FOLGER'S Folger's - Stanford Universityzp492gk4305/zp492gk4305.pdf · The Folger'sproject is part of a largerscale research ... In order to evaluate thenature and scope of the ...

"

"

"

INNOVATION AND BRAND CHOICE: THE FOLGER'S INVASION

by

Ronald E. Frank and William F. Massy*

Introduction

The Folger's project is part of a larger scale research program concerned

with the analysis of the determinants of brand choice for frequently purchased

food and household products. The study is sponsored by the Graduate School of

Business, Stanford University, with funds from a recent Ford Foundation grant

for research dealing with the quantitative and environmental analysis of busi-

ness problems.

In order to evaluate the nature and scope of the Folger's project it is

useful to view it in the context of the other two brand chdice researches in

which Professor William F. Massy and I are engaged.

The objective of the first of these investigations is to develop esti-

mates of the effects of a firm's price and dealing policies in selected market

segments for a frequently purchased, low-priced food product. Because of the

form of the data base for this investigation (which is Market Research Corpora

tion of America consumer panel data for a major metropolitan area) we are able

to group individual purchases by market segment; the term segment refers to

typologies based on package size, chain store identity, and degree of customer

loyalty.

The strategy of market segmentation pursued by many firms is based on

what is often the implicit assumption that different customers or customers

faced with different retail environments are apt to have different degrees of(Presented at the American Marketing Association Winter Conference, atBoston, Mass., on December 27, 1963.)

*Both Assistant Professors of Business Administration, Graduate School ofBusiness, Stanford University

%

2

"

"

"

response to changes in the various components of a firm's marketing program.

This project is aimed at estimating the nature and degree of these differencesas well as drawing up their implications for the formulation of marketing

policies.

The second investigation is concerned with determining the relationship

between the personality and demographic characteristics of individual house-

holds and their brand choices with respect to beer, tea and coffee. The

data base for this investigation consists of approximately a quarter of a

million purchase decisions generated by about 3700 members of the J. Walter

Thompson panel, together with the results of a personality test (the Edwards

Personal Preference Schedule) for both the husband and wife (where present)

of each household. This data base, therefore, permits not only an analysis

of the relation between personality and brand choice but also of the rela-

tive importance of the personalities of the husband and wife in determiningpurchasing behavior. These data, which are now in a form ready for analysis,

were generously provided by the Advertising Research Foundation.

These two projects are concerned with analyzing customer response to

an existing set of brands. In contrast, the Folger's investigation is focused

on customer response to a change in the set of existing brands.

The objectives of the Folger's investigation ares

1. To determine the nature and extent of differences between families

who adopted a newly introduced brand of a frequently purchased product and

those who remained with established brands.

2. To determine the nature and extent of differences between familieswho adopted this newly introduced brand at varjLous times during the first 18

months after its introduction.

3

"

"

"

3= To determine the nature and extent of the relationship between a

household's initial purchase history for a new brand and the likelihood that

the household will stay with the brand in the future.

The results reported in this paper are concerned with the first of these

three objectives* Future publications will deal with the other two.

The ultimate goal of research of this type is to provide an improved

basis for identifying a probable purchaser of a new brand of a frequently

purchased product, thereby contributing toward the formulation of a firm's

promotional strategy with respect to its introduction into an established

market.

The introduction of Folger's coffee to the Chicago market in March of

1959 provides the setting for our analysis. We have chosen to study this

particular brand (Folger's) and product category (regular coffee) for five

reasons

1. Folger's introduction represents one of the most intensive intro-

ductory campaigns that has ever been launched in a single market. More than

1,500,000 coupons, each good for a free pound of coffee, were distributed to

consumers in the Chicago market. In contrast to many changes in markets like

this, the Folger's introduction was abrupt and large in magnitude. A change

of this character may tend to bring out variations in consumer response that

are relatively subtle and would not be observable under usual market

conditions.

2. Folger's obtained virtually 1007» distribution as it entered the

market. Therefore, variation in the extent of distribution over time is not

apt to cause differences in the time at which different families first tried

the product. This is another way of saying that variations in physical

4

"

"

"

distribution are not likely to be confounded with the answers to our

questions.

