determine

whether participants’ believed if purchasing biodegradable products averted

the threat. Participants rated their responses on the following statements:

“Purchasing biodegradable products is a highly effective way of preventing

diseases due to plastic pollution”, “Buying biodegradable products will

significantly lower my risk of being affected by diseases caused by plastic

pollution”, “Buying biodegradable products is an effective method of

reducing threats caused by plastic pollution to human health”. These items

were combined into a single response efficacy score.

178

Self efficacy

Self efficacy was measured using a three item seven point scales

where 1 = Strongly Disagree and 7 = Strongly Agree to determine whether

participants’ believed if they were capable of averting the threat. Participants

rated their responses on the following statements: “I am capable of identifying

and purchasing biodegradable products”, “I can easily switch over to

biodegradable products to prevent future health problems”, “It is not difficult

for me to check if products contain plastic or not”.

6.4.6 Results of Experiment 2 - Threat Levels and Goal Framing

This study was conducted with the watch stimuli to evaluate the

effect of different threat levels (low/high) and goal frames (gain vs. loss) on

the PMT variables. The effect of PMT variables on involvement and the

subsequent influence of involvement on attitudes and purchase intention were

also evaluated.

6.4.6.1 Manipulation checks

If the participant under high threat condition chose any other

answer, apart from the generic “diseases”, the manipulation was considered

successful. A chi-square test, comparing the observed frequencies of cases

with the correct evaluation of the threat with the expected frequencies,

revealed that the threat manipulation was successful only in the high threat

condition. The results can be seen in Table 6.41 below. The threat condition

that was assigned to them was correctly identified by 94.2% of the

participants in the high threat condition. This showed that the manipulation

worked for the high threat condition. Even under low threat condition

179

participants viewed the threat as high (64.7% of them viewed the threat as

high).

Table 6.41 Experiment 2: Manipulation check for threat levels

Threat_level * MC_THREAT Crosstabulation Count

MC_THREAT TotalCancer and Resp Generic

Threat_level high 31 4 35low 22 12 34

Total 53 16 69

Chi-Square TestsValue df Asymp. Sig. (2-

sided) Exact Sig. (2-sided)

Exact Sig. (1-sided)

Pearson Chi-Square

5.515a 1 .019

Continuity Correctionb

4.256 1 .039

Likelihood Ratio 5.707 1 .017Fisher's Exact Test

.024 .019

N of Valid Cases 69 a. 0 cells (.0%) have expected count less than 5. The minimum expected count is 7.88. b. Computed only for a 2x2 table

The frame manipulation was not successful as there was no significant main effect of the frame manipulation (loss vs. gain) in both the conditions (Table 6.42).

Similar to Experiment 1, unsuccessful manipulation checks were

not of great concern and did not indicate that the manipulation of the

independent variable failed (Sigall & Mills 1998). Hence further analyses

were conducted.

180

Table 6.42 Experiment 2: Manipulation check for frame type

ANOVASum of Squares df Mean Square F Sig.

MC_GAIN Between Groups 1.503 1 1.503 .535 .467Within Groups 188.265 67 2.810Total 189.768 68

MC_LOSS Between Groups 4.304 1 4.304 2.247 .139Within Groups 128.333 67 1.915Total 132.638 68

6.4.6.2 Scale reliability

The internal consistency of the scales was assessed using Cronbach

. The Table 6.43 below shows the reliability scores.

Table 6.43 Experiment 2: Reliability scores using watch stimulus

Construct Cronbach

Perceived severity 0.84

Perceived vulnerability 0.71

Response Efficacy 0.64

Self Efficacy 0.56

Message involvement 0.80

Fear 0.92

Attitude towards ad 0.86

Attitude towards brand 0.91

Purchase intention 0.91

Environmental concern 0.78

All the variables except self-efficacy had reliability scores exceeding 0.6.

Self-efficacy also has adequate reliability in this case. The results suggest that

181

the instrument was reasonably reliable. The scale reliability for perceived

severity increased with the revision.


variables

Tables 6.44 and 6.45 show the distribution characteristics and the

group wise means of the protection motivation variables. It can be seen that

the average values for perceived severity and perceived vulnerability are high

and closer to the maximum score. The group-wise means also show that there

are not much variations in the scores across the groups.

Table 6.44 Experiment 2: Distribution characteristics of the protection motivation variables (watch stimuli)

Minimum Maximum Mean Std. Deviation

PERC_SEV 1.00 7.00 6.22 1.00

PERC_VUL 2.33 7.00 5.31 1.20

RESP_EFFICACY 3.33 7.00 5.92 0.82

SELF_EFFICACY 2.00 6.67 4.88 1.26

FEAR 1.00 7.00 4.34 1.60

Table 6.45 Experiment 2: Group wise mean values of protection motivation variables for the (watch stimuli)



Response Efficacy

Self Efficacy

Fear

Threat level: High 6.25 5.32 5.90 4.91 4.25

Threat level: Low 6.17 5.29 5.94 4.85 4.42

Goal frame: Gain 6.27 5.43 6.07 5.06 4.49

Goal frame: Loss 6.16 5.19 5.78 4.72 4.20

182

To test the hypotheses regarding the effect of manipulations on the

PMT variables, MANOVA was conducted to test the hypotheses. Hypothesis

5 stated that participants who viewed advertisements with higher threats

levels would report higher severity and vulnerability when compared to

consumers who viewed weaker threats. A one-way MANOVA was conducted

to test this hypothesis. The one-way MANOVA results were: Pillai’s


Root = 0.002, F(2,66)=0.057, p > 0.05) (Table 6.46a). The results indicate

that there was no statistically significant difference in severity and

vulnerability based on threat levels and hence Table 6.46b was not further

interpreted. Therefore hypothesis 5 (H5) was not supported.


Multivariate Testsb

Effect Value F Hypothesis df Error df Sig.


Wilks' Lambda .023 1387.169a 2.000 66.000 .000


Roy's Largest Root 42.035 1387.169a 2.000 66.000 .000

Threat_level Pillai's Trace .002 .057a 2.000 66.000 .945

Wilks' Lambda .998 .057a 2.000 66.000 .945


Roy's Largest Root .002 .057a 2.000 66.000 .945

a. Exact statistic b. Design: Intercept + Threat_level

183





Square F Sig.

Corrected Model

PERC_SEV .112a 1 .112 .111 .740PERC_VUL .015b 1 .015 .010 .919


Threat_level PERC_SEV .112 1 .112 .111 .740PERC_VUL .015 1 .015 .010 .919

Error PERC_SEV 67.627 67 1.009PERC_VUL 97.167 67 1.450


Corrected Total

PERC_SEV 67.739 68

a. R Squared = .002 (Adjusted R Squared = -.013) b. R Squared = .000 (Adjusted R Squared = -.015)

The results also indicate that there was no statistically significant difference in severity and vulnerability based on frame type (Pillai’s Trace=0.010; Wilks’ lambda = 0.990; Hotelling’s Trace and Roy’s Largest Root = 0.010, F(2,66)=0.341, p > 0.05). Therefore hypothesis 6 (H6) was not supported. The following tables (Tables 6.47a and 6.47b) show the results.


Multivariate Testsb


Error df Sig.

Intercept Pillai's Trace .977 1392.001a 2.000 66.000 .000 Wilks' Lambda .023 1392.001a 2.000 66.000 .000

184


Effect Value F Hypothesisdf

Error df Sig.

Hotelling's Trace 42.182 1392.001a 2.000 66.000 .000 Roy's Largest Root 42.182 1392.001a 2.000 66.000 .000

frame Pillai's Trace .010 .341a 2.000 66.000 .712Wilks' Lambda .990 .341a 2.000 66.000 .712Hotelling's Trace .010 .341a 2.000 66.000 .712Roy's Largest Root .010 .341a 2.000 66.000 .712

a. Exact statistic b. Design: Intercept + frame





Square F Sig.

Corrected Model

PERC_SEV .194a 1 .194 .192 .663PERC_VUL .991b 1 .991 .690 .409


frame PERC_SEV .194 1 .194 .192 .663PERC_VUL .991 1 .991 .690 .409

Error PERC_SEV 67.545 67 1.008PERC_VUL 96.191 67 1.436


Corrected Total

PERC_SEV 67.739 68PERC_VUL 97.182 68

a. R Squared = .003 (Adjusted R Squared = -.012) b. R Squared = .010 (Adjusted R Squared = -.005)

It can be seen from Tables 6.48a and 6.48b that the proposed

interaction between threat levels and frames was also not supported (Pillai’s

185


Root = 0.086 (F2,64)=2.539, p>0.05). Therefore hypothesis 7 (H7) was not

supported.


Multivariate Testsb


df Error

df Sig.


Wilks' Lambda .023 1379.959a 2.000 64.000 .000


Roy's Largest Root

43.124 1379.959a 2.000 64.000 .000

frame Pillai's Trace .010 .325a 2.000 64.000 .724

Wilks' Lambda .990 .325a 2.000 64.000 .724


Roy's Largest Root

.010 .325a 2.000 64.000 .724

Threat_level Pillai's Trace .002 .061a 2.000 64.000 .941

Wilks' Lambda .998 .061a 2.000 64.000 .941


Roy's Largest Root

.002 .061a 2.000 64.000 .941

frame * Threat_level

Pillai's Trace .074 2.539a 2.000 64.000 .087

Wilks' Lambda .926 2.539a 2.000 64.000 .087

Hotelling's Trace .079 2.539a 2.000 64.000 .087

Roy's Largest Root

.079 2.539a 2.000 64.000 .087

a. Exact statistic b. Design: Intercept + frame + Threat_level + frame * Threat_level

186

From Table 6.48b it could be seen that the interaction of the independent

factors had an effect on perceived vulnerability. Hence a followup ANOVA

revealed this effect. The ANOVA results are shown in Table 6.48c. Figure

6.2 shows that gain frames and high levels of threat produced higher

vulnerability scores. Under loss frame conditions, low threat produced higher

scores of perceived vulnerability.

Table 6.48b Experiment 2: Hypothesis 7: multivariate tests (watch stimuli)




df Mean

Square F Sig.

