# An Examination of Consistency in Coupon Usage By Households ACRoss Product Classes

^{[ to cite ]:}

Kapil Bawa and Robert W. Shoemaker (1986) ,"An Examination of Consistency in Coupon Usage By Households ACRoss Product Classes", in NA - Advances in Consumer Research Volume 13, eds. Richard J. Lutz, Provo, UT : Association for Consumer Research, Pages: 277-281.

^{[ direct url ]:}

http://acrwebsite.org/volumes/6503/volumes/v13/NA-13

This study examines whether households behave consistently in their level of coupon usage in different product classes. It also examines whether the degree of store loyalty is consistently high or low for these households. Several interesting empirical results are found. First, households are considerably more consistent in their purchasing than would be expected by chance. Second, there exists a significant segment of households with above average coupon usage for most product classes and low store loyalty for most products. The managerial implications of these findings are discussed.

INTRODUCTION

In recent years, a number of studies have been conducted on the growing use of coupons, the profitability of coupon promotions and the level of coupon redemption rates. While these studies have addressed a number of important couponing issues, it appears that no studies have examined whether households are consistent in their level of coupon usage across different product classes. For example, if a household uses coupons to an above average extent in one product class, does it also use coupons to an above average extent for its purchases in other product classes? If it does, it can be considered a "consistently" heavy user of coupons.

When examining coupon usage across product classes, it appears desirable to consider the extent of store loyalty as well. A household that is consistently store loyal and a heavy coupon user is likely to be responsive to manufacturers' coupons for products that are not available in its favorite store, and much less responsive to store coupons offered by other stores. On the other hand, a household that follows a store-switching strategy for most product classes and is a heavy user of coupons is likely to respond favorably to both manufacturer and store coupons.

This study therefore examines consistency across product classes in terms of both coupon usage and store loyalty. The major questions addressed here are:

1. To what extent are households consistent in their use of coupons across product classes? (i.e. to what extent do they consistently use an above average percent of coupons in all classes or to what extent do they consistently use a below average percent of coupons across product classes?)

2. To what extent are households consistent in their store loyalty across product classes? (i.e. if they are store loyal for one class, are they also loyal for other classes?)

3. What is the relationship, if any, between a household's degree of consistency in coupon usage and its degree of consistency in store loyalty?

PRIOR RESEARCH ON CONSISTENCY IN PURCHASING BEHAVIOR ACROSS PRODUCT CLASSES

While no previous studies have examined the consistency of coupon usage across products, a number of authors have examined the general question of whether households utilize similar buying strategies across product classes.

Cunningham (1956) studied households' loyalty to their favorite brand (measured as the share of purchases devoted to the favorite brand) in seven product classes. He computed pairwise rank correlation coefficients for each of the 21 possible pairs and found only 2 of the pairs to be significant. He concluded that "....loyalty proneness (across product classes) does not exist to a significant degree." This conclusion was later echoed by Massy, Frank and Lodahl (1968) who conducted a similar analysis using purchase data for beer, coffee and tea.

A third analysis covering 38 product classes was conducted by Wind and Frank (1969). The two variables they observed were brand loyalty (measured as the market share of the favorite brand) and private brand proneness (market share of private label purchases). As in earlier studies, they correlated each measure between all possible pairs of product classes and concluded that "on the average, only an extremely small percentage of the variation from household to household in loyalty to brands in one product class is associated with variation in brand loyalty in another class."

A more recent study of buying consistency across product classes was conducted by Blattberg, Peacock and Sen (1976). In their analysis, heavy buyers were classified into one of sixteen "purchase strategy" categories such as "Loyal Buyers of National Brands" or "Deal Oriented Buyers of Private Label Brands." They then measured the extent to which buyers utilized the same buying strategy across 2 pairs of product classes.

They found that 35% of the households (n=108) used the same buying strategy for aluminum foil and waxed paper, while the corresponding figure for liquid detergent and facial tissue was 13% (n=235). They also examined the extent to which buyers used similar but not identical buying strategies for the two pairs of products. In general, the authors found greater consistency in buying strategies than the earlier studies. However, the level of consistent behavior across product classes was still not high.

