MA1-2

Proceedings of the Asia Pacific Industrial Engineering & Management Systems Conference 2014

1

A model for improving the customers’ purchase willingness

considering their latent intentions and media contacts

Keisuke Korenaga

Graduate school of Science and Engineering

Aoyama Gakuin University, Japan

Tel: (+81) 90-1770-0981, Email: [email protected]

Satoshi Kumagai

Industrial and Systems Engineering

Aoyama Gakuin University, Japan

Tel: (+81) 42-759-6312, Email: [email protected]

Hiroki Nakano

SmartPlatform, NIFTY Corporation, Japan

Tel (+81) 3-5471-4923, Email: [email protected]

Abstract: Due to advertising budget limitations and the diverse nature of customer purchase values,

advertisements need to focus on the most effective customers, that is, the customers most likely to make purchases.

The effectiveness of advertising varies depending on customer purchase values. The purpose of this study is to

construct a model that quantifies the extent to which customer values and frequency of media contact influence

customer purchase willingness. We used single source data that connected customer attribute data, media contact

data, and buying behavior data for particular customer IDs. Using the example of two beverage products, we

constructed a model through factor analysis and structural equation modeling. The analysis consisted of three

steps. First, we extracted seven kinds of latent intentions from the available customer value data. We named these

as High-quality, Consideration, Lifestyle, Design and Trend, Economical, Brand, and Ecological intentions.

Second, we divided consumers into three segments according to the frequency of media contact. Finally, we

analyzed those segments simultaneously, using multiple analysis in structural equation modeling. Thus, we

quantified the effect of each latent intention on purchase willingness, also keeping frequency of media contact in

view. The results indicate that the Design and Trend intention of the customer group of the second segment of the

three.

Keywords: Single source data, Structural Equation Modeling, advertising effectiveness, purchase willing

ness

1. Introduction

Promotion activity is aimed at increasing product

recognition and sales. Because advertising reaches many

unspecified customers, it is one of the primary means of

business promotion. However, advertising is very expensive.

Therefore, it is essential to measure the effectiveness of the

substantial advertisements costs incurred in a promotion

strategy. On the other hand, the values according to which

customers make purchases are very diverse. Therefore, it is of

critical importance to construct a model that measures

advertising effectiveness while keeping in view the values of a

variety of consumers.

Figure 1 shows the movements in total advertisement

costs in Japan from 2006 through 2011. These costs fell sharply

after the bankruptcy of Lehman Brothers in 2008. Figure 2

shows the component ratios of advertising costs in that period.

Television advertising accounted for about 30% of all

advertising costs. Television was therefore the primary

advertising medium before Internet advertising became

widespread. However, according to Figure 3, costs for

television advertising also fell in 2008 as a result of the

Lehman Brothers bankruptcy,

Companies have to utilize limited costs effectively, and it

is therefore necessary for them to increase product recognition

and sales. Understanding the effect of advertising and the

7

Korenaga, Kumagai and Nakano

________________________________________

manner in which it influences customer purchase decisions

helps in building a strategy for placement of advertisements.

Figure 1: The transition of total advertisement costs.

Figure 2: The component ratio of advertisement costs.

Figure 3: The transition of advertisement costs of television

media.

2. Purpose of study

This study closely examines the factors that influence

customer purchase willingness. Purchase willingness is the

level of a customer’s desire to buy a specific product. Figure 4

shows the AIDMA model, a typical advertising

communication model. The AIDMA model illustrates the

entire process, from a customer’s awareness of an

advertisement to final buying behavior. Purchase willingness

corresponds to Desire in the AIDMA model. Purchase

willingness is the most important part of the purchasing

process, so company promotion activities aim to generate

purchase willingness. In terms of the AIMDA model, our study

focuses on how to move the customer from the Attention stage

to the Desire stage.

This study assumed that a primary factor to improve

customers’ purchase willingness is customer values and media

contacts. This study used single source data (from the Nomura

Research Institute, Ltd. INSIGHT SIGNAL) that connected

the attribute data of customers, media contact data, and buying

behavior data by specific customer ID. Based on this data, and

focusing on television, we constructed a model to quantify the

effects of advertising in the context of diverse customer

purchase values and frequency of media contact.