Corrected Model

PERC_SEV .743a 3 .248 .240 .868

PERC_VUL 7.784b 3 2.595 1.886 .141


PERC_VUL 1940.361 1 1940.361 1410.803 .000

Frame PERC_SEV .180 1 .180 .175 .677

PERC_VUL .907 1 .907 .659 .420

Threat_level PERC_SEV .128 1 .128 .124 .726

PERC_VUL .050 1 .050 .036 .850

frame * Threat_level

PERC_SEV .442 1 .442 .428 .515

PERC_VUL 6.781 1 6.781 4.930 .030

Error PERC_SEV 66.996 65 1.031

PERC_VUL 89.398 65 1.375


PERC_VUL 2042.111 69

Corrected Total

PERC_SEV 67.739 68

PERC_VUL 97.182 68

a. R Squared = .011 (Adjusted R Squared = -.035) b. R Squared = .080 (Adjusted R Squared = .038)

187

Table 6.48c Experiment 2: Hypothesis 7: interaction effect of goal frames and threat levels on perceived vulnerability

Tests of Between-Subjects Effects Dependent Variable:PERC_VUL



Partial Eta

Squared

Noncent. Parameter

Observed Powerb

Corrected Model

7.784a 3 2.595 1.886 .141 .080 5.659 .467

Intercept 1940.361 1 1940.361 1410.803 .000 .956 1410.803 1.000 Threat_level .050 1 .050 .036 .850 .001 .036 .054frame .907 1 .907 .659 .420 .010 .659 .126Threat_level * frame 6.781 1 6.781 4.930 .030 .071 4.930 .590

Error 89.398 65 1.375Total 2042.111 69Corrected Total 97.182 68

a. R Squared = .080 (Adjusted R Squared = .038) b. Computed using alpha = .05

Figure 6.2 Experiment 2 – effect of threat levels and goal on perceived vulnerability

188

In order to examine H8a regression analyses was done with three

predictors: framing, environmental concern and an interaction term of these

variables with purchase intention as the dependent variable. Framing was

dummy coded with the loss-frame message condition allocated a value of 0

and the gain-frame message condition a value of 1.

The interaction terms were calculated as a product of frame type

and environmental concern (frame x environmental concern). The hypothesis

was not supported as the interaction between the variables did not predict

purchase intention. Since the model was significant, a follow up stepwise

regression revealed that only environmental concern predicted purchase

intention. Table 6.49a shows this interaction.

Table 6.49a Experiment 2: Hypothesis 8a: interaction of frame and environmental concern on purchase intention (watch stimuli)

Model Summary


Estimate

1 .0.444a 0.197 0.147 1.25387

a. Predictors: (Constant), TOTAL_ENV_CONCERN, FRAME_CODED, ENV_CONC_X_FRAME

ANOVAb


df Mean

Square F Sig.

1 Regression 21.394 3 7.131 4.461 .007a

Residual 103.901 65 1.598

Total 125.295 68

a. Predictors: (Constant), TOTAL_ENV_CONCERN, FRAME_CODED, ENV_CONC_X_FRAME b. Dependent Variable: PURCHASE_INTENTION

189


Coefficientsa

Model



t Sig.

BStd.

Error Beta

1 (Constant) -.688 4.717 -.146 .884

FRAME_CODED .127 3.294 .047 .039 .969

ENV_CONC_X_FRAME .007 .523 .018 .014 .989

TOTAL_ENV_CONCERN .886 .750 .401 1.181 .242


Similarly H8b was also not supported as message involvement did

not interact with frame type to produce an effect on purchase intentions.

Table 6.49b shows this interaction. Since the model was significant, a follow

up stepwise regression revealed that only message involvement significantly

predicted purchase intention.

Table 6.49b Experiment 2: Hypothesis 8b: interaction of frame and message involvement on purchase intention (watch stimuli)

Model Summary


Estimate

1 .0.515a 0.266 0.232 1.18986

Predictors: (Constant), MESS_INV_X_FRAME, MESSAGE_INVOLVEMENT, FRAME_CODED

190


ANOVAb


df Mean

Square F Sig.

1 Regression 33.270 3 11.090 7.833 .000a

Residual 92.025 65 1.416

Total 125.295 68

a. Predictors: (Constant), MESS_INV_X_FRAME, MESSAGE_INVOLVEMENT, FRAME_CODED b. Dependent Variable: PURCHASE_INTENTION

Coefficientsa



B Std. Error Beta

1 (Constant) 3.679 2.110 1.744 .086

FRAME_CODED -1.331 1.470 -.493 -.905 .369

MESSAGE_INVOLVEMENT .254 .400 .204 .634 .528

MESS_INV_X_FRAME .270 .276 .637 .979 .331


H8c was not tested as the sample consisted of mostly male subjects

(95.7%). Therefore gender based variations could not be investigated.

The summary of results is shown in Table 6.50. The table clearly

highlights the fact that the factors were not successful in producing the

hypothesized main effects or interaction effects.

191

Table 6.50 Experiment 2: Summary of the effect of manipulations on PMT variables with the watch stimulus



H5 Threat level X X

H6 Goal frame X X

H7 Threat level * Goal Frame X X

- -

- -

Purchase Intention

H8a Environmental concern * Goal Frame

X

H8b Message Involvement * Goal Frame

X


involvement, attitudes and intentions

Similar to Experiment 1, standard simple or multiple linear

regression (ordinary least squares (OLS)) analyses were run to ascertain the

effect of the predictor variables. Only one or two predictors were considered

at a time. Similarly assumptions regarding like multicollinearity,

homoscedascity and linearity were met in all the scenarios.


perceived severity and perceived vulnerability predicted fear. H9a was

supported as both perceived severity ( =0.318, t(66)=2.622, p<0.05) and

perceived vulnerability ( =0.273, t(66)=2.254, p<0.05) significantly predicted

fear. The model fit was also good (R2=0.261).

192

Perceived severity also predicted both the efficacy variables.

Tables 6.51a, 6.51b and 6.51c show the regression results.

Table 6.51a Experiment 2: Hypothesis 9a: effect of threat appraisal components on fear (watch stimuli)

Model Summary


1 .510a 0.261 0.238 1.39272


ANOVAb


df Mean

Square F Sig.

1 Regression 45.110 2 22.555 11.628 .000a

Residual 128.018 66 1.940

Total 173.128 68


b. Dependent Variable: FEAR

Coefficientsa

Model



B Std. Error Beta

1 (Constant) -.755 1.099 -.687 .495

PERC_SEV .508 .194 .318 2.622 .011

PERC_VUL .365 .162 .273 2.254 .028

a. Dependent Variable: FEAR

193

Table 6.51b Experiment 2: Hypothesis 9b: effect of threat appraisal components on response efficacy (watch stimuli)

Model Summary


1 .382a 0.146 0.120 0.77197


ANOVAb


df Mean

Square F Sig.

1 Regression 6.700 2 3.350 5.621 .006a

Residual 39.332 66 .596

Total 46.032 68


Coefficientsa



B Std. Error Beta

1 (Constant) 3.959 .609 6.497 .000

PERC_SEV .196 .107 .237 1.821 .073

PERC_VUL .141 .090 .205 1.569 .121


From Table 6.51b it can be seen that although the model was

significant, the predictors were not significant. Hence H9b was not supported.

A stepwise regression was conducted to evaluate which one of the variables

contributed to the model. The results showed that perceived severity alone

predicted response efficacy as shown in Table 6.51 b1 (R2=0.114 =0.337,

t(66)=2.931, p<0.01).

194

Table 6.51b1 Experiment 2: Effect of threat appraisal components on response efficacy (watch stimuli)

Model Summary


1 .337a 0.114 0.100 0.78036


ANOVAb


df Mean

Square F Sig.

1 Regression 5.232 1 5.232 8.592 .005a

Residual 40.800 67 .609

Total 46.032 68

a. Predictors: (Constant), PERC_SEV b. Dependent Variable: RESP_EFFICACY

Coefficientsa



B Std. Error Beta

1 (Constant) 4.195 .597 7.027 .000

PERC_SEV .278 .095 .337 2.931 .005


Excluded Variablesb

Model Beta In t Sig. Partial

Correlation

Collinearity Statistics

Tolerance

1 PERC_VUL .205a 1.569 .121 .190 .762

a. Predictors in the Model: (Constant), PERC_SEV b. Dependent Variable: RESP_EFFICACY

195

The results showed that perceived severity alone signficantly

predicts self efficacy as shown in Table 6.51c (R2=0.090 =0.331,

t(66)=2.459, p<0.01). Hence H9c was not supported.

Table 6.51c Experiment 2: Hypothesis 9c: effect of threat appraisal components on self efficacy (watch stimuli)

Model Summary


1 .300a 0.090 0.062 1.22004


ANOVAb


df Mean

Square F Sig.

1 Regression 9.721 2 4.861 3.265 .044a

Residual 98.240 66 1.488

Total 107.961 68

a. Predictors: (Constant), PERC_VUL, PERC_SEV b. Dependent Variable: SELF_EFFICACY

Coefficientsa



B Std. Error Beta

1 (Constant) 2.734 .963 2.839 .006

PERC_SEV .418 .170 .331 2.459 .017

PERC_VUL -.084 .142 -.080 -.593 .555

a. Dependent Variable: SELF_EFFICACY

Simple regression analyses were run with each of the PMT

variables and each one of them except self-efficacy significantly predicted

message involvement. Perceived severity ( =0.261 t(67)=2.213,

196

p<0.05,R2=0.06), perceived vulnerability ( =0.334 t(67)=2.898,

p<0.001,R2=0.111), fear ( =0.554 t(67)=5.447, p<0.001,R2=0.307), and

response efficacy ( =0.458 t(67)=4.221, p<0.05,R2 = 0.210) significantly

predict message involvement. The results are shown in the following tables

Tables 6.52a, 6.52b, 6.52c and 6.52d. Therefore H10a, H10b, H10c, H10d

were supported and H10e was not supported.

Table 6.52a Experiment 2: Hypothesis 10a: effect of perceived severity on message involvement (watch stimuli)

Model Summary


1 .261a 0.068 0.054 1.06091

a. Predictors: (Constant), PERC_SEV

ANOVAb


df Mean

Square F Sig.

1 Regression 5.512 1 5.512 4.898 .030a

Residual 75.411 67 1.126

Total 80.924 68

a. Predictors: (Constant), PERC_SEV b. Dependent Variable: MESSAGE_INVOLVEMENT

Coefficientsa



B Std. Error Beta

1 (Constant) 3.420 .812 4.214 .000

PERC_SEV .285 .129 .261 2.213 .030


197

Table 6.52b Experiment 2: Hypothesis 10b: effect of perceived vulnerability on message involvement (watch stimuli)

Model Summary


1 .0.334a 0.111 0.098 1.03598

a. Predictors: (Constant), PERC_VUL

ANOVAb


df Mean

Square F Sig.

1 Regression 9.016 1 9.016 8.400 .005a

Residual 71.908 67 1.073

Total 80.924 68


Coefficientsa



B Std. Error Beta

1 (Constant) 3.576 .572 6.255 .000

PERC_VUL .305 .105 .334 2.898 .005


Table 6.52c Experiment 2: Hypothesis 10c: effect of fear on message involvement (watch stimuli)

Model Summary


1 .554a 0.307 0.297 0.91495


198

Table 6.52c (Continued)

ANOVAb


df Mean

Square F Sig.

1 Regression 24.836 1 24.836 29.668 .000a

Residual 56.088 67 .837Total 80.924 68

a. Predictors: (Constant), FEAR b. Dependent Variable: MESSAGE_INVOLVEMENT

Coefficientsa




FEAR .379 .070 .554 5.447 .000


Table 6.52d Experiment 2: Hypothesis 10d: effect of response efficacy on message involvement (watch stimuli)



ANOVAb


df Mean

Square F Sig.