The purpose of the current analysis is to extend prior research in several ways. First, the study focuses on a new variable, namely the consistency of a household's coupon usage across product classes. Second, this study simultaneously analyzes consistency across four different product classes. While prior studies considered data from several product classes, the analyses were based on comparing only two product classes at a time. Finally, statistical tests are used to test the hypothesis that usage of coupons or degree of store loyalty is decided independently for each of the four product classes. "Independence" in this study refers to statistical independence; that is, if the joint probability of observing similar behavior (such as above average coupon usage) in different product classes is statistically no different from the product of the unconditional probabilities of observing such behavior in each of these product classes, the behavior is considered independent across the classes.

HYPOTHESES

It is assumed that households gain utility when they use coupons in any product class, in the form of savings resulting from the lower prices paid. However, there are certain costs associated with coupon usage as well. These consist of (a) a fixed cost, which must be incurred if one wishes to use any coupons at all, and which includes the opportunity cost of scanning newspapers or direct mail, reading coupon inserts, and organizing the coupons for future use, and (b) variable costs for each coupon used, arising from the opportunity cost of cutting, sorting, and redeeming individual coupons.

Thus an individual household's coupon usage is assumed to be a function of the savings obtained relative to the costs associated with coupon usage. The notion of a "fixed cost" of coupon usage suggests, moreover, that if a household uses coupons at all, it is in the household's interest to use coupons in many product classes in order that the total benefits from coupon usage exceed the fixed cost. Thus if the household is able to redeem enough coupons such that the total savings exceed the total costs of coupon usage, it will tend to be a relatively "heavy" user of coupons for most product classes. However, if the total savings are always less than the total costs, it will tend to be a light or non-user of coupons for most product classes.

In either case, the level of coupon usage by individual households is likely to be highly correlated across product classes "that is, they will be relatively consistent in their use of coupons across classes, rather than deciding on their coupon usage independently for each product class. This reasoning leads to the following hypothesis on coupon usage:

H1: More households will consistently make an above (or below) average share of their purchases with a coupon in several product classes than would be expected if coupon usage was decided independently for each product class.

For store loyalty too, one would expect to find consistently high (or low) levels across product classes. Here it is assumed that if a household buys in several different stores for one product class, it can then buy in several stores for most product classes at little or no additional cost. This leads to the following hypothesis:

H2: More households will consistently shop in different stores (exhibit low store loyalty) or primarily in one store (exhibit high store loyalty) in several product classes than would be expected if store loyalty was independent from one product class to another.

THE DATA

The panel data used to test the hypotheses were obtained by NPD Research. The panel consisted of over 4,000 households. The data analyzed cover four product classes: ready-to-eat cereal, facial tissue, shampoo, and deodorants and anti-perspirants. The period analyzed covers the 12 months of 1975.

In order to provide an adequate basis for classifying households, the analysis was based on all households who made at least five purchases in each of the four product classes. This yielded a sample of 385 households.

A brief summary of the data is presented in Table 1. The first row of the table shows the average number of purchases in each of the four product classes. It ranges from 10 for shampoo and deodorants to 46 for cereal. Row two shows the average percent of volume purchased with a coupon (e.g. on average 9.82 of all facial tissue purchases for this sample were made with either a manufacturer or store coupon). Row three shows the percent of households whose fraction of volume was above the average for all households. The corresponding figures for store loyalty are shown in rows four and five.

SUMMARY STATISTICS ON COUPON USAGE AND STORE LOYALTY BY PRODUCT CLASS

METHOD OF ANALYSIS

As a household purchases each product class over an extended period, such as a year, the household makes some percent of its purchases with a coupon. Let PC_{i,c} denote household i's percent of purchase volume that is made with a coupon in product class c. The percent of volume purchased with a coupon is used rather than the total volume of coupon purchases since households differ widely in their total volume of purchases in each product class, and a light user household might otherwise be classified as a light coupon user. The average of PC_{i,c} over all households is denoted by APC_{c}. If PC_{i,c} exceeds APC_{c} the household is considered to be a "heavy" user of coupons for this product class. [While it was possible to split the sample into more than two groups, for instance into "heavy," "medium" and "light" coupon users, this would reduce the sample size in each of the groups and thus might invalidate the statistical procedures used subsequently.]

