Do Grocery Consumers Make Trial Purchases?
Malcolm Wright and Helen Peat, Massey University
Abstract
The assumption that consumers trial new brands and adopt or reject them depending on their experience is integral both to consumer behaviour and to trial-repeat models in marketing science. However, recent research has suggested grocery products may experience nearinstant loyalty, raising questions about the assumptions of trial-repeat models. We investigated this using scanner panel data for a frequently purchased grocery product (n=439).
The results showed little evidence of a trial-adoption process actually taking place. So called
‘adopters’ did not continue to buy, and the well known conversion to higher depth of repeat classes turned out to be present for an established brand to the same degree as the new brand, suggesting that it is just an expression of the distribution of light and heavy buyers. On the other hand, repeat purchase rates were lower than expected for both new and established brands. Overall, the results suggest that the trial process was having an effect on repeat purchase, but only at the margin for light buyers, and in a way that is not captured by trialrepeat models. Some patterns of consumer behaviour assumed by trial repeat models are present in the market, but appear to have nothing to do with new brand performance. Thus, a near-instant loyalty hypothesis is to be preferred to a trial-repeat hypothesis, and an important area of market modeling appears to be based on false assumptions.
Introduction
We generally expect a new grocery brand to go through an introductory period. During this time, potential adopters will make one or more trial purchases, evaluate the consumption experience, and decide whether to adopt or reject the brand. These assumptions are embedded in the familiar information processing models of consumer behaviour, but interestingly they also play an important role in the trial-repeat models of new product performance common in the marketing science literature.
The marketing science trial-repeat models initially used panel data to attempt to predict brand share from early data. Fourt and Woodlock (1960) assumed that, for a new brand, greater depth of repeat buying was more likely to lead to conversion. They analysed conversion ratios over successive repeat levels and made cumulative predictions of brand performance.
However, while the model gave a good fit for short time frames, it provided poor predictions for longer time frames. Parfitt and Collins (1968) had more success by combining a buying rate index (controlling for light/heavy category purchase) with penetration predictions and repeat purchasing estimates. Eskin’s (1973) well-known contribution was to add methods for estimating the conversion of triers from one repeat class to the next, although the empirical success of his approach was also mixed.
Later approaches moved away from panel data to obtain pre-launch performance assessments from survey data and experiments. TRACKER was well known at the time (Blattberg and
Golanty 1978), but by far the most influential models were Silk and Urban’s (1978)
ASSESSOR which took account of awareness, availability, trial and repeat purchase, and the extended ASSESSOR (Kalwani and Silk 1980) which expanded the depth of repeat purchase
ANZMAC 2002 Conference Proceedings 2401
classes examined. These later trial-repeat models have had a great influence on commercial pre-launch forecasting products, such as ACNielsen’s BASES.
While these models have employed slightly different assumptions and techniques, a key element is that consumers are assumed to undertake trial purchase to determine whether the new brand is of comparable or superior value to established brands, and therefore worth subsequent purchasing (Kalwani and Silk 1980). Other common themes include identification of the actual trial rates, identification of actual conversion rates, and determining conversion to higher classes of depth of repeat purchase.
But could the trial-repeat models be wrong? What if the assumption of an identifiable trial, followed by an adoption/rejection decision cannot be empirically justified? What if conversion to higher classes of depth of repeat purchase is the same for a new brand as for an existing brand? There is evidence to suggest that patterns of purchase loyalty become normal almost immediately (Wellan and Ehrenberg 1988, Ehrenberg and Goodhardt 2000, Wright and Sharp 2001) which seems inconsistent with the themes of trial-repeat modeling.
One explanation is that the trial-adoption process is over very quickly – although this is not the assumption of the trial-repeat models. Another explanation is that the high proportion of light buyers present in grocery markets are being mistaken for those following a trial-rejection pattern of purchase. If so, it implies that the commercial trial-repeat models are based on a false premise. Although some researchers might be happy with theoretically incorrect models that provide instrumentally useful predictions, we believe that diagnosing and correcting theoretical problems is a pre-requisite for improving the scope and accuracy of our modelling.
Objectives
This research sought to assess whether grocery consumers make identifiable trial purchases preceding the adoption or rejection of new grocery products, and whether patterns of repeat purchase for a new brand were any different from those for existing brands. In particular, following the recommendations of Armstrong, Brodie and Parsons (2001), we sought to compare the explanatory power of two competing hypotheses: the trial-repeat hypothesis, that new brands follow a pattern of trial purchasing followed by a discrete adoption or rejection, with conversion to higher depth of repeat purchase playing an important role in adoption, and the near-instant loyalty hypothesis, that new brands behave like established brands right from the early stages of introduction. To evaluate the assumptions of trial-repeat models in the light of these competing hypotheses, we examined three propositions.