Figure 4: AIDMA model.

3. Approach

3.1. Summary of Single Source Data

Table 1 shows a summary of the single source data used

in this study. Breaking down this data, we investigate purchase

willingness and buying behavior of one product and several

individual customers for March and April 2012. The data set

enables us to analyze how media contact brought about

changes in buying behavior during this period.

Table 1: Summary of the Single Source Data

Customer’s attribute, customer’s value, and buying

behavior.

8


________________________________________

Frequency of television viewing and listening

(From March 1, 2012 to April 30, 2012)

Frequency of magazine reading

(From February 25, 2012 to April 27, 2012)

Frequency of internet access

(From March 1, 2012 to April 30, 2012)

Frequency of ad placements on television

Frequency of ad placements in magazines

3.2. Approach

We constructed our model through the following process:

Step 1: Choice of products for analysis

Step 2: Extraction of details on customers whose purchase

willingness improved from March to April 2012

Step 3: Extraction of latent intentions by factor analysis of

customer values data

Step 4: Quantification of the effect of latent intentions on

purchase willingness (through structural equation modeling)

Step 5: Classification of customers according to the frequency

of contact with televisions advertisements for the target

products

Step 6: Quantification of the factors that increase purchase

willingness, using multiple analysis in the structural equation

modeling

4. Construction of the model considering latent intentions

This study uses two beverage products, Orangina and

Ihoras, as examples. We focused on increases in purchase

willingness, and sought to quantify the effects of customer

values and media contact on purchase willingness. We

therefore started by extracting the names of customers whose

purchase willingness increased from March to April.

4.1. Extraction of customers whose purchase

willingness improved

The questionnaire item on purchase willingness included

the following options: ‘Not willing to buy’ ‘Unknown’,

‘Willing to buy’, and ‘Highly willing to buy’. The consumers

whose purchase willingness changed were those who changed

their responses to this question between March and April. The

attitude change was from ‘Not willing to buy’ or ‘Unknown’ to

‘Willing to buy’ or ‘Highly willing to buy’.

4.2. Extraction of latent intentions through factor

analysis

One purposes of this study is to quantify the influence of

customer values on purchase willingness. The data set obtained

had 32 questionnaire items on customer values. Table 2 shows

some of these items. The number of variables is very high,

given that we had to take 32 value items into consideration. We

have assumed that certain latent factors influence customer

values, and have therefore extracted these latent factors

through factor analysis.

Table 2: Questionnaire items on customer values

Variables Questionnaire items

S5 Have a brand of regular purchase

S6 Color and design is first priority rather than

easiness of use

S7 Color and design is first priority even for TVs

and PCs

S11 Trend is important for selecting products

S14 Uniqueness is important for selecting products

We analyzed each customer’s purchase values through

factor analysis. Based on individual values, a variety of factors

or combinations of factors can influence customer purchase

willingness. We used maximum-likelihood estimation and

normalized varimax as analytical methods for factor analysis.

We extracted seven latent factors pertaining to both beverages.

These latent factors were named to be based on the variables

that factor loading is high. Table 3 shows these factors, which

we labeled as “latent intention” factors.

Table 3: The latent intention factors

Orangina Irohas

High-quality intention High-quality intention

Consideration intention Consideration intention

Lifestyle intention Lifestyle intention

Ecological intention Ecological intention

Design and Trend intention Trend intention

Economical intention Economical intention

Brand intention Brand and Design intention

4.3. Construction of the model considering latent

intentions

We constructed a model to quantify which of the latent

intentions influences customer purchase willingness and how

it does so. The model has been constructed for both drinks.

While it deploys factor analysis, this model is constructed so

as to also take account of factor correlation, which is not

possible in the case of factor analysis alone.