Similarly, let PS_{i,c} denote household i's percent of purchases for product class c that are made in household i's favorite store for this product class. This is a measure of store loyalty for product class c. The "favorite" store is defined as the one in which the household shops most frequently for that product class. The average of PS_{i,c} over all households is denoted by APS_{c}. If PS_{i,c} exceeds APS_{c} the household is considered to be relatively store loyal for product class c.

Households are considered to be relatively consistent in their coupon usage behavior across product classes to the extent that they are consistently above or below the average for each of the 4 product classes. Similarly, they are considered consistent in their store loyalty across classes to the extent that their store loyalty is uniformly higher or lower than the average for each class. Thus "consistency" in purchase behavior is used here in the sense of systematic repetition in behavior across product classes. Table 2 contains hypothetical values of PC_{i,c} and PS_{i,c} for four households in each of four product classes. This illustrates how households might vary in their coupon usage and store loyalty across product classes. For household 1, PC_{i,c} is above the average (APC_{c}) for all classes. That is, the percent of volume purchased with a coupon for this household is consistently greater than the average percent of volume purchased with a coupon for each of the four classes. This household is also above average in store loyalty for all four product classes. Thus household 1 is considered to be highly consistent both in terms of its coupon usage and store loyalty. Household 2, on the other hand, is consistently store loyal across classes but is not consistent in its coupon usage; it exhibits lower than average coupon usage for 2 of 4 classes.

EXAMPLES OF POSSIBLE PURCHASING PATTERNS FOR ONE YEAR

Let DC_{i,c} and DS_{i,c} denote dummy variables which represent above average coupon usage and store loyalty, respectively, in household i's purchases of product class c. That is,

DC_{i,c}=1 if PC_{i,c} > APC_{c} and = 0 otherwise.

DS_{i,c}=1 if PS_{i,c} > APS_{c} and = 0 otherwise.

While some information is lost by using dummy variables rather than the original percentages, this procedure allows us to use certain statistical tests (such as the chi-square test) which would not be possible otherwise, and captures the phenomena of interest - households' coupon usage and store loyalty relative to the average for the product class.

Let n(CU)_{i} and n(SL)_{i} be the sums, respectively, of DC_{i,c} and DS_{i,c} across the product classes:

n(CU)_{i} = ^{E}_{c} DC_{i,c}

and

n(SL)_{i} = ^{E}_{c} DS_{i,c}

The variables n(CU)_{i} and n(SL)_{i} are measures of consistency in coupon usage and store loyalty across product classes. Note that consistency is greater to the extent that n(CU)_{i} and n(SL)_{i} take on extreme values. Thus, for c=1,...,4, n(CU)_{i} varies from O to 4, with high consistency in coupon usage indicated by values of either 0 (the household is consistently below average in its coupon usage across product classes) or 4 (above average in coupon usage for all classes). Similarly, n(SL)_{i} equals 4 if a household is consistently store loyal and equals 0 if the household is consistently below average in loyalty.

The values for n(CU)_{i} and n(SL)_{i} for the four families in Table 2 are shown in Table 3.

CHARACTERIZING HOUSEHOLDS' CONSISTENCY ACROSS PRODUCT CLASSES (BASED ON DATA SHOWN IN TABLE 2)

Each household can thus be character set by the vector [n(CU)_{i},n(SL)_{i}]. For the 4-product-class case, each of the 2 elements in the vector can have 5 values. The 25 possible outcomes can be represented as cells in a 5X5 matrix as shown in Table 4. In this matrix, the corner cells [O,O], [0,4], [4,0] and [4,4] represent highly consistent behavior; households which fall into these cells are consistently below or above average in both their coupon usage and store loyalty. On the other hand, households which fall into cells along the borders of the matrix (where either n(CU)_{i}=0 or 4 and 0<n(SL)_{i}<4, or n(SL)_{i}=0 or 4 and 0<n(CU)_{i}<4, are consistent on one variable but not the other. Finally, households which fall into the interior cells (where 0<n(CU)_{i},n(SL)_{i}<4) have little or no detectable consistency across product classes

Each of the households in the sample was assigned to one of the cells in the matrix on the basis of its observed n(CU)_{i} and n(SL)_{i} values. The resulting distribution of households across cells is shown in Table 4, along with the expected frequencies for the cells under the assumption of independence in purchase behavior across product classes.