3. 18 months after its introduction, Folger's remained one of the top

five brands in the market; during the introductory period it became the lead-

ing brand (with 21% of the market). Therefore, the number of families who

participated in the introduction, as well as those who remained with the

brand afterward, is large enough to permit a detailed analysis based on

panel data..

4. The regular coffee purchase records of 538 families over the period

of 1958-1960, which embraces the time of Folger's introduction, are available

for analysis from the Chicago Tribune's Consumer Panel- .2/5. One of us— has intensively worked with the data from the Tribune's

Consumer Panel from 1956 through 1958 for regular and instant coffee in an

attempt to determine whether relatively simple probability models could be

used as a basis for learning something about the nature of the mechanisms

that create brand loyalty. The other has extensively studied the response of

consumers to the introduction of a new durable good--monochrome television3/receivers — . While the results obtained in this investigation are not directly

- We wish to thank both Pierre Martineau and Patrick Luby of the ChicagoTribune for their cooperation with respect to making the data availableat a nominal cost.

2/- Ronald E. Frank, "The Prediction of Brand Choice Using Simple ProbabilityModels," (unpublished Ph.D. dissertation, Graduate School of Business,University of Chicago, 1960); and "Brand Choice as a Probability Process,"Journal of Business, Vol. XXXV, #1 (1962), pp. 43-56.

3/- William F. Massy, "Innovation and Market Penetration," (unpublished Ph.D.dissertation in Economics, Massachusetts Institute of Technology, 1960).

5

"

"

"

comparable to the questions envisioned in the present study, the same kind of

statistical problems are involved.

The Data

The Tribune- panel consists of about 700 families who keep a chronological

record of their purchases of food and household items. For each purchase in a

given product class, information is available as to family code number, brand

purchased, date, quantity, price, type of outlet, and whether or not a deal

was used inmaking the purchase. In addition, data about a number of demographic

and economic characteristics of each of the households is available.

The 536 families included in our analysis turned in weekly diaries for

at least 153 of 156 weeks during the 3-year period, from 1958 through 1960.

The raw material for the analysis consists of approximately 35,000 pur-

chase decisions, each recorded on an IBM card. These data were summarized

into the variety of purchasing measures discussed below, and became the inputs

for discriminatory and other types of statistical analysis.

The Analysis

The results reported in the following paragraphs represent our first try

at determining the degree to which selected household socio-economic and pur-

chasing characteristics provide a basis for identifying which households will

become loyal to the new brand (Folger's).

After analyzing the weekly time series for the market shares and relative

prices for the 15 major brands in the market, the 1958-60 time period was

divided into three parts. These were: (1) a 62-week period "before" the

introduction; (2) a 32-week period "during" the introduction; and (3) a

62-week period "after" the introduction. The proportion of pounds devoted to

Folger's and each of the 14 other brands during the period after the introduction

6

"

"

"

was computed, far. each household. Based on these results, each household was

placed into one of the following four categories:

Primary Folger's Loyal: Those households that purchased a higher pro

portion of pounds of Folger's coffee than for any other brand.

Secondary Folger's Loyal; Those households for which Folger's was

brand.the second most frequently purchased

3. Other Folger's Households; Those households not falling in (1) or

at least once.(2) for which Folger's was purchased

4. Non-Folger's Households: Those households who purchased regular

coffee but never purchased Folger's.

The objective is to develop a model capable of predicting in which of the

above four categories a household will fall by observing its socio-economic

and purchasing characteristics during the 62-week period "before" Folger's

introduction. Our initial hypotheses led us to a list of 13 socio-economic

and 7 purchasing characteristics which might be related to the degree to which

a household would adopt Folger's. These are listed in Exhibit 1. Of these

20 variables,all but 3, have reasonably obvious definitions. Variable 10,

Status of Wife, indicates whether or not the household has a living husband.

Variable 14 is the proportion of purchases devoted to the brand purchased

most frequently by the household. Households who tend to devote a high

proportion of their purchases to their favorite brand might be less likely

to shift to a new alternative than those with a lesser degree of loyalty.