Table 4 shows the degree of goodness of the fit for each

model. GFI and AGFI are examined, and the fit is deemed to

be good if each indicator is at 0.9 or over. CFI how long shows

goodness of fit was improved as compared with an

independent model. We judge the model fit to be good if the

CFI is around 1. RMSEA shows estrangement with the

distribution of the model and the true distribution for quantity

9


________________________________________

per 1 degree of freedom. RMSEA values at or below 0.05

imply a good fit. Each indicator level obtained from the data

set was judged good in terms of fit.

Table 4: Goodness of fit.

Indicator Orangina Irohas

GFI 0.927 0.851

AGFI 0.904 0.817

CFI 0.833 0.811

RMSEA 0.047 0.049

Table 4 and Table 5 show the path coefficients to quantify

which latent intentions influence customer purchase

willingness in each model.

First, we examined the Orangina path coefficient. Here,

the Design and Trend intention and the Economical intention

were shown to be highly effective factors. Customers’ desire to

be seen as trendy or buy well-designed products and their

preference for cheap products significantly improve purchase

willingness. On the other hand, the Brand intention appeared

to have a negative effect; the tendency to prefer big-name

brand products appears to decrease purchase willingness in the

case of the two drinks.

Next, we examined the Irohas path coefficient. Here, the

Brand and Design intentions showed negative effects. Thus, in

this case, the customer preference in favor of both big-name

brands and well-designed products greatly decreases purchase

willingness.

Table 5: Orangina Path coefficient

Latent intentions Path coefficient

High-quality intention 0.47

Consideration intention -0.12

Lifestyle intention -0.47

Design and Trend

intention 1.09

Economical intention 1.13

Brand intention -1.27

Ecological intention -0.57

Table 6: Irohas Path coefficient

Latent intentions Path coefficient

High-quality intention 0.14

Consideration intention -0.30

Lifestyle intention -0.55

Brand and Design

intention -3.07

Economical intention 0.22

Trend intention -0.45

Ecological intention 0.13

5. Construction of the model considering latent intentions

and frequency of media contact

In this section, we add frequency of media contact to the

factors included in the model so far. We focus only on

Orangina in this section. Because we don’t have enough data

of customers whom purchase willingness improved in Irohas,

we couldn’t construct the model of Irohas.

5.1. Segmenting customers according to level of media

contact

To construct our model, we segmented customers

according to level of media contact. Figure 5 shows the

relationship between frequency of contact with television

advertisements and purchase willingness. We categorized

customers into five grades, and we graphed the ratio of

improvement in purchase willingness based on frequency of

media contact. We named this indicator the Improvement rate

of purchase willingness. The results indicate that when contact

with the relevant television advertisements occurs 0-5 times

and 6-10 times, the Improvement rate of purchase willingness

grows. The rate is at the maximum after 11-15 incidents of

contact, but tends to stagnate above 16 incidents.

We divided the customers into three segments according

to the tendency of Improvement rate of purchase willingness

and frequency of contacts with advertisements on television.

Customers whose contact with television advertisements

occurred 0 to 10 times have been placed in segment 1. Their

Improvement rate of purchase willingness is high. Customers

whose contact occurred 11 to 15 times have been placed in

segment 2. Their Improvement rate of purchase willingness is

the highest. Customers whose contact occurred over 16 times

have been placed in segment 3. Their Improvement rate of

purchase willingness has leveled off.

Figure 5: Improvement rate of purchase willingness.

5.2. Construction of the model considering latent

intentions and frequency of media contact

10


________________________________________

5.2.1. Multiple analysis by SEM

Structural equation modeling involves a method called

multiple analysis. The purpose of multiple analysis is to

analyze multiple populations simultaneously. This enables

comparison between different models for calculation of the

goodness of fit and for selection of the optimal aggregate

model for multiple populations. Multiple analysis also enables

researchers to examine the homogeneity and heterogeneity of

the model in respect of multiple populations. The process of

construction is displayed below:

Model 1: Arrangement Constant model

This model has equal variables for each factor between

populations. Therefore, the path diagram is identical between

all populations.

Model 2: Weakly Measurement Constant model

This model involves equal factor loadings for each factor.

Model 3: Measurement Constant model

In addition to model 2, this model involves equal variance and

covariance for each factor.