1. Consumers who adopt a new brand continue as adopters of that brand.
2. Over time adopters convert to higher depth of repeat classes.
3. Repeat purchase rates will be lower for new brands than for existing brands.
These propositions are all consistent with the trial-repeat hypothesis. However, the nearinstant loyalty hypothesis suggests that: Proposition 1 would not be supported, as light buyers may not buy in the following period; Proposition 2 is just as true for existing brands as for new brands, and; Proposition 3 is only true for a very short period of time.
Given these expectations, we can usefully add a fourth proposition that is consistent with the near-instant loyalty hypothesis, but inconsistent with the trial-repeat hypothesis.
ANZMAC 2002 Conference Proceedings 2402
4. The new brand will behave exactly like an established brand within a very short time.
Data and Method
We obtained a data set for a frequently purchased grocery product from ACNielsen’s New
Zealand HomeScan panel. Due to confidentiality requirements, the exact category cannot be named, but it is purchased at least weekly by most households. The data consisted of 52 weeks of brand purchase information for 439 households who were buyers of the category.
Twenty-one brands were initially present in the category, with the market was dominated by three main brands. A new brand was introduced into the market in the 20 th week of the data.
We divided the data following the new brand launch into two sixteen week periods, treating the first sixteen weeks as an introductory period, and the second sixteen weeks as post-launch market maturity. We examined the performance of the new brand and, for comparative purposes, an established brand of similar size. The 16-week penetration of these two brands varied from 14% to 18%, while average purchase frequency varied from about 4 to about 5.
To evaluate Proposition 1, we classified households who purchased a brand twice within the first sixteen weeks as ‘adopters’ and all others as ‘rejecters’. For the second sixteen weeks, we similarly classified households as ‘buyers’ or ‘non-buyers’. A cross-tabulation of adopters/rejecters with buyers/non-buyers then allowed Proposition 1 to be evaluated.
To evaluate Proposition 2, we identified those that had made a purchase in the first sixteen weeks, and examined how many had made at least one purchase in the second sixteen weeks.
We then calculated conversion to each higher depth of repeat class in the second sixteen weeks; for example, of those who made at least one purchase, what proportion made more than one, of those who made at least two, what proportion made more than two, and so on.
To evaluate Proposition 3, we calculated observed repeat purchase rates between the two sixteen week periods, and used the observations from the first period to calculate Negative
Binomial Distribution theoretical norms against which these observations could be compared.
The results of these analyses allowed us to make a qualitative assessment of Proposition 4.
Results
Table 1 shows an analysis of the adopters of the new brand. We selected 2 purchases as the cutoff for adopter classification as clearly the first purchase could only be seen as a trial.
Table 1: Changes in Adoption Status (New Brand)
Maturity
Introduction
Adopter 28 20
Rejecter 16 373
ANZMAC 2002 Conference Proceedings 2403
Table 1 shows that 48 households made 2 or more purchases of the new brand in the introductory period. However, only 28 maintained this behaviour in maturity. Conversely,
16 of the rejecters from the first period became buyers (the equivalent of ‘adopters’) in the second period. Thus, households who ‘adopt’ a new brand do not necessarily continue to purchase it at all. Proposition 1 is not supported.
Table 2 undertakes a similar analysis for an established brand of similar size.
Table 2: Changes in Adoption Status (Comparative Brand)
Maturity
Introduction
Adopter 23 15
Rejecter 15 386
The same patterns are apparent here, except that ‘adopters’ who drop out in the second period are exactly matched by ‘rejectors’ coming into the market. As this is an established brand, all those who bought in either period must actually be ongoing buyers of the brand; it is just that many of them are light buyers whose purchases vary from period to period. This is classic behaviour for a stationary grocery category with some ‘churn’ in light buyers.
The lack of balance in Table 1 between those who ‘discontinue’ (20) and those who freshly
‘adopt’ (16) in the second period may reflect a new brand effect, or it may just be random variation. However, any new brand effect clearly operates at the margin of light buying, and does not justify the appellation of a high-involvement trial-adoption process.
Table 3 reports conversion to higher depth of repeat classes for those who made any purchase at all in the ‘introductory’ period. Thus, for buyers of the new brand, 54% bought again in the second period. Of these 54%, 67% made two or more purchases, and so on. This process of identifying conversion to higher depth of repeat classes is a common theme in trial-repeat modeling. In this case there is a lower ‘bar’ for adoption that in the previous tables; here a single purchase will do, as we interested in migration from lower to higher levels of purchase.
Table 3: Conversion Percentages by Depth of Repeat Class
One Purchase
Two Purchases
Three Purchases
Four+ Purchases
Brand
54
67
69
73
Brand
54
68
71
74
The results for the new brand are consistent with Proposition 2 and the trial-repeat hypothesis.
