On the Predictive Accuracy of Subjective Purchase Probabilities


Donald H. Granbois and John O. Summers (1972) ,"On the Predictive Accuracy of Subjective Purchase Probabilities", in SV - Proceedings of the Third Annual Conference of the Association for Consumer Research, eds. M. Venkatesan, Chicago, IL : Association for Consumer Research, Pages: 502-511.

Proceedings of the Third Annual Conference of the Association for Consumer Research, 1972      Pages 502-511


Donald H. Granbois, Indiana University

John O. Summers, Indiana University

[Donald H. Granbois is Professor of Marketing, John O. Summers is Associate Professor of Marketing, Indiana University.]

Empirical study of the household decision process for major purchases has relevance for at least three theoretical topics in consumer behavior. Purchases of items such as automobiles, major appliances and vacations are infrequent and costly, and they often involve joint consumption or use behavior. Therefore, these purchases are likely to involve more than one family member (Davis, 1970, p. 170). Such decisions are thus ideal for investigating family decision-making role structure. For similar reasons, the process of deliberation and searching out information on available alternatives and their characteristics is likely to be more complex and fully articulated than is true for nondurables (Katona & Mueller, 1954), and these decisions therefore enable study of the character of these phases of the decision process. Finally, the concept of intention (or problem recognition) and its relationship with the outcome of the decision process is ideally studied in the context of major purchases, since the length of time separating intention and outcome is apt to be long enough to permit relatively independent estimates of each (Howard & Sheth, 1969, p. 133). The study we will report here is most relevant for this third theoretical topic and considers the degree to which purchase intentions measured with a subjective probabilities instrument can predict purchase behavior.

From a policy-making perspective, both producers and governmental policy makers are concerned with such operational problems as short-run forecasts of demand for total durable goods. The magnitude and volatility of this demand has important implications for both marketing decisions and economic policy. Forecasting models typically include purchase intentions and fulfillment rates as variables. Conceptually, period-to-period changes in durable goods demand result from non-offsetting changes in the proportion of households having purchase intentions, the rate of intentions fulfillment, and the purchase rate among "non-intenders " (Juster, 1966). Empirical study of the decision process and the role of intentions is relevant here, too, since discovery of determinants and patterns of interrelationships among these three components may suggest significant improvements in forecasting models based upon data regularly collected in purchase intentions surveys (Granbois, 1971a). We think our findings may have relevance for such model building.


Purchase intentions data have been integrated into forecasts since the Survey Research Center at the University of Michigan first started collecting them in 1945. However, these data have been considerably less predictive of actual behavior than desired. Longitudinal studies in which respondents' later behavior has been measured have revealed that the traditional survey data are poor predictors of short-run changes in demand because fulfillment rates are low, purchase rates among "non-intenders" are high, and yet these rates are not highly correlated over time so they cannot be systematically . adjusted by the analyst (Juster, 1966).

At least five strategies seem feasible for improving the accuracy of forecasts based upon intentions data:

(1) Improve the initial measure of intentions, so as to reduce the magnitude and variability of the "unplanned" purchase rate and to increase the magnitude and reduce the variability of the fulfillment rate. In the last few years, simple intentions questions have been replaced by measures of subjective purchase probabilities that elicit responses in terms of "Chances out of 10" (or 100) of purchase during three-, six- or twelve-month future periods. These probabilistic questions provide significantly better predictions of purchase rates than do intentions questions (Juster, 1966; McNeil & Stoterau, 1968; Granbois & Willett, 1968; Clawson, 1971).

(2) Study the relationships between intentions, fulfillment and unplanned purchase rates and the contingencies (unexpected expenses or income, changes in prices, unemployment, etc.) with which they may vary. At least one attempt has been made to prepare separate forecasts of these contingencies and to modify the forecasted rates accordingly (Kosobud & Morgan, 1964).

(3) Study the consumer search process intervening between intentions and outcome to find the effects of new information on consumer intentions fulfillment and unplanned purchase rates (Pratt, 1965).

(4) Disaggregate so that separate forecasts are made for categories of products for which relationships between intentions, fulfillment and unplanned purchase rates differ, or for categories of products for which these rates respond differently to exogenous variables, such as price levels and income change. Categories may involve different generic products (refrigerators, dishwashers, etc.) for which such dimensions as length of planning period may vary (Pratt, 1968); different kinds of purchase circumstance, such as first-time acquisition or replacement (Heald, no date); or different degrees of urgency or priority of acquisition (McFall, 1969). Another basis for disaggregation might be demographic segments, such as income groups. (Namais, 1960).