Model 4: Strong Measurement Constant model

In addition to model 2, this model involves equal error variance

for each variable.

Model 5: Equal Population Parameter model

This model involves equal parameters.

These five models strengthen restrictions as we advance

from model 1 to model 5. We determine what model should be

adopted according to goodness of fit.

5.2.2. Construction of the model through multiple

analysis in the context of latent intentions and frequency of

media contact

We postulate that the effect of customers’ latent intentions

on purchase willingness varies according to the frequency of

media contact. We therefore used multiple analysis to quantify

the effect of each latent intention on customer purchase

willingness based on the categorization into three segments in

section 5.1.

First, we constructed the arrangement constant model,

where the shapes of the models are identical. The arrangement

constant model enables comparison of the path coefficients

between segments of customers according to identical model

shapes.

Table 7 shows the goodness of fit of the arrangement

constant model constructed according to each latent intention.

The GFI and AGFI are over 0.9 in all models; hence, we judged

the model fit is good. The RMSEA and CFI levels also

indicated the acceptable goodness of fit. Thus, we judged the

path diagram is identical according to all latent intentions

between three segments.

Table 7: Goodness of fit in the Arrangement Constant model

AGFI GFI RMSEA CFI

High-quality

intention 0.953 0.984 0.047 0.979

Design and

Trend

intention

0.965 0.985 0.000 1.000

Economical

intention 0.908 0.960 0.087 0.765

Brand

intention 0.914 0.963 0.087 0.890

Lifestyle

intention 0.965 0.991 0.036 0.986

Consideration

intention 0.941 0.986 0.079 0.926

Ecological

intention 0.997 1.000 0.000 1.000

Table 8 shows the path coefficient pertaining to the

influence of latent intentions on purchase willingness. We

observed a significant gap between the influence of the Design

and Trend intention on customers in different segments. There

does not appear to be a similar gap as regards other intentions

across segments.

Table 8: Path coefficient in the Arrangement Constant model.

Segment 1 Segment 2 Segment 3

High-quality

intention 0.16 0.14 0.15

Design and

Trend

intention

-0.55 1.41 0.36

Economical

intention 0.22 0.05 0.36

Brand

intention 0.16 0.09 0.28

Lifestyle

intention 0.15 0.29 0.20

Consideration

intention 0.21 0.21 0.09

Ecological

intention -0.33 -0.71 -0.14

Figure 6 shows the constructed arrangement constant

model. We observed that the shape of the path diagram is the

same in each segment. However, each path coefficient is

different.

11


________________________________________

Figure 6: Arrangement Constant model considering media

contacts and latent intention

Next, we constructed a weakly measurement constant

model to correspond to model 2 described in section 5.2.1. The

weakly measurement constant model enables us to argue

whether or not the path coefficient is same between segments

of a population.

Table 9 shows the goodness of fit of the weakly

measurement constant model. Ecological intention greatly

degraded the goodness of fit here as compared with the

arrangement constant model. Thus, we chose to adopt the

arrangement constant model. The other latent intentions

examined that goodness of fit is good. However, Design and

Trend intention has a path coefficient which latent intention

influence to purchase willingness has distinguished gap in the

Arrangement constant model.

Table 9: Goodness of fit in the Weakly Measurement Constant

model.

AGFI GFI RMSEA CFI

High-quality

intention 0.960 0.980 0.030 0.987

Design and

Trend

intention

0.949 0.970 0.041 0.938

Economical

intention 0.922 0.954 0.075 0.749

Brand

intention 0.916 0.951 0.085 0.856

Lifestyle

intention 0.968 0.984 0.033 0.979

Consideration

intention 0.962 0.983 0.050 0.944

Ecological

intention 0.693 0.898 0.328 0.000

Table 10 shows the path coefficient of the influence of

five latent intentions on purchase willingness in the weakly

measurement constant model. The path coefficients are

identical between each segment in the weakly measurement

constant model. Moreover, each path coefficient is small.

Table 10: Path coefficients in the Weakly Measurement

Constant model.