1 Regression 17.001 1 17.001 17.820 .000a

Residual 63.922 67 .954Total 80.924 68

a. Predictors: (Constant), RESP_EFFICACY b. Dependent Variable: MESSAGE_INVOLVEMENT

199


Coefficientsa



B Std. Error Beta

1 (Constant) 1.594 .861 1.852 .068

RESP_EFFICACY .608 .144 .458 4.221 .000


Table 6.52e Experiment 2: Hypothesis 10e: effect of self efficacy on message involvement (watch stimuli)


1 .236a 0.056 0.042 1.06800a. Predictors: (Constant), SELF_EFFICACY

ANOVAb


df Mean

Square F Sig.

1 Regression 4.502 1 4.502 3.947 .051a

Residual 76.422 67 1.141Total 80.924 68

a. Predictors: (Constant), SELF_EFFICACY b. Dependent Variable: MESSAGE_INVOLVEMENT

Coefficientsa




SELF_EFFICACY .204 .103 .236 1.987 .051 a. Dependent Variable: MESSAGE_INVOLVEMENT

200

Multiple regression analysis was conducted to evaluate how well

coping variables predicted attitude towards ad and purchase intention. Both

response efficacy ( =0.474, t(67)=4.409, p<0.001, R2=0.225) and self

efficacy ( =0.271, t(67)=2.306, p<0.05,R2=0.074) predicted attitude towards

ad. However they did not predict purchase intention. Therefore H11a and

H11b were supported but H11c and H11d were not. These results can be seen

in Tables 6.53a, 6.53b, 6.53c and 6.53d.

Table 6.53a Experiment 2: Hypothesis 11a: effect of response efficacy on attitude towards ad (watch stimuli)

Model Summary


1 .474a 0.225 0.213 1.01172


ANOVAb


df Mean

Square F Sig.

1 Regression 19.897 1 19.897 19.439 .000a

Residual 68.579 67 1.024

Total 88.477 68


Coefficientsa



B Std. Error Beta

1 (Constant) 1.739 .892 1.950 .055

RESP_EFFICACY .657 .149 .474 4.409 .000


201

Table 6.53b Experiment 2: Hypothesis 11b: effect of self efficacy on attitude towards ad (watch stimuli)

Model Summary


1 .271a 0.074 0.060 1.10608


ANOVAb


df Mean

Square F Sig.

1 Regression 6.508 1 6.508 5.320 .024a

Residual 81.968 67 1.223

Total 88.477 68

a. Predictors: (Constant), SELF_EFFICACY b. Dependent Variable: ATTITUDE_AD

Coefficientsa



B Std. Error Beta

1 (Constant) 4.434 .537 8.261 .000

SELF_EFFICACY .246 .106 .271 2.306 .024


Table 6.53c Experiment 2: Hypothesis 11b: effect of response efficacy on purchase intention (watch stimuli)

Model Summary


1 .185a 0.034 0.020 1.34386


202


ANOVAb


df Mean

Square F Sig.

1 Regression 4.296 1 4.296 2.379 .128a

Residual 120.999 67 1.806

Total 125.295 68


Table 6.53d Experiment 2: Hypothesis 11d: effect of self efficacy on purchase intention (watch stimuli)

Model Summary


1 .0.110a 0.012 -0.003 1.35918


ANOVAb


df Mean

Square F Sig.

1 Regression 1.521 1 1.521 .824 .367a

Residual 123.773 67 1.847

Total 125.295 68

a. Predictors: (Constant), SELF_EFFICACY b. Dependent Variable: PURCHASE_INTENTION

H12a, H12b, H12c and H12d were not supported as environmental

knowledge did not significantly predict perceived severity, perceived

vulnerability, fear or message involvement. Tables 6.54a , 6.54b 6.54c, 6.54d

illustrate the results.

203

Table 6.54a Experiment 2: Hypothesis 12a: effect of environmental knowledge on perceived severity (watch stimuli)

Model Summary


1 0.130a 0.017 0.002 0.99695


ANOVAb


df Mean

Square F Sig.

1 Regression 1.147 1 1.147 1.154 .286a

Residual 66.592 67 .994

Total 67.739 68


Table 6.54b Experiment 2: Hypothesis 12b: effect of environmental knowledge on perceived vulnerability (watch stimuli)

Model Summary


1 0.092a 0.009 -0.006 1.19921


ANOVAb


df Mean

Square F Sig.

1 Regression .829 1 .829 .577 .450a

Residual 96.353 67 1.438

Total 97.182 68


204

Table 6.54c Experiment 2: Hypothesis 12c: effect of environmental knowledge on fear (watch stimuli)

Model Summary


1 0.077a 0.006 -0.009 1.60273


ANOVAb


df Mean

Square F Sig.

1 Regression 1.021 1 1.021 .398 .530a

Residual 172.107 67 2.569

Total 173.128 68


Table 6.54d Experiment 2: Hypothesis 12d: effect of environmental knowledge on message involvement (watch stimuli)

Model Summary


1 0.103a 0.011 -0.004 1.09322


ANOVAb


df Mean

Square F Sig.

1 Regression .850 1 .850 .711 .402a

Residual 80.073 67 1.195

Total 80.924 68


205

Similar simple regression analysis was conducted to evaluate if

involvement predicted attitude towards ad. Environmental concern was

positively related to message involvement ( =0.457 t(66)=4.203, p<0.001,

R2=0.209) and therefore H13a was supported. H13b was also supported as

environmental concern significantly predicted attitude towards ad ( =0.499

t(66)=4.716, p<0.001, R2=0.249). Environmental concern significantly

predicted purchase intention ( =0.408 t(66)=3.661, p<0.001, R2=0.167) and

therefore H13c was supported. The following tables (Tables 6.55a, 6.55b,

6.55c) show the regression results.

Table 6.55a Experiment 2: Hypothesis 13a: effect of environmental concern (enduring involvement with the environment) on message involvement (watch stimuli)


1 0.457a 0.209 0.197 0.97766a. Predictors: (Constant), TOTAL_ENV_CONCERN

ANOVAb


Square F Sig.

1 Regression 16.883 1 16.883 17.664 .000a

Residual 64.040 67 .956Total 80.924 68


Coefficientsa

Model



t Sig. B Std.

Error Beta

1 (Constant) .112 1.215 .092 .927TOTAL_ENV_CONCERN .811 .193 .457 4.203 .000


206

Table 6.55b Experiment 2: Hypothesis 13b: effect of environmental concern (enduring involvement with the environment) on attitude towards the ad (watch stimuli)

Model Summary


1 0.499a 0.249 0.238 0.99571


ANOVAb


df Mean

Square F Sig.

1 Regression 22.050 1 22.050 22.241 .000a

Residual 66.426 67 .991

Total 88.477 68


Coefficientsa



B Std. Error Beta

1 (Constant) -.175 1.237 -.141 .888



Table 6.55c Experiment 2: Hypothesis 13c: effect of environmental concern (enduring involvement with the environment) on purchase intention (watch stimuli)

Model Summary


1 0.408a 0.167 0.154 1.24834


207


ANOVAb


df Mean

Square F Sig.

1 Regression 20.884 1 20.884 13.401 .000a

Residual 104.410 67 1.558

Total 125.295 68


Coefficientsa



B Std. Error Beta

1 (Constant) -.536 1.551 -.345 .731



It can be seen from Table 6.56 that H14 was supported as message

involvement predicted attitude towards ad ( =0.718 t(66)=8.452, p<0.001).

The model fit was also good with R2 =0.516.

Table 6.56 Experiment 2: Hypothesis 14: effect of message involvement on attitude towards ad (watch stimuli)

Model Summary


1 0.718a 0.516 0.509 0.79944


208


ANOVAb


df Mean

Square F Sig.

1 Regression 45.657 1 45.657 71.439 .000a

Residual 42.820 67 .639

Total 88.477 68


Coefficientsa

Model



t Sig.B

Std. Error

Beta

1 (Constant) 1.732 .471 3.674 .000



H15 and H16 were also supported as attitude towards the ad

significantly predicted the attitude towards the brand ( =0.747 t(66)=9.210,

p<0.001, R2=0.559). Table 6.57 illustrates this result. Attitude towards the

brand significantly predicted the purchase intention as seen in Table 6.58

=0.647, t(66)=6.942, p<0.001, R2=0.418).

Table 6.57 Experiment 2: Hypothesis 15: effect of attitude towards adon attitude towards brand (watch stimuli)

Model Summary


1 0.747a 0.559 0.552 0.74046


209


ANOVAb


df Mean

Square F Sig.

1 Regression 46.512 1 46.512 84.832 .000a

Residual 36.735 67 .548Total 83.246 68

a. Predictors: (Constant), ATTITUDE_AD b. Dependent Variable: ATTITUDE_BRAND

Coefficientsa




ATTITUDE_AD .725 .079 .747 9.210 .000a. Dependent Variable: ATTITUDE_BRAND

Table 6.58 Experiment 2: Hypothesis 16: effect of attitude towards brand on purchase intention (watch stimuli)


1 0.647a 0.418 0.410 1.04293a. Predictors: (Constant), ATTITUDE_BRAND

ANOVAb


df Mean

Square F Sig.

1 Regression 52.419 1 52.419 48.192 .000a

Residual 72.876 67 1.088Total 125.295 68

a. Predictors: (Constant), ATTITUDE_BRAND b. Dependent Variable: PURCHASE_INTENTION

210


Coefficientsa



B Std. Error Beta 1 (Constant) .757 .640 1.183 .241

ATTITUDE_BRAND .794 .114 .647 6.942 .000a. Dependent Variable: PURCHASE_INTENTION

Table 6.59 below shows the summary of the hypotheses tests

regarding the relationship among the PMT variables, involvement, attitudes

and purchase intention.

Table 6.59 Experiment 2: Summary of hypotheses tests regarding relationship between PMT variables, involvement, attitudes and intentions (H9 – H16) (watch stimuli)



R2

Unstandardisedcoefficient B

Standardised coefficient


Fear 0.261 0.238 0.5080.365

0.318*0.273*

H9b Perceived Severity and Perceived vulnerability

Response Efficacy

Not supported (Only perceived severity)


Self efficacy 0.300 0.090 0.418 0.331*


Message Involvement

0.06 0.05 0.285 0.261*

H10b Perceived Vulnerability

Message Involvement

0.11 0.09 0.305 0.334**


0.30 0.29 0.37 0.55***

211




R2Unstandardised



Message Involvement

0.21 0.19 0.608 0.458***


Not Significant


Attitude towards ad

0.22 0.21 0.65 0.47 ***


0.07 0.06 0.246 0.271 *


Purchase Intention

Not Significant


Not Significant


Perceived Severity

Not Significant


Perceived Vulnerabilty

Not Significant

H12c EnvironmentalKnowledge


H12d EnvironmentalKnowledge

Message Involvement

Not Significant


Message Involvement

0.20 0.19 0.81 0.45***


Attitude towards ad

0.25 0.24 0.92 0.49***


Purchase Intention

0.16 0.15 0.902 0.408***


Attitude towards ad

0.51 0.50 0.751 0.718***



0.55 0.55 0.725 0.747***


Purchase Intention

0.41 0.41 0.79 0.64***

***p <.001 **p <.01 *p <.05; n= 68

212


This experiment examined the effects of varying threat levels and

goal frames on PMT variables and the subsequent effects of the PMT

variables on message involvement, attitudes and purchase intention. The

manipulation checks were moderately successful as under higher threat

conditions, participants identified the threat levels correctly. Internal

consistency of the self-efficacy scale successfully improved to 0.56. There

was no main effect of threat levels on perceived severity of threat, perceived

vulnerability and fear. The goal frames did not influence the threat appraisal

variables as hypothesized. There was also no significant interaction effect

between the factors as hypothesized. However independent variables had an

interaction effect on perceived vulnerability. The interaction showed that gain

frames and higher threat levels increased feelings of perceived vulnerability.