(5) Study patterns of relationships between intentions, preferences and outcome for both husbands and wives (and perhaps older children) since certain family members may be better predictors of household behavior than others, and since degree of conflict or difference may be a key determinant of fulfillment rates. (Granbois, 1971b). Comparisons of husbands' and wives' responses have revealed discrepancies in studies of family participation in major purchase decisions (Granbois & Willett, 1970; Davis, 1970) and in measures of major purchase plans and plan fulfillment. (Wolgast, 1958).

The design of our study was influenced by three of these strategies. We collected subjective purchase probabilities for future major purchases (Strategy #1) from husbands and wives independently and from the same couples jointly (Strategy #5), and later determined the actual purchase behavior of these families. Our sample, though small, permitted some disaggregation by product category and demographic segment (Strategy #4). We did not study intervening search behavior (Strategy #3) or the effects of contingencies (Strategy #2), although our method could easily be modified to include these factors.


A group of 77 married couples participated in a behavioral laboratory study of decision processes for major expenditures, [This was part of a joint study conducted by Donald Granbois and Douglas Longman. Responses to three other instruments used and detail on methodological findings are found in Longman (1970).] in which one instrument asked for major purchases ($100 or more) planned during the coming year. A subjective probability in the form of an eleven-point "Chances out of 10" scale, estimated cost, and expected month of purchase were required for each item listed. The instrument was first administered to each husband and wife separately; each couple then repeated the exercise, discussing their responses and completing a joint form representing their consensus. Responses were not restricted to durable goods plans since these were thought to interact with other major expenditures (vacations, investments, etc.).

A mail questionnaire one year later determined actual purchase behavior. The follow-up questionnaire asked for purchase details (or reasons for not purchasing) for each purchase plan listed in both the individual and joint phases of the laboratory study, and for any other major expenditures during the period. Changes in household circumstances, income, working status of husband ant wife, and other expenses were also measured.


Our results involve three types of evaluation of subjective purchase probability responses:

(1) Analysis of the comparative predictive accuracy of aggregated husbands' responses, aggregated wives' responses, and aggregated responses given by husbands and wives jointly.

(2) Evaluation of the predictive accuracy of response for individual purchase item categories, such as automobiles, vacations and travel, etc.

(3) Evaluation of several possible determinants of the variability in predictive accuracy of response including age, income and wife's work status.

Analysis of Aggregate Responses

The 77 couples produced more plans responding jointly than either husbands or wives responding singly; expected purchase rates (mean subjective purchase probabilities) for each set were quite similar; and purchase rates varied somewhat over the three response sets (see Table 1).

Subjective probabilities were somewhat concentrated on even numbers, probably because verbal cues were attached to these numbers on the response form. With a few exceptions, purchase rates tended to increase uniformly from low to high probabilities, as expected. In each set, expected rates considerably exceeded actual purchase rates, a finding also reported by Clawson (1971) for a quite different set of items, but contrasting with Juster's (1966) results. On the basis of this simple aggregate analysis, there is little basis for favoring husbands', wives' or joint subjective probabilities as more accurate predictors of purchase rates, although the discrepancy between actual and expected purchase rates was greatest for wives and least for joint responses. Because these averages were aggregated over several item categories and over respondents varying in such demographic characteristics as age and income, they may have disguised several important variations in incidence of plans and in the relationship between purchase probability and outcome.

Variations by Item Category

Purchase plans were classified into nine categories and incidence of plans, expected purchase rates (mean subjective purchase probabilities) and purchase rates were computed for each category. Table 2 summarizes these data for each of the three response sets.





Wives listed many more plans for furniture and carpeting than did husbands, and husbands' plans were more evenly distributed among categories than were wives' responses, but the rank order of plans by category was-quite similar for all three sets of responses. Spearman rank correlations were .979 comparing husbands' and wives', .954 comparing husbands' and joint, and .962 comparing wives' and joint responses. Clothing appears as a category because several respondents anticipated "clustered" purchases expected to cost $100 or more. In almost every category, more plans were generated by joint responses than by husbands or wives responding individually, partly because "new" plans were created during discussion and partly because joint responses represented a merger of items originally listed by just one spouse but agreed upon by both during discussion.