The expected frequency for each cell was computed by considering all the possible combinations of DC_{i,c} and DS_{i,c} that would result in that particular [n(CU)_{i}, n(SL)_{i}] vector, computing the joint probability of each such combination as the product of the observed proportions for each product class, and multiplying the sum of these joint probabilities by the total sample size.

For example, the only possible combination that can result in the cell [0,0] is DC_{i,c}=0 and DS_{i,c}=0 for all c. Thus the expected frequency for that cell, denoted E_{0,0}, is computed as:

E_{0,0} = N. prob[ observing DC_{i,c} =DS_{i,c}=0 for c=1,...,4]

= N.P1,00-P2,00-P2,00-P3,00-P4,00

where N = the total number of households in the sample, and

P_{c,rs} = the overall proportion of households with DC_{i,c}=r and DS_{i,c}=s, where c=1,...,4, r=0,1 and s=0,1.

Similarly, there are four combinations of DC_{i,c} and DS_{i,c} values that can result in the cell [0,1] since there are four product classes and DS_{i,c} can equal 1 for each of them in turn. The probability of each combination is computed from the appropriate Pc,rs values and summed to yield the probability for the cell [0,1]. Probabilities for the other cells in the matrix are computed using the same procedure.

Note that we are treating Pc rs as the "guessing" distribution for households across c, r and s. That is, if a household is drawn at random from the sample, the probability that it will have DC_{i,c}=r and DS_{i,c}=s for a given product class c is P_{c,rs}. By computing the joint probability of observing a specific combination of DC_{i,c} and DS_{i,c} values for each class as the product of the P_{c,rs} values, we are making the assumption that the "guessing" distributions are independent across classes. Thus if the distributions are not in fact independent, the theoretical probabilities computed in this manner will differ from the proportions observed in the data.

RESULTS

As Table 4 shows, the observed frequency of households in the cells along the borders of the matrix (i.e. where n(CU)_{i}=0 or 4, and/or n(SL)_{i}=0 or 4) is much higher than the expected frequency. This suggests that households are more consistent in their coupon usage and store loyalty across product classes than would be expected by chance. In other words, for the households in this sample, information about the extent of coupon usage and store loyalty for one product class can be a useful predictor of these variables for the other product classes.

We first consider the presence of consistency in coupon usage alone. Table 5 shows the observed and expected frequencies of households for different values of n(CU)_{i} (these are simply the row totals from Table 4). A chi-square test revealed that the two were significantly different at the .001 level, thereby rejecting the null hypothesis that coupon usage was independent across product classes. As can be seen, the actual frequency was much higher than expected for extreme values of n(CU)_{i}. For n(CU)_{i}=0, the observed frequency was nearly twice that expected (41% of the sample versus 22%), while for n(cU)_{i}=4, the observed was nearly seven times as large (7Z versus 1%). Thus 48% of the households in the sample were consistently above or below average in their coupon usage for all 4 classes.

A similar pattern was observed for store loyalty. Table 6 shows the distribution of households for different values of n(SL)_{i}, along with the expected distribution if store loyalty was independent across classes (these are the column totals from Table 4.) Again, the chi-square test rejected the hypothesis of independence at the .001 level. Of the 385 households in the sample, 21% were below average in store loyalty for all 4 classes (compared to an expected value of g:), while 17% were consistently above average in store loyalty for all 4 classes (compared to 5% expected). Taking the two together, 37% of the households were consistent in their degree of store loyalty across product classes.

NUMBER OF HOUSEHOLDS WHOSE DEGREE OF STORE LOYALTY WAS ABOVE AVERAGE FOR n PRODUCT CLASSES

Finally, in order to examine the interaction between consistency in coupon usage and in store loyalty, we considered the frequency of households that were either below or above average in coupon usage for 3 or more classes (n(CU)_{i} =0,1,3,4), and below or above average in store loyalty for 3 or more classes (n(SL)_{i}=0,1,3,4). [The cells were defined in this manner to ensure that the sample size in each was large enough to permit the use of the chi-square test.] These households, therefore, were those that exhibited moderate to high consistency in both coupon usage and store loyalty. Table 7 shows the frequency of households under this classification scheme.