Variable 15 is a measure of the degree to which the proportion of pounds de-

voted to the household's favorite brand remained stable during the time period4/under investigation. —

4/For a discussion of the stability measure used in this investigation seeRonald E. Frank, "Brand Choice as a Probability Process," Journal of Busi-ness, Vol. XXXV, No. 1 (January, 1962) pp. 53-56.

7

"

"

"

Exhibit 1

Household Socio-Economic and Purchasing CharacteristicsInitially Selected for Investigation

A Socio-economic characteristics

1. Metropolitan zone

2. Number of adults3. Number of children4. Age of female head

5. Wife's employment status

6. Income

7. Occupation

8. Education9. Race

10. Status of wife11. Size of building

12. Tenure

13. Number of television sets

Purchasing characteristicsB

the household's most frequently14. Share of purchases devotedpurchased brand.

to

15. Stability over time of the proportion of purchases devoted tothe most frequently purchased brand

16. Number of stores in which the household shopped for regularcoffee.

17. Number of brands of regular coffee that were purchased.

18. Average number of pounds purchased per regular coffee shoppingtrip.

19. The total number of pounds of regular coffee purchased

20. The total number of proceeds purchased on some form of a deal(i.e., a lc off coupon or label on a can).

8

"

"

"

Two way multiple discriminant analysis is the principal technique used

to generate the results to be reported. It is similar to the more familiar

tool of multiple regression analysis. Regression analysis takes a continuous

dependent variable, such as durable good expenditures, and estimates the de-

gree to which each of a set of independent variables like income and age are

related to it. In contrast, two way multiple discriminant analysis is a tool

for estimating the relationship between a set of independent variables and a

dichotomous attributes such as whether families are Folger's loyal or non-

Folger's customers. The equation can be viewed in the following way:

efficients. The latter reflect the associated variable's importance for dis-

criminating between the members of the two customer groups.

Thus far the parameters for about 40 two way multiple discriminant

analyses have been estimated. They are based on the following three dichotomies

1. Primary and Secondary Folger's loyal versus non-Folger's households.

2. Primary loyal versus non-Folger's households.

3. Primary versus Secondary loyal households.

The most promising results obtained thus far have been generated from

the first of these splits; they are reported in the following paragraphs.

Comparison of Group Means

Table 1 presents the mean values for each of 16 household characteristics

for each of the following two groups of households: (1) Those who were primary

or secondary loyal to Folger's and (2) those who never purchased Folger's

during the period after the brand's introduction. The list of characteristics

V = ax, + bx„ + . . . . kx,12 k

here V is a dummy variable reflecting group membership, X, . . , X

re the independent variables, and a, b, . . . k are the discriminant co-

9

"

"

"

does not completely match those listed in Exhibit 1 as some were eliminated

on the basis of earlier findings. Six of the 7 statistically significant

differences between the two groups pertain to purchasing characteristics as

opposed to socio-economic characteristics.

Based on this table, it would appear that purchasing behavior is more

closely associated with household loyalty than are such socio-economic

variables as income and occupation, which are often used as indicators of

a household's social class.

These results are univariate in nature. When comparing income levels,

for example, there is nothing in the underlying methodology that takes into

account the interaction between income and the other variables. These might

produce a spurious effect or confound the true effects if any are present.

Discriminant Function Results

In order to simultaneously evaluate the effect of the factors listed in

Table 1 a two way multiple discriminant analysis was corlducted. The results

of the analysis contribute to two goals:

1. Predicting the innovative behavior of households.

2. Gaining insight into the factors that determine innovative

behavior.

Household Loyalty Predictions

Imagine that we wanted to make as many correct predictions about house-

hold" loyalty as possible, knowing only that the proportion of households

that were loyal to Folger's was 24.8 per cent. We would predict that any

household drawn at random would not be loyal. This way only 24.8 per cent

or 99 of the households would be misclassif ied.

Do the results of the discriminant analysis do any better? Using the

10

"

"

"

Table 1

Average Values by Characteristicfor Primary and Secondary Folger's Loyal Households (PSFL)

and for Non-Folger's Households (NF) £'

PSFL NLDescription DifferenceNumber of Households99 299

Socio-economic characteristics

Purchasing characteristics

a/- The metric for an individual characteristic is often expressed in terms ofset categories which are dissimilar to the units in which the raw variablemight be usually expressed. The relative magnitudes of the averages for agiven attribute are of inportance, whereas the absolute values are of littledirect significance.b/The values of these variables have been transformed into logarithms in orderto adjust for what appears to be nonlinearities in the raw data.c/Significant at the 5% level (based on a two tailed test).d/Significant at the 1% level (based on a two tailed test).