This result confirms the results of previous studies (Rothman et al 1993;

Mann et al 2004) that gain frames work better in the case of preventive

behaviour. This result also shows that goal framing can be used to promote

pro-environmental behaviour by accentuating intrinsic goals related to health

and well-being (Lindenberg & Steg 2007; Pelletier & Sharp 2008).

In contrast to previous studies (Cox & Cox 2001; Meyers-Levy &

Maheswaran 2004; van ‘t Riet et al 2008;O’Keefe & Jensen 2009; Janssens

et al 2010; Updegraff 2013) loss frames did not increase threat perception.

Similarly there was no relationship between involvement and framing. Both

environmental concern and message involvement did not interact with frames

to produce an effect on purchase intention.

Perceived severity and perceived vulnerability significantly

predicted fear as proposed by PMT (Rogers & Prentice-Dunn 1997; Floyd et

al 2000) and other studies that apply this theory (Milne et al 2000; de Hoog et

al 2008). Since the participants judged the threat to be high, fear levels

213

increased and coping appraisal was also intiated (Maddux & Rogers 1983;

Boer & Seydel 1996; Milne et al 2000). In the case of the previous

experiment only response efficacy and perceived vulnerability predicted

message involvement. In this experiment all the PMT variables (perceived

severity, perceived vulnerability, fear and response efficacy) except self

efficacy significantly predicted message involvement. This shows that higher

levels of health risk increased elaboration and thereby increased their

involvement with the message as observed in previous studies (Bloch &

Richins 1983; Richins & Bloch 1986; Keller & Block 1996; de Hoog 2005).

Therefore fear can increase more effortful processing in environment related

communication (Meijnders et al 2001). The relationship between the PMT

variables and message involvement is similar to the results presented by

recent research (Cauberghe et al 2009). Environmental knowledge did not

have any effect on the hypothesized variables. Unlike the previous

experiment, environmental concern had a significant effect on message

involvement, attitudes and intentions. This confirms ELM’s proposition that

issue involvement has an effect on attitudes and intentions (Petty & Cacioppo

1986). Both environmental concern and message involvement had a

significant influence on attitudes and intention similar to other advertising

studies (Gardner 1985; Park & Young 1986; Muehling & Laczniak 1988).

The manipulation checks were comparatively successful than

Experiment1 for the threat level perception. Therefore Experiment 3 was

designed using these two factors (threat level and goal frames) by refining the

stimuli. Since high scores for perceived severity and vulnerability were

reported in both the scenarios, the threat levels were modified, such that the

low threat level contained very generic statements about pollution issues,

whereas the high threat levels highlighted the perceived vulnerability to the

threat, by mentioning risks of cancer and respiratory illness. The next

214

experiment used the issue of e-waste (based on mobile phone stimuli) to

observe if similar effects were observed.

6.5 STUDY 2: EXPERIMENT 3: THREAT LEVELS AND GOAL

FRAMING (MOBILE PHONE STIMULI)

Experiment 3 was designed to use threat levels and goal frames to

influence the PMT variables. This experiment was conducted with mobile

phones as the chosen product. The mobile phone stimulus was chosen for the

experiment and the changes to the stimulus were made based on the

information gained from Experiment 2.


A 2 (frame type: gain vs. loss) x 2 (threat level: high vs. low)

between subjects experimental design was utilized to investigate the

hypotheses. This resulted in four possible combinations of the stimuli. One

hundred and ninety undergraduate engineering students from a large South

Indian University (52.4 % male, median age=20) were randomly assigned to

the four possible conditions. Males and female respondents were represented

almost equally. The respondents’ age ranged from 18-22. The participants

were exposed to the stimulus and data collection was through a paper and

pencil questionnaire.The questionnaire (Q4) is shown in Appendix 7. On

completing the questions on the dependent variables, the respondents were

given the filler questionnaire followed by the counterbalanced questionnaires

related to the environment (environmental concern and environmental

knowledge). This was similar to the questions in Experiment 1. The

personality variable CFC was also not included in this questionnaire.

215

6.5.2 Stimuli and Treatment Validity


cells: high threat level and gain frame; high threat level and loss frame; low

threat level and gain frame; low threat level and loss frame. The

advertisements listed the features of the green mobile phone and specified that

the mobile phone is 82% recyclable. In Experiment 2, the participants judged

low levels of threat as a severe threat. Hence in this experiment the threat

levels were toned down in the low threat level conditions.

In the low threat condition, the advertisement emphasized that e-

waste was difficult to dispose and did not specifically mention a health threat.

In the high threat condition, the message emphasized health threats like

respiratory illness and highlighted personal vulnerability towards the threat.

The gain frame exhorted the respondents to protect themselves and the loss

frame message highlighted the potential losses incurred when not purchasing

a green product. These were also made stronger. The advertisements were

shown to the panel as in the previous experiments to check their validity. The

advertisements are presented in Appendix 8 (Figure A8.1, Figure A8.2, Figure

A8.3 and Figure A8.4).


Threat level manipulations were checked by verifying if the

perceived severity and perceived vulnerability varied for different threat

levels. Frame manipulations were checked by asking the respondents (similar

to Experiment 2) to rate the following questions: “I can gain health benefits

by buying recylable products”, “I can lose important health benefits if I don’t

buy recylable products”. The response to this items was measured using seven

point Likert scales anchored from 1 = Strongly Disagree and 7 = Strongly

Agree.

216

6.5.4 Dependent Variables

Most of the dependent variables were the same as Experiment 2.

The items were changed to reflect the issue of e-waste and changes are

described below.

6.5.4.1 Protection motivation theory variables

Perceived severity

Perceived severity was measured using a three item seven point


asked to indicate their responses on the following statements. “I believe that

e-waste pollution is a serious threat to human health”, “I believe that e-waste

disposal may cause severe health issues”, “I believe that e-waste pollution is

extremely harmful”. The three items were averaged into a single perceived

severity score.




the participants’ susceptibility to the threat. Participants were asked to

indicate their responses on the following statements: “It is possible that I

might get affected by diseases caused by e-waste pollution.”, “It is probable

that I will suffer from various diseases caused by e-waste pollution”, “I am at

risk for getting health problems caused by e-waste pollution”. These items

were collapsed into a single perceived vulnerability score.

217

Response efficacy

Response efficacy was also measured using a three item seven


whether participants’ believed if purchasing recyclable products averted the

threat. Participants rated their responses on the following statements:

“Purchasing recyclable products is a highly effective way of preventing

diseases caused by e-waste pollution”, “Buying recyclable products will

significantly lower my risk of being affected by diseases caused by e-waste

pollution”, “Buying recyclable products is an effective method of reducing

threats to human health caused by e-waste pollution” These items were

combined into a single response efficacy score.

Self efficacy

Self efficacy was measured using a three item seven point

scaleswhere 1 = Strongly Disagree and 7 = Strongly Agree to determine

whether participants’ believed if they were capable of averting the threat.

Participants rated their responses on the following statements: “I am capable

of checking if products contain recyclable materials”, “I can easily switch

over to recyclable products to prevent future health problems”, “I can identify

and purchase recyclable products” These items were combined into a single

self efficacy score. The rest of the dependent variables remained the same as

Experiment 2. The questionnaire (Q4) is shown in Appendix 7 as discussed

previously.

6.5.5 Demographics

There were very few demographic variables collected from the

participants, as these variables were not the major focus of the research

questions. Table 6.60 shows the demographic details of the sample based on

218

the usable responses out of the 190 students. Male and female students were

represented in almost equal proportions. Most of the students were aged 20

and above.

6.5.6 Data Screening

Unlike previous experiments, this study planned to use PLS-SEM.

Hence to ensure data integrity, data screening was conducted using SPSS 19

to ensure the validity of the data prior to hypotheses testing. Data was first

screened for missing values and outliers (univariate and multivariate). Apart

from this, other multivariate statistical assumptions (normality, linearity and

homoscedasticity) were also investigated.

Table 6.60 Demographic details

Variable Count %

Gender MaleFemale Total

9685181

4753100

Age1819202122

13596445

0.619.353.024.32.8

Course Civil Engineering Electrical and Electronics EngineeringInformation Technology Mining Engineering Printing Technology

4256311537

23.230.917.18.320.4

219

6.5.6.1 Missing Value Analysis

On inspection, there were very few missing values. Only one value

was missing in the dependent variable. Hair et al (2009) recommend mean

imputation as a suitable method for replacing missing values in such cases.

Hence mean imputation was used to replace this missing value. After

imputation of the missing value, the items were summated to determine the

composite score of the variables.

6.5.6.2 Outlier Analysis

Univariate outliers

Box plots were used to identify univariate outliers among

dependent variables. Outlier analyses were conducted on composite variables

to reduce the effect any variations that single indicators might cause. Few

outliers were identified.

Multivariate outliers

Multivariate outliers are unusual combinations of variable values

(Hair et al 2009). Mahalanobis' distance was used to identify multivariate

outliers. A conservative level of significance of 0.001 was used to identify

outliers. Four cases were identified as outliers.

6.5.6.3 Decision regarding outliers

Since there were no data entry errors or other anomalies, univariate

outliers were not deleted as deleting them might impact the generalizability of

the data (Hair et al 2009). Only the multivariate outliers were removed as they

represented a small proportion (2.1%) of the total number of cases. This

resulted in 181 usable responses.

220

6.5.6.4 Univariate normality

Skewness and kurtosis were assessed to determine univariate

normality. Certain variables (perceived severity, response efficacy, self

efficacy, message involvement) did not follow a normal distribution.

For instance, respondents perceived high severity in both high and low threat

conditions and therefore the perceived severity value was negatively skewed.

The skewness and kurtosis values are shown in Table 6.61.

Table 6.61 Univariate normality

Mean Skewness Kurtosis

Statistic Statistic Zskew Statistic Zkurtosis

PERC_SEV 5.6427 -.900 -4.97238 .696 1.938719

PERC_VUL 4.7403 -.681 -3.76243 .402 1.119777

RESP_EFFICACY 5.6446 -.816 -4.50829 1.287 3.584958

SELF_EFFICACY 4.6961 -.330 -1.8232 -.434 -1.20891

MESSAGE_INVOLVEMENT 5.1860 -.766 -4.23204 .755 2.103064

FEAR 3.9945 -.221 -1.22099 -.563 -1.56825

ATTITUDE_AD 5.3168 -.632 -3.49171 .500 1.392758

ATTITUDE_BRAND 5.1731 -.321 -1.77348 -.341 -0.94986

PURCHASE_INTENTION 4.5783 -.544 -3.00552 -.143 -0.39833

ENV_KNOW 6.6133 .360 1.98895 -.076 -0.2117

TOTAL_ENV_CONCERN 5.9982 -1.420 -7.8453 3.451 9.612813

Most of the skewness and kurtosis values range from -1 to +1,

except environmental concern. Further tests showed that Z-values of the

skewness and kurtosis values of most variables were negative. This is not

surprising as the computed mean values are very high for the threat and fear

variables. Hair et al (2009) also recommend Z-tests for testing the skewness

221

of the variables and based on significance at 0.01 levels, seven variables

exceeded the value of 2.58. Most variables are negatively skewed.