Expected purchase rates by purchase category exhibit two striking characteristics: Substantial variation appears across categories (54.5% to 87.5% for joint responses, for example) and very close correspondence across all three sets of responses. Spearman rank correlations were .817 comparing husbands' and wives', .944 comparing husbands' and joint, and .870 comparing wives' and joint responses. Highest mean probabilities were assigned to clothing, services and investments, and miscellaneous items (an extremely heterogeneous category including cameras, art, watches, sporting goods and recreational equipment, etc.). Lowest expected purchase rates were given to home entertainment equipment, appliances, and furniture and carpeting plans. High probabilities appeared to be associated with items purchased on a seasonal or time-specific basis ("clustered" clothing purchases and travel and vacations) and with items for which a specific item was contemplated (services and investments and miscellaneous). Items for which low probabilities were assigned were either in categories where an uncertain or discretionary need for replacement triggered the plan, where the item was purely discretionary (and therefore both postponable and likely to compete with other discretionary purchases) or where the purchase was contingent on finding the "right" style, color, or price, as in furniture and carpeting.

The difference between actual and expected purchase rates can be regarded as a measure of the predictive accuracy of subjective purchase probabilities. Except for auto and clothing plans, expected rates always exceeded purchase rates, with considerable variation in the magnitude of these differences across purchase categories. Inspection reveals a slight systematic relationship between the magnitude of the expected purchase rate and the difference between actual and expected rates, in that higher discrepancies tended to be associated with lower expected rates. Variations in the difference measure are perhaps also explained by the characteristics of each purchase category involved. Subjective probabilities predicted auto, travel and clothing purchase rates best. [Clothing and service and investment data are based on very small cell sizes, and not much significance should be attributed to findings for these categories.] The time-specific nature of clothing and travel plans may have made them more predictable, and the fact that most respondents had made several auto purchases in the past may have improved the accuracy of these probabilities. Home addition and remodeling plans are likely to.involve many uncertainties with respect to costs, materials, design and availability of craftsmen. Thus, unexpected difficulties and delays may have depressed the actual purchase rates. Appliance and home entertainment equipment plans, both low in expected rates, exhibited large discrepancies between actual and expected rates, perhaps because of the uncertainty associated with replacement need and the almost totally discretionary nature of many items in these categories.

The general pattern of differences between actual and expected rates was quite similar across all three sets of responses, suggesting that the predictive accuracy of subjective probabilities varies much more across purchase categories than it does among the three respondent types.

Effects of Husband's Age, Income and Wife's Work Status

In contrast to the analysis of the effects of purchase category, few systematic relationships appeared when incidence of plans, expected purchase rate, and predictive accuracy were computed for husband's age, husband's income and wife's work status categories. Table 3 summarizes these findings. A slight tendency for older men to assign higher subjective probabilities to their purchase plans is evident, but the small cell sizes for respondents over 50 and the relatively high probabilities for young men (under 30) suggest caution in generalizing this tendency. Discounting the two lowest income groups because of small cell size, there is a somewhat stronger positive relationship between expected purchase rates and income for joint and husbands' responses. Since wives' responses do not exhibit this relationship, however, this finding should also be interpreted with caution. No clear pattern of relationship between predictive accuracy and either husband's age or husband's income appears. Wife's work status seems unrelated to any of the three dependent variables and, somewhat surprisingly, number of plans per respondent was not related to any of the three demographic variables. With a few exceptions, patterns of response across joint, husbands' and wives' responses were quite similar. Thus, as in the aggregate and purchase category analyses, respondent choice seems to make no difference in the investigation of demographic groupings.


Unlike earlier studies, the design here did not limit respondents to a predetermined list of possible purchases, but rather, they originated their own purchase plans in a kind of unaided response environment. Perhaps for this reason, more plans were generated by husbands and wives interacting than when each spouse listed plans independently. Should the method be used specifically for generating estimates of the magnitude of short-run changes in durable goods demand, it is likely that better estimates could be made using joint response data than responses from either husbands or wives. This conclusion should be tested using a predetermined list of durables instead of the unaided technique used here.

Mean subjective purchase probabilities exceeded actual purchase rates and, although actual purchase rates were positively related to the expected purchase rates, enough variation in the predictive accuracy of subjective probabilities was found to encourage the pursuit of strategies for improving their performance.

On the basis of the findings reported here, disaggregating by purchase category and (to a lesser extent) by demographic characteristics seems to be a more promising strategy for improving predictive accuracy than using husbands and wives responding jointly as respondents in preference to using husbands or wives alone. Research on the comparative responses of husbands and wives should continue, however, since respondent couples in the present study were somewhat homogeneous with respect to education and social class, and the behavioral laboratory environment may have resulted in somewhat different responses than would have occurred in an at-home survey setting. Greater differences between husbands and wives, as well as greater variations among demographic segments, might therefore be discovered in future research.