Metropolitan zone

Northwest city .192 .114 c/.078-Northwest suburb .091 .054 .037

Number of adults 2.091 2.097 -.006Number of children 1.869 1.572 297Wife's employment status

b/1nc0me-

.2731.911

.4151.878

- . 142^.033

Occupation 1.687 1.733 -.046Status of wife 1.091 1.144 -.053Size of building— 2.101 2.545 -.444

Primary brand share .630 .691 -.061^Stability of primary share -1.629 -.962 -.667^Number of stores- -3.848 3.438 .41^Number of brands 5.141 4.107 1.034^Average No. of pounds per regularcoffee shopping trip£'

1-1 /Total No. of pounds-

-1.48544.747

1.33137.579

.154^7 . 169-^

Total No. of pounds purchasedon a deal 5.354 3.662 1.691

11

"

"

"

proper break-point value for the discriminant function, 100 or 25.2 per cent

of the households would be misclassified. The discriminant value for a given

household is computed hy summing the product of the discriminant coefficient

for each variable times the value of the variable for the households, across

household whose discriminant valueall of the variables in the equation. A

is at the break- point would have an equal probability of being either

i i -5/loyal or not.—

— The break- point value (z*) was computed from the following formula:

where :

z- is the mean discriminant value of the loyal group

value of the non- loyal group

1

is the mean discriminantA2 is the pooled variances of the discriminant scores of each of

the two groupscT-p

P is the proportion of observations in the loyal group

P is the proportion of observations in the non-loyal group.

Computing the break-even score in this fashion helps to take into ac

count the -fact that the two groups are of unequal size. For a discussion

of the theory underlying this procedure see:

William W. Cooley and Paul R. Lohnes : Multivariate Procedures for the

Behavioral Sciences (New York: John Wiley &

Sons,

Inc. 1962) p. 118.

z* ■2<J2 - zL ) l4~<+ 2 < l°*e T^ 1

12

"

"

"

Taken by itself this piece of evidence would seem to imply that the dis-

criminant function is of little value to one faced with the problem of pre-

dicting innovative potential.

Fortunately., -however, our ability to predict is not in as poor a state

as is implied by. this result.. Though the misclassif ication rate test of

predictive efficacy is often used to evaluate the results of discriminant anal

ysis it .fails to take into account all of the information that is available

to the user of such a model. It uses only information as to which side of the

break-pointvalue a given household's score falls. This is because the under-

lying question is couched in terms of making dichototnous classifications

for particular households. This is an extremely stringent demand. Even if

we cannot make predictions of this sort successfully it may still be possible

to improve our ability to predict by using information as to the conditional

probabulity of being Folger's loyal given a household's discriminant score;.

Figure 1 presents the relationship between the value of the discriminant

function and the probability of a household being loyal to Folgers.-''' This

probability gradient rises from a Folger's loyal probability of about 8% to

10% at low discriminant values to over 50% for the highest values. Incorpora-

ting the discriminant function information increases the probability of Fol-

ger's loyalty by a factor of 5.

6/- For an example of this same distinction in the context of predicting brandchoice see :

Alfred A. Kuehn, "Demonstration of a Relationship Between PsychologicalFactors and Brand Choice," Journal of Business, Vol. XXXVI, No. 2 (April1963), pp. 237-241. ~~

13

"

"

"

Probability

-2Discriminant Score (XlO )

Figure 1. The relationship between a household's discriminantscore and the probability of its being Folger'sloyal.

14

"

"

"

Determinants of Household Innovative Behavior

Table 2 presents the standardized coefficients and the F-ratio for the

function-.. The standardized coefficients are calculated as the product of

the raw coefficients times the standard deviation of the associated variable.