6.5.6.5 Multivariate Statistical Assumptions

Multivariate normality

Multivariate normality was assessed based on Mardia’s coefficient

(Mardia 1970) using IBM AMOS 18. A high critical ratio of the coefficient

(26.512) indicated that data was significantly not normal as it exceeded the

cut-off value of 5.0 as suggested by Bentler (2006).

Linearity and homoscedasticity

Linearity and homoscedasticity were assessed among the variables

by using the regression residual and scatter plots. The variables met the

assumptions of linearity and homoscedasticity.

6.5.6.6 Conclusions from Data Screening

There was very few missing data and data imputation was done to

handle missing data. Data analysis showed that most variables were

negatively skewed and had a non-normal distribution. MANOVA is robust to

the violations of normality (Leech et al 2011) and therefore hypotheses

involving MANOVA were conducted using IBM SPSS 19. This study was

planned to be analysed using partial least squares based SEM (PLS-SEM) and

therefore this choice ensured that non normal data distribution did not pose a

problem for further analysis. Additionally the fact that PLS also works well

for a series of cause and effect relationships is to the study’s advantage

(Bontis et al 2007). PLS bootstrap also provides a more accurate and efficient

estimation of structural model parameters when compared to MLE and

222

Bollen-Stine SEM bootstraps when there are fewer than 200 observations

(Sharma & Kim 2013).

6.5.7 Validity and reliability analyses

Validity and reliability was assessed in two different ways. For the

first level of analysis, to assess the effect of the factors on the PMT variables

MANOVA is to be used. Hence Cronbach’s & EFA (Exploratory Factor

Analysis) were used to analyse the reliability and validity of the PMT

variables.

Next, the results of path analysis were to be analysed using PLS-

SEM. Hence, the measurement model was checked to ensure the reliability

and validity criteria associated with the formative and reflective measurement

model.

6.5.7.1 Validity and reliability of the PMT variables

Exploratory Factor Analysis (EFA) was used to examine

component loadings for the PMT constructs. The prescribed minimum sample

size for EFA is 100 (Hair et al 2009). On completion of the EFA, scale

reliabilities were assessed using the reliability coefficient (Cronbach ). The

PMT constructs (perceived severity, perceived vulnerability, fear, response

efficacy and self efficacy) loaded on five factors. Since PMT hypothesizes

close relationship between the constructs, oblique rotation was used and

factor loading above 0.4 was used to interpret each factor (Wu et al 2005).

None of the items loaded on more than one factor. Next, the internal

reliabilities of the constructs were tested using Cronbach’s and were found

adequate.

223

Scale reliability of the PMT constructs

Table 6.62 lists the reliabilities of the PMT constructs used. All the

scales had adequate reliabilities except response efficacy which had a

moderate reliability of 0.56. However, other PMT studies have also used this

variable despite achieving such moderate reliability scores (e.g. Milne et al

2002; Wu et al 2005; Daley et al 2009). Hence, this variable was retained in

the study.

Table 6.62 Experiment 3: Means and reliabilities of the PMT constructs

Scale Items Item-total correlation ( ) MEAN SDPerceived Severity 0.67 5.65 0.93

Perceived Vulnerability 0.62 4.76 1.13

Response Efficacy 0.56 5.64 0.82

Self Efficacy 0.61 4.7 1.18

Fear 0.87 4.0 1.321

6.5.7.2 Reliability of the reflective constructs by assessing the

measurement/outer model

Indicator reliability and composite reliability were used to assess

the reliability of the reflective measurement model. Cronbach’s exceeded

0.6 for all constructs (Table 6.62), except response efficacy which had a

moderate reliability as discussed previously. Composite reliability was now

used to prioritise indicators during estimations. Indicator reliability was

checked by examining the indicator loadings. Table 6.63 shows the indicator

loading and the composite reliability. The loadings ranged from 0.40 to 0.94

and most of them exceeded 0.70 (Fornell & Larcker 1981). Loadings below

0.7 are candidates for deletion if deleting these indicators leads to an increase

in composite reliability above the threshold value of 0.70. However composite

224

reliability scores ranged from 0.76 to 0.95 and therefore exceeded the 0.7 cut-

off (Hair et al 2011; Hair et al 2012). Only MI2 (“The ad's message seemed

relevant to me.”) from the message involvement scale had a low value but it

was not necessary to remove this indicator as the AVE and CR values were

above requisite cut-off criteria (Hair et al 2011) .

Common method bias

Common method variance is a potential problem in social science

research as data is collected through surveys based on self reporting. This

research tried to minimise the effect of common method bias by (a) increasing

the clarity of questions by using iterative pretests (b) not collecting sensitive

personal data as it might induce social desirability responses (Herath & Rao

2009; Mohan et al 2013). Harman’s One-Factor Test was employed to test if

a single factor emerged from the analysis or if a general factor explained

majority of the variance (Podsakoff et al 2003). The results of an exploratory

factor analysis on IBM SPSS 19 showed that multiple factors were present

and the major factor accounted for only 29 % of the total variance. However

this method has its own limitations (Podsakoff et al 2003; Chin et al 2012).

Therefore the correlations matrix of the latent variables was observed and the

largest correlation was 0.65 which is lower than the correlations that suggest

common method bias (r > 0.9).

6.5.7.3 Validity

Convergent and discriminant validity was also assessed for the

measurement model. The average variance extracted (AVE) was higher than

the requisite 0.50 for all the constructs except environmental concern which is

a second order construct. The values are shown in Table 6.63. Discriminant

validity was evaluated by the examination of the cross loading of the variable

and the Forner-Larcker criterion.

225

Table 6.63 Experiment 3: Composite reliability, indicator reliability and convergent validity

Construct Indicator Outer

Loading>0.7

Composite Reliability

>0.7

AVE

>0.5Perceived Severity PS1 0.8361 0.824 0.619

PS2 0.8671PS3 0.6251

Perceived Vulnerability PV1 0.8741 0.791 0.568PV2 0.5212PV3 0.8177

Response Efficacy RE1 0.7463 0.778 0.539RE2 0.7732RE3 0.6812

Self Efficacy SE1 0.6773 0.761 0.525SE2 0.5718SE3 0.8864

Fear F1 0.8286 0.904 0.661F2 0.8871F3 0.8813F4 0.7723F5 0.6578

Message Involvement MI1 0.6895 0.859 0.512MI2 0.4081MI3 0.7837MI4 0.7461MI5 0.7964MI6 0.7968

Attitude towards brand AAB1 0.917 0.951 0.866AAB2 0.9402AAB3 0.935

Attitude towards ad AAD1 0.8815 0.901 0.753AAD2 0.8629AAD3 0.8559

226


Construct Indicator Outer

Loading>0.7

Composite Reliability

>0.7

AVE

>0.5Purchase Intention PI1 0.9222 0.949 0.860

PI2 0.9301PI3 0.933

Biospherical-Concern Animals 0.711 0.844 0.573Birds 0.778Plants 0.803

Children 0.741Egoistic-concern Me 0.833 0.844 0.538

My Future 0.839My Health 0.708

My LifeStyle 0.643Altruistic - Others All People 0.715 0.805 0.626

My children 0.684People in

my country 0.714

Marine 0.736Environmental concern Bio 0.859 0.604

Egoistic Altruistic

The AVE of the latent construct must be greater than the latent

construct’s highest squared correlation with other constructs (Fornell &

Larcker 1981). It can be seen from Table 6.64 that the Forner-Larcker

criterion is satisfied. Table 6.65 shows the details of the second-order

construct. Here too, the criterion was met.

The main diagonal in Table 6.64 and 6.65 show the AVE of the

constructs. The scales satisfied the discriminant validity criteria. The loadings

and cross-loading of item to other constructs were also inspected to evaluate

227

discriminant validity. Items loaded more on their constructs when compared

to other constructs as required.

Table 6.64 Experiment 3: Discriminant validity

AAD AAB FEAR MI PS PV PI RE SE

AAD 0.753

AAB 0.498 0.866

FEAR 0.071 0.068 0.661

MI 0.313 0.281 0.26 0.512

PS 0.0595 0.091 0.148 0.127 0.619

PV 0.053 0.094 0.086 0.089 0.164 0.568

PI 0.243 0.428 0.132 0.231 0.075 0.034 0.86

RE 0.088 0.071 0.094 0.181 0.261 0.158 0.047961 0.539

SE 0.033 0.035 0.024 0.071 0.076 0.031 0.09 0.127449 0.525

Table 6.65 Experiment 3: Discriminant validity of the second order construct (environmental concern)

Egoistic Altruistic Biospheric

Egoistic 0.538

Altruistic 0.099 0.573

Biospheric 0.130 0.298 0.626

The analysis shows that the reflective measurement model for the

(both first-order and second-order) variables used in this research are reliable

and valid.

Validity of the formative construct

Hair et al (2011) recommend the examination of convergent

validity, collinearity among indicators and use previous theory to retain

indicators that do not have significant outer loading to asses the formative

228

constructs and indicators. The weight of the indicators was examined by

resampling using bootstrapping (181 observations per subsample, 5,000

subsamples and no sign changes) in SmartPLS as this is the primary statistic

for examining the indicators (Hair et al 2012). The t-values were significant

for most of the indicators except EK7, EK8, EK14 and EK15 (p<0.05). Hence

these indicators were removed from the model. Subsequently variance

inflation factor (VIF) was used to test the multicollinearity among the

remaining environment knowledge indicators. The results show minimal

collinearity among the indicators as the VIF of all items ranged between

1.073 and 1.33, below the common cut off value of 5. Therefore, the

assumption of multicollinearity was not violated (Chin, 2010).

6.5.8 Results of Experiment 3: threat levels and goal framing

This experiment was conducted to verify goal framing and threat

levels. The mobile phone stimulus was chosen, since electronic waste seemed

to be a less familiar issue when compared to biodegradability.

This study was conducted with the mobile phone stimuli to evaluate

the effect of different threat levels (low/high) and goal frames (gain vs. loss)

on the PMT variables. The effect of PMT variables on involvement and the

subsequent influence of involvement on attitudes and purchase intention were

evaluated using the path model.

6.5.8.1 Manipulation checks

One-way ANOVA was conducted to examine the effectiveness of

the message in manipulating the perceived severity, perceived vulnerability

based on threat levels. In both the threat levels, the mean for the perceived

severity remained above 5.5 and perceived vulnerability scores ranged above

4.9. Therefore there was no statistically significant effect of the threat level on

229

these variables. Similarly the manipulation checks of the frames indicated that

there was no statistically significant effect of the frames on the variables

included for manipulation checks. Tables 6.66a, 6.66b, 6.66c and 6.66d show

the results of the manipulation check.