Attention should be directed to exploration of additional bases for disaggregation not touched on here. In particular, replacement versus first-time acquisition plans and the extent of previous purchase histories may affect the predictive accuracy of subjective probabilities. Finally, the strategies not examined here--studying the consumer search process and taking into account the role of contingencies--deserve investigation.


Clawson, C. J. How Useful are 90-day Purchase Probabilities? Journal of Marketing, 1971, 35, 43-47.

Davis, H. L. Dimensions of Marital Roles in Consumer Decision Making. Journal of Marketing Research, 1970, 7, 168-177.

Granbois, D. H. Decision Processes for Major Durable Goods. In G. Fisk (ed.), New Essays in Marketing Theory. Boston: Allyn and Bacon, 1971 a, pp. 172-205.

Granbois, D. H. A Multi-level Approach to Family Role Structure Research. In D. M. Gardner (ed.), Proceedings of the 2nd Annual Conference of the Association for Consumer Research, 1971 b, pp. 99-107.

Granbois, D. H. & Willet, R. P. An Empirical Test of Probabilistic Intentions and Preference Models for Consumer Durables Purchasing. In R. L. King (ed.), Marketing and the New Science of Planning. Chicago: American Marketing Association, 1968, pp. 401-408.

Granbois, D. H. & Willett, R. P. Equivalence of Family Role Measures Based on Husband and Wife Data. Journal of Marriage and the Family, 1970, 32, 68-72.

Heald, G. The Use of Consumer Attitude and Buying Intentions Data as an Aid to Government Policy Making. (Unpublished manuscript).

Howard, J. A. & Sheth, J. N. The Theory of Buyer Behavior. New York: John Wiley, 1969.

Juster, F. T. Consumer Buying Intentions and Purchase Probability: An Experiment in Survey Design. Journal of the American Statistical Association, 1966, 61, 658-696.

Katona, G. & Mueller, E. A Study of Purchase Decisions. In L. Clark (ed.), The Dynamics of Consumer Reaction. Consumer Behavior, Volume 1. New York: New York University Press, 1954, pp. 30-87.

Kosobud, R. F. & Morgan, J. N. Consumer Behavior of Individual Families Over Two and Three Years. Ann Arbor: Survey Research Center, University of Michigan, 1964.

Longman, D. S. A Laboratory Study of Husband-wife Interaction in Consumption Decision Making. Unpublished doctoral dissertation, Graduate School of Business, Indiana University, 1970.

McFall, J. Priority Patterns and Consumer Behavior. Journal of Marketing, 1969, 33, 50-55.

McNeil, J. M. & Stoterau, T. L. The Census Bureau's New Survey of Consumer Buying Expectations. 1967 Proceedings of the Business and Economic Statistics Sections American Statistical Association, 97-113.

Namais, J. Intentions to Purchase Related to Consumer Characteristics. Journal of Marketing, 1960, 25, 32-36.

Pratt, R. W., Jr. Understanding the Decision Process for Consumer Durable Goods. In P. D. Bennett (ed.), Marketing and Economic Development. Chicago: American Marketing Association, 1965, pp. 244-260.

Pratt, R. W., Jr. Using Research to Reduce Risk Associated with Marketing New Products. In R. Moyer (ed.), Changing Marketing Systems. Chicago: American Marketing Association, 1967, pp. 98-104.

Wolgast, E. Do Husbands or Wives make the Purchasing Decisions? Journal of Marketing, 1958, 2 , 151-158.



Donald H. Granbois, Indiana University
John O. Summers, Indiana University


SV - Proceedings of the Third Annual Conference of the Association for Consumer Research | 1972

Share Proceeding

Featured papers

See More


B5. Money Doesn’t Buy Happiness, But What About Buying Trust? The Effectiveness of Financial Compensation in Restoring Trust After Double Deviation

Valentina Ortiz Ubal, Universidade Federal do Rio Grande do Sul (UFRGS)
Cristiane Pizzutti, UFRGS
Katja Gelbrich, Catholic University Eichstätt-Ingolstadt

Read More


O5. The Effect of Synchrony on Non-Human Objects Involved in the Synchronous Performance

Xiaoyin YE, Xiamen University
Jun YE, Xiamen University

Read More


Can Making Family Salient Improve Retirement Contributions? Evidence from Field Experiments in Mexico

Avni Shah, University of Toronto, Canada
Matthew Osborne, University of Toronto, Canada
Jaclyn Lefkowitz, IDEAS42
Andrew Fertig, IDEAS42
Dilip Soman, University of Toronto, Canada
Nina Mazar, Boston University, USA

Read More

Engage with Us

Becoming an Association for Consumer Research member is simple. Membership in ACR is relatively inexpensive, but brings significant benefits to its members.