Those who purchase in the ‘introductory’ period and re-purchase in the ‘maturity’ period, show a steady propensity to increase to greater levels of depth of repeat purchase. However, as expected from the near-instant loyalty hypothesis, Table 2 also shows an identical pattern for the established comparative brand! This suggests that Proposition 2 may be misexpressed; rather than conversion to higher depth of repeat classes, all that really appears to be present is the standard distribution of light and heavy buyers seen for mature brands.
ANZMAC 2002 Conference Proceedings 2404
Overall, these results must be said to support the near-instant loyalty hypothesis; the new brand is behaving like an existing by the second period (Proposition 4).
Table 4 shows the proportion of repeat buyers from the first to the second period, together with NBD theoretical norms from the first period. Results are reported for both the new brand and the comparative brand. By way of linking the analyses it is worth pointing out that 81 households bought the new brand in first period, and of these 44 bought again second period; this corresponds to the values in Table 3 (44/81 = .54) and Table 4 (44/439 = .10)
Table 4: Repeat Purchase Rates
New
Brand
Comparative
Brand
NBD Theoretical 0.14 0.11
The results show that, as expected from the trial-repeat hypothesis, the observed repeat purchase rate is lower than the theoretical norm for a brand of that size. Surprisingly, however, this is also true for the established comparative brand. This may be due to some
‘shopping around’ trying the new brand instead of repeat purchasing the established brand, or it may be due to some other aspect of market non-stationarity. Whatever the explanation, these results support Proposition 3 and the trial-adoption hypothesis rather than the nearinstant loyalty hypothesis.
Conclusions and Implications
As expected from the near-instant loyalty hypothesis, Proposition 1 is not supported and socalled ‘adopters’ turn out not to be adopters at all; the pattern of ‘adoption-rejection’ in the data is almost certainly due to a light buying effect. Similarly, Proposition 2 is supported equally for the new and existing brands, and so seems to be a claim about the distribution of purchases for all brands, not just new brands. These are heavy blows against the trial-repeat hypothesis. The trial-repeat hypothesis does gain some relief from the support for Proposition
3, that repeat purchase is lower than normal for the new brand. However, the best explanation of this is that there is something occurring at the margin of light buying, such as a temporary minor reallocation of purchases: this is not captured by the standard trial-repeat models.
Thus, while trial-repeat patterns are present in the market, they appear to have little to do with new brand performance, except insofar as new brand performance is exactly like established brand performance. This supports the near-instant loyalty hypothesis and Proposition 4.
Further research could usefully examine more categories and different countries. However, these results already suggest that it is time to revisit some long-standing models of new product performance, and the related commercial pre-launch forecasting products. While the commercial products are relatively accurate, there is still a degree of unresolved error; perhaps the next stage of their development should be greater emphasis on understanding light buying and temporary changes to repeat purchase. At the very least, this paper should encourage the developers of commercial pre-launch forecasting products to compare observations for new brands with equivalent observations for existing brands. They may well find the results surprising.
ANZMAC 2002 Conference Proceedings 2405
References
Armstrong, J.S., Brodie, R.J., and Parsons, A., 2001. Hypotheses in Marketing Science:
Literature Review and Publication Audit. Marketing Letters. 12 (2), 171-187.
Blattberg, R.C. and Golanty, J., 1978. Tracker: An Early Test-Market Forecasting and
Diagnostic Model for New Product Planning. Journal of Marketing Research. 15, 192-202.
Ehrenberg, A.S.C. and Goodhardt, G.J., 2000. New Brands: Near Instant Loyalty. Journal of
Marketing Management. 16, 607-617.
Eskin, G.J., 1973. Dynamic Forecasts of New Product Demand Using a Depth of Repeat
Model. Journal of Marketing Research. 10, 115-129.
Fourt, L.A. and Woodlock, J.W., 1960. Early Prediction of Market Success for New Grocery
Products. Journal of Marketing. 24, 31-38.
Kalwani, M.U. and Silk, A.J., 1980. Structure of Repeat Buying for New Packaged Goods.
Journal of Marketing Research. 17, 316-322.
Parfitt, J.H. and Collins, B.J.K., 1968. Use of Consumer Panels for Brand Share Prediction.
Journal of Marketing Research. 5, 131-145.
Silk, A.J. and Urban, G.L., 1978. Pre-Test Market Evaluation of New Packaged Goods: A
Model and Measurement Methodology. Journal of Marketing Research. 15 (171-191).
Wellan, D.M. and Ehrenberg, A.S.C., 1988. A Successful New Brand: Shield. Journal of the
Market Research Society. 30 (1), 35-45.
Wright, M. and Sharp, A., 2001. The Effect of a New Brand Entrant on a Market. Journal of
Empirical Generalisations in Marketing Science. 6, 15-29.
ANZMAC 2002 Conference Proceedings 2406