Table 6.66a Experiment 3: Manipulation check: effect of threat level on perceived severity

Descriptive Statistics Dependent Variable:PERCEIVED SEVERITY

Threat Level Mean Std. Deviation Nhigh 5.7753 .88731 89low 5.5145 .96383 92

Total 5.6427 .93362 181

Tests of Between-Subjects Effects Dependent Variable:PERCEIVED SEVERITY

Source Type III Sum of Squares df Mean Square F Sig.Corrected Model 3.077a 1 3.077 3.580 .060Intercept 5765.935 1 5765.935 6709.822 .000Threat_level 3.077 1 3.077 3.580 .060Error 153.820 179 .859Total 5920.000 181Corrected Total 156.896 180a. R Squared = .020 (Adjusted R Squared = .014)

Table 6.66b Experiment 3: Manipulation check: effect of threat level on perceived vulnerability

Descriptive Statistics

Dependent Variable:PERCEIVED VUNERABILITY

Threat Level Mean Std. Deviation N

High 4.8052 1.02840 89

low 4.6775 1.24177 92

Total 4.7403 1.14049 181

230



Dependent Variable:PERCEIVED VULNERABILITY

Source Type III Sum of Squares df Mean Square F Sig.

Corrected Model .738a 1 .738 .566 .453

Intercept 4067.903 1 4067.903 3119.890 .000

Threat_level .738 1 .738 .566 .453

Error 233.391 179 1.304

Total 4301.333 181



Manipulation checks also revealed that the frame manipulation was

not successful as there was no significant main effect of the frame

manipulation (loss vs. gain) in both the conditions (Table 6.66c and 6.66d).

Table 6.66c Experiment 3: Manipulation check: effect of frame type (gain)


Dependent Variable:MC_GAIN



Intercept 5045.855 1 5045.855 3251.132 .000

frame_type .816 1 .816 .526 .469

Error 277.813 179 1.552

Total 5328.000 181



231

Table 6.66d Manipulation check: effect of frame type (loss)


Dependent Variable:MC_LOSS



Intercept 3916.154 1 3916.154 1520.422 .000

frame_type .021 1 .021 .008 .928

Error 461.051 179 2.576

Total 4378.000 181



Similar to the other experiments , failed manipulation checks were

not of great concern and therefore further analysis on the data was conducted.


variables

The hypothesized effect of goal frames and threat levels on the

PMT variables was analyzed using MANOVA. The dependent variables

(perceived severity, perceived vulnerability, fear, response-efficacy and self-

efficacy) were only moderately correlated (0.21 – 0.51) and therefore there

was no risk of multicollinearity to pose a hindrance to MANOVA. Tables

6.67 and 6.68 show the distribution characteristics and the group wise means

of the protection motivation variables. Similar to previous experiments, the

perceived severity and perceived vulnerability to the threat are on the higher

side. The group wise means do not seem to differ much similar to the

previous experiments.

232

Table 6.67 Experiment 3: Distribution characteristics of the protection motivation variables watch stimuli

Minimum Maximum Mean Std. Deviation

PERC_SEV 2.67 7.00 5.64 0.93

PERC_VUL 1.00 7.00 4.74 1.14

RESP_EFFICACY 2.33 7.00 5.64 0.83

SELF_EFFICACY 1.67 7.00 4.70 1.20

FEAR 1.00 7.00 3.99 1.33

Table 6.68 Experiment 3: Group wise mean values of protection motivation variables for the mobile phone stimuli



Response Efficacy

SelfEfficacy

Fear

Threat level: High

5.77 4.80 5.71 4.78 3.98

Threat level: Low

5.51 4.67 5.57 4.60 4.00

Goal frame: Gain

5.72 4.72 5.71 4.66 3.86

Goal frame: Loss

5.56 4.75 5.57 4.72 4.12

A one-way MANOVA was conducted to test hypothesis 5 (H5) that

stated that participants who viewed advertisements with higher threats levels

would report higher levels of severity and vulnerability when compared to

consumers who viewed weaker threats. The results did not show significant

differences between the groups (Pillai’s Trace=0.20; Wilks’ lambda = 0.980;

Hotelling’s Trace and Roy’s Largest Root = 0.020, F(2,178) =1.782, p >0.05)

and hence the hypothesis was not supported (Table 6.69a and 6.69b).

233


Multivariate Testsb


df Error

df Sig.


Wilks' Lambda .023 3712.945a 2.000 178.000 .000

Hotelling's Trace

41.718 3712.945a 2.000 178.000 .000

Roy's Largest Root

41.718 3712.945a 2.000 178.000 .000

Threat_level Pillai's Trace .020 1.782a 2.000 178.000 .171

Wilks' Lambda .980 1.782a 2.000 178.000 .171

Hotelling's Trace

.020 1.782a 2.000 178.000 .171

Roy's Largest Root

.020 1.782a 2.000 178.000 .171

a. Exact statistic b. Design: Intercept + Threat_level





df Mean

Square F Sig.

Corrected Model

Perceived severity

3.077a 1 3.077 3.580 .060

Perceived vunerability

.738b 1 .738 .566 .453

Intercept Perceived severity

5765.935 1 5765.935 6709.822 .000


4067.903 1 4067.903 3119.890 .000

234





df Mean

Square F Sig.

Threat_level Perceived severity

3.077 1 3.077 3.580 .060


.738 1 .738 .566 .453

Error Perceived severity

153.820 179 .859


233.391 179 1.304

Total Perceived severity

5920.000 181


4301.333 181

Corrected Total

Perceived severity

156.896 180


234.129 180


Hypothesis 6 (H6) was not supported as the results indicated that

there was no statistically significant difference in severity and vulnerability

based on frame type (Pillai’s Trace=0.010; Wilks’ lambda = 0.990;

Hotelling’s Trace and Roy’s Largest Root = 0.010, F(2,178) = 0.872, p >0.05)

(Table 6.70a and 6.70b). Therefore frame type did not increase threat

perception.

235


Multivariate Testsb


df Error

df Sig.


Wilks' Lambda .024 3659.959a 2.000 178.000 .000

Hotelling's Trace

41.123 3659.959a 2.000 178.000 .000

Roy's Largest Root

41.123 3659.959a 2.000 178.000 .000

frame_type Pillai's Trace .010 .872a 2.000 178.000 .420

Wilks' Lambda .990 .872a 2.000 178.000 .420

Hotelling's Trace

.010 .872a 2.000 178.000 .420

Roy's Largest Root

.010 .872a 2.000 178.000 .420

a. Exact statistic b. Design: Intercept + frame_type





df Mean

Square F Sig.

Corrected Model

Perceived severity

1.124a 1 1.124 1.292 .257


.054b 1 .054 .041 .840


5764.188 1 5764.188 6623.707 .000


4065.598 1 4065.598 3109.007 .000

236





df Mean

Square F Sig.

frame_type Perceived severity

1.124 1 1.124 1.292 .257


.054 1 .054 .041 .840


155.772 179 .870


234.075 179 1.308


5920.000 181


4301.333 181

Corrected Total

Perceived severity

156.896 180


234.129 180


The proposed interaction between threat levels and frames was also

not supported (Pillai’s Trace=0.011; Wilks’ lambda = 0.989; Hotelling’s

Trace and Roy’s Largest Root = 0.011, F(3,175) = 0.630, p >0.05).

Therefore hypothesis 7 (H7) was not supported. Tables 6.71a and 6.71b show

the MANOVA results. Therefore the factors did not interact to produce any

significant results.

237


Multivariate Testsb


df Error

df Sig.


Wilks' Lambda .023 3720.629a 2.000 176.000 .000

Hotelling's Trace

42.280 3720.629a 2.000 176.000 .000

Roy's Largest Root

42.280 3720.629a 2.000 176.000 .000

frame_type Pillai's Trace .010 .876a 2.000 176.000 .418

Wilks' Lambda .990 .876a 2.000 176.000 .418

Hotelling's Trace

.010 .876a 2.000 176.000 .418

Roy's Largest Root

.010 .876a 2.000 176.000 .418


Wilks' Lambda .980 1.810a 2.000 176.000 .167

Hotelling's Trace

.021 1.810a 2.000 176.000 .167

Roy's Largest Root

.021 1.810a 2.000 176.000 .167

frame_type * Threat_level

Pillai's Trace .010 .901a 2.000 176.000 .408

Wilks' Lambda .990 .901a 2.000 176.000 .408

Hotelling's Trace

.010 .901a 2.000 176.000 .408

Roy's Largest Root

.010 .901a 2.000 176.000 .408

a. Exact statistic b. Design: Intercept + frame_type + Threat_level + frame_type * Threat_level

238






Corrected Model

Perceived severity 5.210a 3 1.737 2.027 .112

Perceived vunerability 2.467b 3 .822 .628 .598

Intercept Perceived severity 5766.088 1 5766.088 6728.366 .000

Perceived vunerability 4065.379 1 4065.379 3106.127 .000

frame_type Perceived severity 1.140 1 1.140 1.330 .250

Perceived vunerability .046 1 .046 .035 .851

Threat_level Perceived severity 3.115 1 3.115 3.635 .058

Perceived vunerability .777 1 .777 .593 .442


Perceived severity 1.029 1 1.029 1.201 .275

Perceived vunerability 1.673 1 1.673 1.278 .260

Error Perceived severity 151.686 177 .857

Perceived vunerability 231.662 177 1.309

Total Perceived severity 5920.000 181

Perceived vunerability 4301.333 181

Corrected Total Perceived severity 156.896 180

Perceived vunerability 234.129 180


239

A regression analysis was done with three predictors: framing,

environmental concern and the interaction term with purchase intention as the

dependent variable to test H8a. Framing was dummy coded with the loss-

frame message condition allocated a value of 0 and the gain-frame message

condition a value of 1. The interaction terms were calculated as a product of

frame type and environmental concern (frame x environmental concern) from

these variables. It can be seen from Table 6.72 that the hypothesis was not

supported as interaction between the variables did not predict purchase

intention. Hence the two variables did not have the hypothesized effect.

Table 6.72a Experiment 3: Hypothesis 8a: interaction of frame and environmental concern on purchase intention (mobile phone stimuli)

Model Summary


Estimate

1 .0.127a 0.016 -0.001 1.45332

a. Predictors: (Constant), TOTAL_ENV_CONCERN, FRAME_CODED, ENV_CONC_X_FRAME

ANOVAb


df Mean

Square F Sig.

1 Regression 6.123 3 2.041 .966 .410a

Residual 373.796 177 2.112

Total 379.919 180

a. Predictors: (Constant), TOTAL_ENV_CONCERN, frame_coded, frame_x_env_concern

b. Dependent Variable: PURCHASE_INTENTION

240


Coefficientsa

Model



t Sig. B

Std.Error

Beta

1 (Constant) 5.118 2.463 2.078 .039

frame_coded -1.237 1.682 -.427 -.735 .463

frame_x_env_concern .186 .278 .413 .668 .505

TOTAL_ENV_CONCERN -.059 .407 -.033 -.146 .884


Similarly Table 6.72b shows that H8b was not supported as message involvement did not interact with frame type to produce an effect on purchase intentions. However, the model was significant, a follow up stepwise regression revealed that only message involvement significantly predicted purchase intention.