Standardization facilitates the comparison of the magnitude of the effects

by stating them in terms of equally likely changes. The F-ratio is 2.101

which implies that it is just significant at the 1% level.—

The relative magnitude of the standardized coefficients (ignoring signs)

are quite informative. Of the four factors which exert the greatest effect

three have to do with rates of purchasing activity. Of these, total pounds

is by far the most important factor. Households with high purhcase rates

for the product during the "before" period are much more apt to be either

primary and secondary loyal after the introduction. The same result holds

for households who tend to purchase larger amounts of coffee per shopping

trip and shop in a larger number of stores.

Possibly households that purchase coffee relatively often are better in-

formed of new developments in the marketplace (if for no other reason than

they are exposed to it more often) and/or are more willing to experiment with

new alternatives because of the importance of the product in the family diet.

Families who tend to purchase large quantities of coffee per trip may tend to

be those who are more economy minded. Particularly during the introductory

period Folger's used its relative price as a basis for attracting customers

- See T. W. Anderson, An Introduction to Miltivariate Statistical Analysis.(New York: John Wiley & Sons, Inc., 1958), pp. 108-'lo9.

15

"

"

"

Table 2

Standardized Discriminant Coefficients and F-Ratio for PSFL versus NF

Coefficient RankDescriptionXlO (signs ignored)

Socio-economic characteristicsMetropolitan zone

824 6Northwest city

Northwest suburbNumber of adultsNumber of children

269 12524 7

8400

Wife's employment status

Income

298 10

1.694 3

Occupation 081 15

Status of wifeSize of building

218 13

-1.269 5

Purchasing characteristicsPrimary brand shareStability of primary shareNumber of StoresNumber of brands

.004 16

-.2862.254

112

9.396Average number of pounds perregular coffee shopping trip 1.380

22.2794

Total number of pounds 1

Total number of poundspurchased on a deal .154

(2.101)-714

F-Ratio

a/— Significant at just about the 1% level (one tailed test)

16

"

"

"

and therefore may have disproportionately attracted this type of shopper. The

number of stores shopped may also tend to be correlated with economy mindedness

and hence lead, to Folger's loyalty. Setting economy aside, it is possible that

customers who are willing to shop in more retail environments indicate a greater

willingness to graple with a wide range of choices, and may be better able to

integrate new alternatives into their shopping habits.

The findings in Table 2 also suggest that the socio-economic characteris-

tics of households do not play as important a role in influencing innovativebehavior in this context as do purchasing characteristics.

Previous research on the introduction of new farm practices and new

products has tended to place heavy emphasis on these social indicators. It

may be that for changes of this sort a household's reference group (defined

by such proxy variables as income and occupation) are of relatively greater

importance than in the case of new brands for at least two reasons:

1. Changes of the former type are apt to have repercussions over a

broader range of a persons ' activities than are the latter.

2. Changes of the former type are apt to be associated with a greater

degree of ambiguity as to appropriate behavior than are the latter.

These, then, are a few of the highlights of the initial phase of the

investigation.

Plans For Additional Research

The next stage of the analysis will cover the latter two objectives,

namely: (1) building a model capable of making useful predictions as to when

various types of households will make their first Folger's purchase and (2)

building a model aimed at predicting subsequent household loyalty to Folger's

17

"

"

"

which incorporates the frequency and timing of the household's initial

purchase, history for the brand.

Ln. addition, work on this part of the problem will be extended. Thus

far innovative behavior in regular coffee has been placed on the context of

a household's socio-economic characteristics as well as its regular coffee

purchasing habits. Innovative behavior in a given product class is probably

not just related to the past purchasing behavior of that particular product,

but also to the household's purchasing behavior for grocery products in

general- For example, it may be that households with different rates of

per capita consumption, different degrees of budget allocation to work-

saving grocery products, and/or with different store shopping habits will

have different responses to the availability of a new brand in a particular

product class.

In order to investigate this aspect of the problem we have made use of

one-third of a million purchase decisions from the Chicago Tribune panel

covering household purchasing behavior for a 4-month period toward about 88

different products for each household included in the previous analysis. We

have abstracted from this data base about 60 different measures of over-all

purchasing activity which we suspect might play a role in determining the

innovative behavior of a household in a particular product category. Work

has just begun on ths analysis of this set of data.