High-quality

intention 0.16 0.16 0.16

Economical

intention 0.17 0.17 0.17

Brand

intention 0.15 0.15 0.15

Lifestyle

intention 0.24 0.24 0.24

Consideration

intention 0.11 0.11 0.11

Table 11 integrates the path coefficients of each latent

intention between each segment. In the case of the High-

quality intention, Economical intention, Brand intention,

Lifestyle intention, and Consideration intention we adopted the

weakly measurement constant model, while for the Design and

Trend intention and the Ecological intention we adopted the

arrangement constant model.

Table 11: Path coefficient of the model considering latent

intentions and frequency of media contact


High-quality

intention 0.16 0.16 0.16

Consideration

intention 0.11 0.11 0.11

Lifestyle

intention 0.24 0.24 0.24

Economical

intention 0.17 0.17 0.17

Brand

intention 0.15 0.15 0.15

Design and

Trend

intention

-0.55 1.41 0.36

Ecological

intention -0.33 -0.71 -0.14

The results show that the influence of latent intentions on

12


________________________________________

purchase willingness is identical for the five latent intentions

where the weakly measurement constant model was adopted,

even when the frequency of media contact varied. On the other

hand, the two latent intentions for which the arrangement

constant model was adopted showed different path coefficients

in each segment. That is to say, the influence of latent

intentions on purchase willingness varies according to the

frequency of media contact. When we focused on the Design

and Trend intention, the path coefficient of segment 2 was

large. The Design and Trend intention of segment 2 therefore

has a substantial effect on purchase willingness, judging by the

path coefficient of 1.41. Our analysis suggests that the media

contact frequency that shows the best results in terms of

increase in purchase willingness is between 11 to 15 times.

6. Conclusion and challenges

This study uses structural equation modeling to quantify

how customers’ latent intentions and the frequency of media

contact influence purchase willingness. Analysis based on the

model constructed in this study reveals that the effect of latent

intentions on purchase willingness varies according to the

frequency of media contact.

As regards future challenges, since forms of media have

become increasingly diverse and complex in recent years, there

is need for a model that separates and quantifies the effects of

different media such as magazines, newspapers, the internet,

and social networking sites. Moreover, this study focused

only on increasing purchase willingness, that is, on one part of

the purchase process. We also need to construct a model that

enables quantification of the kind and frequency of media

contact that influence the entire purchase process and hastens

its movement towards the Attention to Purchase stage

illustrated in the AIDMA flowchart.

REFERENCES

[1] DENTSU INC. Advertising Expenditures in Japan.

http://www.dentsu.co.jp/knowledge/ad_cost/

[2] Shigeru Kido, Advertising Management (in Japanese),

Asakura Publishing Co, Ltd. ,Tokyo, Japan, 2004

[3] Akihiro Kamei and Satoshi Hikita, New Text on Advertising,

Nikkei Advertising Research Institute, Tokyo, Japan, 2005

[4] Hideki Toyoda, A primer of factor analysis: Newest data

analysis with R (in Japanese) , Tokyo Tosho Co, Ltd., Tokyo,

Japan, 2012

[5] Atsushi Oshio, The First Step to Structural Equation

Modeling Starting with drawing path diagram, ARTE Co Ltd.,

Tokyo, Japan, 2010

[6] Hideki Toyoda, Covariance Structural Analysis –

Structural Equation Modeling: Introduction (in Japanese),

Asakura Publishing Co, Ltd., Tokyo, Japan, 1998


Structural Equation Modeling: Question (in Japanese),

Asakura Publishing Co, Ltd., Tokyo, Japan, 2003


Structural Equation Modeling: Case study (in Japanese),

Kitaooji Shobo Co, Ltd., Kyoto, Japan, 1998

[9] Hiroshi Kumakura, Advertising Effects: Concepts and

Methods, Management systems: a journal of Japan Industrial

Management Association 18(1), 19-25, 2008-04-15

[10] Nomura Research Institute, Ltd. INSIGHT SIGNAL

https://www.is.nri.co.jp/

13

MA1-2

Documents