Table 6.72b Experiment 3: Hypothesis 8b: interaction of frame and message involvement on purchase intention (mobile phone stimuli)


1 .0.474a 0.225 0.212 1.28982Predictors: (Constant), frame_x_MI, MESSAGE_INVOLVEMENT, frame_coded

ANOVAb


Square F Sig.

1 Regression 85.455 3 28.485 17.122 .000a

Residual 294.464 177 1.664Total 379.919 180

a. Predictors: (Constant), frame_x_MI, MESSAGE_INVOLVEMENT, frame_coded b. Dependent Variable: PURCHASE_INTENTION

241


Coefficientsa

Model



t Sig. B Std.

Error Beta

1 (Constant) .830 1.618 .513 .609frame_coded .043 1.060 .015 .040 .968MESSAGE_INVOLVEMENT .768 .307 .509 2.500 .013frame_x_MI -.038 .201 -.079 -.189 .850


A one-way MANOVA analysis of the factors and gender showed

significant differences in risk perception based on gender. Therefore H8c was

supported. Tables 6.72c and 6.72d show the results. It can be seen that gender

has an effect on perceived severity and fear independently ((Pillai’s


Root = 0.061, F(3,171) = 3.495, p <0.05). Gender also interacts with the

factors to produce an effect on the perceived severity of threat (Pillai’s


Root = 0.050, F(3,171) = 2.834, p <0.05).

Table 6.72c Experiment 3: Hypothesis 8c: multivariate tests (mobile phone stimuli)

Multivariate Testsb


dfError df Sig.


Wilks' Lambda

.022 2572.816a 3.000 171.000 .000

Hotelling's Trace

45.137 2572.816a 3.000 171.000 .000

Roy's Largest Root

45.137 2572.816a 3.000 171.000 .000

242


Multivariate Testsb


dfError df Sig.

frame_type Pillai's Trace .032 1.905a 3.000 171.000 .131

Wilks' Lambda

.968 1.905a 3.000 171.000 .131

Hotelling's Trace

.033 1.905a 3.000 171.000 .131

Roy's Largest Root

.033 1.905a 3.000 171.000 .131


Wilks' Lambda

.975 1.471a 3.000 171.000 .224

Hotelling's Trace

.026 1.471a 3.000 171.000 .224

Roy's Largest Root

.026 1.471a 3.000 171.000 .224

GENDER Pillai's Trace .058 3.495a 3.000 171.000 .017

Wilks' Lambda

.942 3.495a 3.000 171.000 .017

Hotelling's Trace

.061 3.495a 3.000 171.000 .017

Roy's Largest Root

.061 3.495a 3.000 171.000 .017


Pillai's Trace .009 .493a 3.000 171.000 .688

Wilks' Lambda

.991 .493a 3.000 171.000 .688

Hotelling's Trace

.009 .493a 3.000 171.000 .688

Roy's Largest Root

.009 .493a 3.000 171.000 .688

243


Multivariate Testsb


dfError df Sig.

frame_type * GENDER

Pillai's Trace .017 .966a 3.000 171.000 .410

Wilks' Lambda

.983 .966a 3.000 171.000 .410

Hotelling's Trace

.017 .966a 3.000 171.000 .410

Roy's Largest Root

.017 .966a 3.000 171.000 .410

Threat_level * GENDER

Pillai's Trace .001 .041a 3.000 171.000 .989

Wilks' Lambda

.999 .041a 3.000 171.000 .989

Hotelling's Trace

.001 .041a 3.000 171.000 .989

Roy's Largest Root

.001 .041a 3.000 171.000 .989

frame_type * Threat_level * GENDER

Pillai's Trace .047 2.834a 3.000 171.000 .040

Wilks' Lambda

.953 2.834a 3.000 171.000 .040

Hotelling's Trace

.050 2.834a 3.000 171.000 .040

Roy's Largest Root

.050 2.834a 3.000 171.000 .040

a. Exact statistic

b. Design: Intercept + frame_type + Threat_level + GENDER + frame_type * Threat_level + frame_type * GENDER + Threat_level * GENDER + frame_type * Threat_level * GENDER

244

Table 6.73d Experiment 3: Hypothesis 8c: tests of between-subjects effects (mobile phone stimuli)



dfMean

Square F Sig.

Partial Eta

Squared

Noncent. Parameter

Observed Powerb

Corrected Model

Perceived severity

16.307a 7 2.330 2.867 .007 .104 20.066 .917


4.223c 7 .603 .454 .866 .018 3.178 .196

Fear 19.299d 7 2.757 1.588 .142 .060 11.119 .649


5760.417 1 5760.417 7088.373 .000 .976 7088.373 1.000


4044.202 1 4044.202 3043.185 .000 .946 3043.185 1.000

Fear 2893.106 1 2893.106 1666.937 .000 .906 1666.937 1.000

Threat_level Perceived severity

3.036 1 3.036 3.736 .055 .021 3.736 .485


.893 1 .893 .672 .414 .004 .672 .129

Fear .006 1 .006 .003 .954 .000 .003 .050

Frame_type Perceived severity

1.223 1 1.223 1.505 .222 .009 1.505 .230


.057 1 .057 .043 .836 .000 .043 .055

Fear 3.577 1 3.577 2.061 .153 .012 2.061 .298

GENDER Perceived severity

4.382 1 4.382 5.392 .021 .030 5.392 .637


.005 1 .005 .004 .950 .000 .004 .050

Fear 12.505 1 12.505 7.205 .008 .040 7.205 .761

Threat_level *Frame_type

Perceived severity

.559 1 .559 .688 .408 .004 .688 .131


1.563 1 1.563 1.176 .280 .007 1.176 .190

Fear .023 1 .023 .013 .909 .000 .013 .051

Threat_level * GENDER

Perceived severity

.063 1 .063 .077 .782 .000 .077 .059


.003 1 .003 .002 .962 .000 .002 .050

Fear .145 1 .145 .084 .773 .000 .084 .060

245



Type IIISum of Squares

dfMean

Square F Sig.

Partial Eta

Squared

Noncent. Parameter

Observed Powerb

Frame_type * GENDER

Perceived severity

.075 1 .075 .093 .761 .001 .093 .061


.776 1 .776 .584 .446 .003 .584 .118

Fear 2.937 1 2.937 1.692 .195 .010 1.692 .253

Threat_level * Frame_type * GENDER

Perceived severity

6.680 1 6.680 8.220 .005 .045 8.220 .814


.999 1 .999 .752 .387 .004 .752 .139

Fear .406 1 .406 .234 .629 .001 .234 .077


140.590 173 .813


229.906 173 1.329

Fear 300.256 173 1.736


5920.000 181


4301.333 181

Fear 3207.560 181

Corrected Total

Perceived severity

156.896 180


234.129 180

Fear 319.554 180

a. R Squared = .104 (Adjusted R Squared = .068) b. Computed using alpha = .05 c. R Squared = .018 (Adjusted R Squared = -.022) d. R Squared = .060 (Adjusted R Squared = .022)

246

Plots were produced to check the interaction effects among the

three factors. The following figures show the various interactions. Figure 6.3

show that women perceive the environmental threat to be more severe to their

health when compared to men. Fear arousal is also greater in women when

compared to men (Figure 6.4). It can be seen from Figure 6.5 that women

perceived high severity in both the low and high threat conditions when gain

framing is used. However men perceived higher levels of severity only under

high threat conditions when gain frames are used. Figure 6.6 shows that under

loss frame condition, higher threat levels evoke higher levels of perceived

severity only in the case of women. Men do not perceive greater threat

severity under loss conditions even when high threat levels are used.

Figure 6.3 Estimated marginal means for perceived severity

247

Figure 6.4 Estimated marginal means for fear

Figure 6.5 Estimated marginal means for perceived severity for gain frames

248

Figure 6.6 Estimated marginal means for perceived severity for loss frames

Table 6.74 shows the effect of the manipulations.

Table 6.74 Experiment 3: Summary of the effect of manipulations on PMT variables with the mobile stimulus



Fear

H5 Threat level X X NA

H6 Goal frame X X NA

H7 Threat level * Goal Frame X X NA

H8c Gender * Goal Frame X

Effect of Interactions on Purchase Intentions Purchase Intention

H8a Environmental concern * Goal Frame X

H8b Message Involvement * Goal Frame X

249


involvement, attitudes and intentions using the structural/ inner

model

As the measurement model was reliable and valid, the structural

model was tested with SmartPLS version 2.0.M3 (Ringle et al 2005). Initially

collinearity among the exogenous constructs was examined and there were no

multicollinearity issues. The main evaluation criterion for the structural model

is the value of the coefficient of determination (R2) as it represents the

explained variance of all the endogenous variables (Hair et al 2011). The level

and significance of the path coefficients (Hair et al 2011) are other important

criteria to judge the model.

The structural model was tested with 5000 sub-samples generated

using bootstrapping to evaluate the significance of the path co-efficients (181

observations per subsample, 5,000 subsamples and no sign changes). The

results of the structural model are shown in Figure 6.7. The significance of the

hypotheses were evaluated based on two-tailed tests (p < 0.05 (t=1.971), p <

0.01 (t= 2.598) and p < 0.001 (t= 3.334)).

The R2 values (shown in brackets) and path coefficients can be seen in Figure

6.7. R2 values greater than 0.33 are substantial and values between 0.19 and

0.33 are moderate (Chin 1998; Henseler et al 2009). Hair et al (2011) suggest

that 0.20 can be considered high for consumer behaviour studies.

Based on Chin’s criteria (Chin 1998) it can be observed that:

The coefficient of determination, R2 is 0.22 for the fear

endogenous latent variable. This means that the three latent

variables (perceived severity, perceived vulnerability, and

250

environmental knowledge) moderately explain 22.0% of the

variance in fear.

Figure 6.7 Hypothesis testing using PLS-SEM

The coefficient of determination, R2 is 0.36 for the message

involvement endogenous latent variable. This means that the six

latent variables. (perceived severity, perceived vulnerability,

response efficacy, self efficacy, fear and environmental

0.25**

-0.01

-0.11

-0.04

0.22**

0.16*

0.29***

-0.22***

0.71***

0.56***

0.07

-0.01

0.02

0.11

0.07

0.35***

0.07

0.04perceived severity

perceived vulnerability

response efficacy

(0.30)

self efficacy (0.07)

fear (0.22)

message involvement (0.36)

attitude towards ad (0.32)

attitude towards brand (0.50)

purchase intention (0.46)

0.18*

environmental concern

0.03

0.62***

environmental knowledge

0.24**

0.40***

251

concern) substantially explain 36.0% of the variance in message

involvement.

The coefficient of determination, R2 is 0.32 for the attitude

towards ad endogenous latent variable. This means that the four

latent variables (response efficacy, self efficacy, message

involvement and environmental concern moderately explain

32.0% of the variance in attitude towards ad.