Of the households that were below average in store loyalty, nearly a third were above average in coupon usage (Cell C in Table 7). At the same time, among those households that were above average in store loyalty, less than a sixth were above average in coupon usage as well (Cell D). These results suggest the presence of a strong interaction effect between coupon usage and store loyalty for households that are relatively consistent in their purchase behavior across classes. Those who are consistently below average in store loyalty are more likely to be above average in coupon usage than those who are consistently store loyal, and vice-versa. A chi-square test of independence showed the interaction to be significant at the .001 level.

Two cells in Table 7 are of particular interest. The households in Cell C are consistently heavy users of coupons and exhibit low store loyalty for most product classes. This group, which might be described as "activist shoppers", constituted 11% of the entire sample. The household in Cell B are quite the opposite. These households (25: of the sample) are consistently store loyal and consistently below average in coupon usage. They might be labeled "inert" shoppers.

The above analyses were repeated for households that made 2 or more purchases over the 12-month period. Similar results were obtained for this larger sample of 1247 households.

CONCLUSIONS

The major finding is that more households are consistent in their usage of coupons and their degree of store loyalty across product classes than would be expected by chance. The finding concerning consistency in coupon usage is an important one, both from a theoretical and a managerial perspective, since it indicates that knowledge about households' purchase behavior in one product class can be a useful predictor of behavior in other classes. Thus managers may be able to predict the effectiveness of coupon promotions in a given product class based on information about coupon redemptions in other product classes in the past.

The fact that households tend to be relatively consistent in their store loyalty also has important implications for retailers. In particular, it indicates that if consumers can be induced to shop in one store for a given product class (for example, by providing a large selection of brands and frequent "specials" for that product class), they will tend to make their purchases of other product classes in that store as well, thus increasing total revenues for that retailer.

One can also identify an "activist" segment consisting of households that are both high in coupon usage and low in store loyalty (see Cell C in Table 7). From a managerial standpoint this is an important segment to study since their response elasticity to coupons and other promotions is probably the highest Future research may be able to profile this segment in terms of its demographic descriptors and media habits, which would allow managers to target their promotional activities more efficiently. Another possibility is to determine if this group would still redeem coupons even if they were for lower face values or were harder to collect.

There is also a larger than expected segment of households that is above average in store loyalty and below average in coupon redemption (Cell B, Table 7). If one is to promote to this "inert" segment, one might consider alternatives to coupons, such as free samples. Alternatively, one might decide that it is not profitable to promote to this group at all.

Yet another possibility is to focus on households that consistently use an above average level of coupons for several product classes (see Table 5). These households are likely to be more responsive to multiple brand coupons (coupons which are redeemable only if two or more brands are purchased). Again, demographics and media habits could be studied for this group and used to reach it more efficiently.

REFERENCES

Blattberg, Robert C., Peter Peacock, and Subrata K. Sen (1976), "Purchasing Strategies across Product Classes," Journal of Consumer Research, 3 (December), 143-54.

Cunningham, Ross M. (1956), "Brand Loyalty - What, Where and How Much", Harvard Business Review, 34 (January-February), 116-128.

Massy, William F., Ronald E. Frank and Thomas Lodahl (1968), Purchasing Behavior and Personal Attributes, Philadelphia: University of Pennsylvania Press.

Wind, Yoram and Ronald E. Frank (1969), "Interproduct Household Loyalty to Brands," Journal of Marketing Research, 6 (November), 434-435.

----------------------------------------

Tweet
window.twttr = (function (d, s, id) { var js, fjs = d.getElementsByTagName(s)[0], t = window.twttr || {}; if (d.getElementById(id)) return; js = d.createElement(s); js.id = id; js.src = "https://platform.twitter.com/widgets.js"; fjs.parentNode.insertBefore(js, fjs); t._e = []; t.ready = function (f) { t._e.push(f); }; return t; } (document, "script", "twitter-wjs"));