The coefficient of determination, R2 is 0.50 for the attitude

towards brand endogenous latent variable. This means that the

the latent variable (attitude towards ad) substantially explains

50.0% of the variance in attitude towards brand.

The coefficient of determination, R2 is 0.46 for purchase

intention. This means that the latent variable - attitude towards

brand substantially explains 46.0% of the variance in purchase

intention.

Based on the inner model loadings and path co-efficients from

Figure 6.7, it can be summarised that:

The hypothesized path relationship between perceived severity,

perceived vulnerability and fear is statistically significant.

perceived severity has a comparatively stronger effect (0.29) on

fear.

perceived severity (0.40) and perceived vulnerability (0.24)

significantly predict response efficacy

The hypothesized relationship between perceived severity and

self efficacy was significant (0.25) whereas perceived

vulnerability was not significantly related to self efficacy (0.07)

252

The hypothesized path relationship between perceived severity

(0.04), perceived vulnerability (0.07), self efficacy (0.11) and

message involvement was not statistically significant. However

fear (0.35) and response efficacy (0.22) had a signficiant

relationship with message involvement.

response efficacy (0.07) and self efficacy (0.02) were not related

to attitude towards ad. While response efficacy (-0.01) was not

related to purchase intention, the hypothesized relationship

between self efficacy (0.18) and purchase intention was

significant.

The hypothesized path relationship between perceived severity

(-0.11), perceived vulnerability (-0.04), message involvement

(0.07) and environmental knowledge was not statistically

significant. However fear (-0.22) had a signficiant relationship

with environmental knowledge

environmental concern was not related to any of the

hypothesized relationships. The path coefficients were not

significant with the hypothesized variables message

involvement (0.07), attitude towards ad (0.03) and purchase

intention.

message involvement (0.56) is a significant predictor of attitude

towards ad

attitude towards ad is a strong predictor (0.71) of attitude

towards brand

Similarly attitude towards brand is a significant predictor (0.62)

of purchase intention.

253

Table 6.75 summarises the result of the hypotheses tests and their associated results.

Table 6.75 Experiment 3: Results of hypothesis testing using PLS-SEM

Hypothesis Path (standard)

t-value

Result

H9a Perceived severity fear 0.29 *** 4.03 Supported

H9a Perceived vulnerability fear 0.16* 2.02 Supported

H9b Perceived severity response efficacy

0.40 *** 5.92 Supported

H9b Perceived vulnerability response efficacy

0.24 ** 3.03 Supported

H9c Perceived severity self efficacy

0.25 * 2.50 Supported

H9c Perceived vulnerability self efficacy

0.07 0.58 Not supported

H10a Perceived severity message involvement


H10b Perceived vulnerability message involvement


H10c Fear message involvement 0.35*** 4.82 Supported

H10d Response efficacy message involvement

0.22** 2.63 Supported

H10e Self efficacy- message involvement

0.11 1.57 Not Supported

H11a Response efficacy attitude towards ad


H11b Self efficacy- attitude towards ad


H11c Response efficacy purchase intention

-0.01 0.22 Not Supported

254


Hypothesis Path (standard)

t-value

Result

H11d Self efficacy- purchase intention

0.18* 2.52 Supported

H12a Environmental knowledge perceived severity


H12b Environmental knowledge perceived vulnerability


H12c Environmental knowledge fear

-0.22*** 3.47 Supported

H12d Environmental knowledge message involvement


H13a Environmental concern message involvement


H13b Environmental concern attitude towards ad


H13c Environmental concern purchase intention


H14 Message involvement attitude towards ad

0.56 *** 9.50 Supported

H15 Attitude towards ad attitude towards brand

0.71 *** 17.7 Supported

H16 Attitude towards brand purchase intention

0.62 *** 10.3 Supported

Note: n=181; Estimates represent 5000 bootstrapping testing

*p<0.05 ; **p < 0:01; * **p <0:001

255

It is also necessary to assess the predictive relevance of the inner

model and therefore the model’s predictive relevance was analyzed using

Stone-Geisser test criterion Q2 (Chin. 2010; Hair et al 2011). This was

determined using the blindfolding procedure in SmartPLS. The omission

distance was chosen as 7, since values between 5 and 10 are advantageous

(Hair et al 2012). Cross-validated measure Q2 was checked and the results are

shown in Table 6.76. The Q2 values for all the endogenous constructs were

greater than zero as required. Table 6.76 also lists the R2 values of the

endogeneous contructs.

Table 6.76 Experiment 3: Model’s predictive relevance

Endogenous Construct R2 Q2

Attitude towards ad 0.32 0.23

Attitude towards brand 0.50 0.43

Purchase intention 0.46 0.40

Message involvement 0.36 0.19

Fear 0.22 0.15

Next, the effect size was calculated to measure the impact of a

predictor on a specific endogenous construct. Effect size represents Cohen’s d.

The values of 0.02 (small), 0.15 (medium) and 0.35 (large) indicate that the

construct has a small, medium or large effect size on the criterion (dependent)

construct respectively (Cohen 1988). The effects are shown in Table 6.77.

Although fear has a small effect on message involvement the effect size 0.14

is very close to the medium threshold. Most of the other effect sizes are small.

However message involvement has a medium effect on attitude towards ad

and attitude towards brand has a medium effect on purchase intention.

Attitude towards ad has a strong effect on attitude towards brand.

256

Table 6.77 Experiment 3: Effect sizes

Endogeneous VariableExogenous

Variable Effect Size f2

Effect size Interpretation

Purchase intention Attitude Towards brand 0.32 Medium

Purchase intention Self efficacy 0.05 Small

Attitude towards brand Attitude towards ad 0.43 Large

Attitude towards ad Message involvement 0.26 Medium

Attitude towards ad Response efficacy 0.01 Small

Attitude towards ad Environmental concern 0.06 Small

Message involvement Fear 0.14 Small

Message involvement Perceived vulnerability 0.01 Small

Message involvement Response efficacy 0.04 Small

Message involvement Self efficacy 0.01 Small

Message involvement Environmental concern 0.01 Small

Fear Perceived severity 0.09 Small

Fear Perceived vulnerability 0.03 Small

Fear Environmental knowledge 0.06 Small

Perceived severity Environmental knowledge 0.01 Small


This experiment examined the effects of varying threat levels and

goal frames on PMT variables and the subsequent effects of the PMT

variables on message involvement, attitudes and purchase intention. The issue

of e-waste (based on mobile phone stimuli) was used to observe if the

hypothesized effects were supported. The levels of perceived severity and

vulnerability remained high in this experiment too. It can be inferred from

Table 6.67 and 6.68 that perceived severity, vulnerability and fear remain

257

high for watch stimuli. Similar to Experiment 2 there was no main effect of

threat level or the goal frames. Similarly the hypothesized interaction effects

were only partially significant. Similar to Experiment 2 loss frames did not

increase threat perception as suggested by other researchers (Cox & Cox

2001; Meyers-Levy & Maheswaran 2004; van ‘t Riet et al 2008;O’Keefe &

Jensen 2009; Janssens et al 2010; Updegraff 2013).

There was no relationship between involvement and framing as

hypothesized. Both environmental concern and message involvement did not

interact with frames to produce an effect on purchase intention. This is in

contrast to the findings by other researchers who imply an effect between

involvement and framing (Maheswaran & Meyers-Levy 1990; Rothman et al

2006; Kim 2013). However gender played a significant role in predicting the

effect of the factors on the PMT variables. Women were more fearful and

perceived higher severity when facing an environmental threat when

compared to men. This confirms earlier findings by Garbarino & Strahilevitz

(2004) who find that women are more risk averse when compared to men.

The result also highlights the gender gap known to exist in environmental

threat perceptions (Flynn et al 1994; Bord & Connor 1997; McCright &

Dunlap 2011; Franzen & Vogl 2013). The interaction between frame and

gender also showed that women generate more negative thoughts when

presented with a negative goal frame when compared to men (Putrevu 2010).

Similar to the previous experiment, perceived severity and

vulnerability significantly influenced fear arousal as proposed by PMT

(Rogers & Prentice-Dunn 1997 ; Floyd et al 2000) and other studies that

apply this theory (Milne et al 2000; de Hoog et al 2008). Coping appraisal

was also initiated (Maddux & Rogers 1983; Boer & Seydel 1996; Milne et al

2000). Response efficacy was significantly influenced by perceived severity

and vulnerability. However self efficacy was only moderately influenced by

258

perceived severity and not by perceived vulnerability. In this experiment too,

all the PMT variables except self-efficacy predicted message involvement.

This finding confirms that higher levels of health risk increase involvement

with the message (Bloch & Richins 1983; Richins & Bloch 1986; Keller &

Block 1996; de Hoog 2005). Fear can therefore be used to increase message

involvement in enviromental communication (Meijnders et al 2001).

Apart from this, the finding also shows that while using goal frames

and threat levels, emphasis on the response efficacy would significantly

increase consumer involvement. The results are also in contrast to the finding

by Punam & Keller (1995) who find that low efficacy promotes more effortful

processing. Recent research emphasizes the importance of response efficacy

and treats it as a key component to message acceptance (Lewis et al 2010).

However, contrary to previous research, the efficacy variables were not

related to attitudes. Self efficacy significantly predicted purchase intentions

confirming the findings of earlier research (Maibach & Murphy 1995;

Luszczynska 2004; Gaston & Prapavessis 2012; Kreausukon et al 2012).

While objective environmental knowledge did not decrease the levels of

perceived severity or vulnerability, it had a negative effect on fear (Averbeck

et al 2011). Knowledge did not affect the levels of message involvement

similar to Experiment 2. Interestingly, environmental concern did not have an

effect on any of the hypothesized variables. This is in contrast to the findings

of Experiment 2 and contrary to the ELM (Petty & Cacioppo 1986). However ,

this finding was related to Experiment1 where environmental concern did not

predict the hypothesized variables. From Table 6.77, it can also be inferred

that message involvement has a stronger effect on attitude towards the ad.

Hence, this experiment based on stimuli related to e-waste showed

similar results as Experiment 2. The major difference was the role of

environmental concern. While Experiment 2 showed an effect of

259

environmental concern, this experiment showed that this variable was not

effective in influencing attitudes or intentions. Apart from the findings

related to environmental concern, there were no major differences between

the issue of e-waste and plastic waste.There were only two other differences.

In case of Experiment 2, perceived severity affected message involvement

and vulnerability affected self-efficacy. Hence, both the experiments highlight

the role of message involvement in promoting attitudes and intentions towards

green advertising.

6.5.10 Gender and Environmental Concern

Some green marketing studies claim that gender has a significant

effect on environmental concern (Shrum et al 1995; Jain & Kaur 2006;

Mostafa 2007). This was not investigated as part of hypothesis testing as it

was not part of the research objective, Surprisingly, a post hoc analysis

revealed that gender did not have any significant effect on environmental

concern in all the three experiments. This confirms the recent finding by other

researchers who do not find a link between environmental concern and gender

(e.g. D’Souza et al (2007)).

11_chapter6

Documents

results of pretest

selected products

popular products

associated results

mobile equalvariances

product preferences

product categories

product category