Buying More Than We Can Use: Factors Influencing Forecasts of Consumption Quantity

Mary Frances Luce, Duke University
ABSTRACT - This paper develops a process description of consumer forecasts concerning the optimal quantity of repeatedly consumed goods. The case where limited availability necessitates a forecast over an unusually long time period is of particular interest. It is hypothesized that extension of the forecast period will induce upward bias in quantity forecasts. It is further hypothesized that reliance upon current preferences in the absence of the ability to predict future preferences is a primary factor underlying this bias. The roles of confidence in current preferences, of safety stock, and of forecast heuristics are also discussed.
[ to cite ]:
Mary Frances Luce (1992) ,"Buying More Than We Can Use: Factors Influencing Forecasts of Consumption Quantity", in NA - Advances in Consumer Research Volume 19, eds. John F. Sherry, Jr. and Brian Sternthal, Provo, UT : Association for Consumer Research, Pages: 584-588.

Advances in Consumer Research Volume 19, 1992      Pages 584-588


Mary Frances Luce, Duke University


This paper develops a process description of consumer forecasts concerning the optimal quantity of repeatedly consumed goods. The case where limited availability necessitates a forecast over an unusually long time period is of particular interest. It is hypothesized that extension of the forecast period will induce upward bias in quantity forecasts. It is further hypothesized that reliance upon current preferences in the absence of the ability to predict future preferences is a primary factor underlying this bias. The roles of confidence in current preferences, of safety stock, and of forecast heuristics are also discussed.

There has been recent interest in individuals' ability (or lack of ability) to understand their own preferences. For instance, theoretical (March 1978) and empirical (Kahneman and Snell 1990; Varey and Kahneman In press) work suggests that people are unable to accurately predict their future preferences. The related issue to be discussed in this paper is how, and with what degree of accuracy, individuals forecast their utility maximizing purchase quantity. Little work has been done to date investigating the processes by which individuals decide how much of a good to buy. Instead, research has concentrated on the manner in which consumers decide what to buy. However, the quantity decision is a very important part of the purchase process. Many sellers, notably those handling food and clothing, rely on consumers to make large, repeat purchases. Therefore, encouraging multiple purchases is often one of the seller's major goals. The quantity forecasting issue will be examined primarily within a special case. This case is that of a relatively long forecast period caused by limited availability of the good under consideration.

It may seem that the quantity decision made during any particular shopping trip is unimportant; one may simply correct a suboptimal quantity decision on the following trip. This would be true if consumption rate were exogenous to inventory level. However, there is some evidence that consumption rate is a function of inventory level (Folkes, Gupta, and Martin 1991). This indicates that each quantity decision may play a role in the determination of total consumption.

It seems that forecasts of consumption quantity will generally display upward bias, especially when limited product availability extends the forecast period. Several cognitive processes which may contribute to this effect are discussed below. This paper does not discuss the effect of limited availability upon preferences, and, therefore, upon optimal purchase quantity. The enhancing effect of limited availability upon preferences is important, but has been well documented elsewhere (e.g., "the rule of the few," in Cialdini 1985). Instead, this paper concerns the effects of scarcity upon consumption forecasts holding preferences constant.

There is anecdotal evidence for the hypothesis that limited availability results in higher forecasts of consumption quantity. The Wall Street Journal reports that Warner-Lambert Co. enjoyed an impressive sales increase in three flavors of gum once the company limited their availability to two months a year. Sales rose from $3.2 million in 1986 to $10.5 million in 1988. Pam Goyette, a Warner-Lambert senior product manager, attributes this success to the fact that consumers "really load up" on the gums as a result of their limited availability (Alsop 1989, p.B1).

There is also intuitive appeal to the notion that limited availability will produce overestimation of one's optimal consumption amount. In general, it seems that people overestimate when consumption forecasts are necessary. The cognitive processes and biases which are investigated below seem to produce a behavioral theme of unfounded optimism concerning one's enjoyment of any item under consideration. Further, this optimism seems to be accentuated when forecasts of optimal quantity must be made in advance. We tend to take more food than we can possibly eat when later access to the buffet table is limited. We consistently forget that we will be tired after the first week of vacation. We do not realize that we will not wear the clearance shoes that we buy in three colors.

Implications resulting from this work will generalize to two classes of marketing situation. First, this analysis is directly relevant to the case of limited product availability. Some consumer items are routinely offered on a restricted basis. For example, many clothing products are available for only one season. Food items, such as beer and candy, are commonly offered seasonally or in limited editions. Many collectibles are offered "for a limited time only." Exclusive distribution restricts a good's availability for consumers who live far from a store which carries the good. Tourist items are of limited availability to the traveler. Any insights as to how consumers make quantity decisions under limited availability should aid marketers of the above product classes. Likewise, this theoretical development provides a rationale for the somewhat counter-intuitive practice of making one's products difficult to obtain.

The second, and more common, case to which these ideas apply is that of sale items. Most products are constantly available, but are periodically available at a sale price. Sales may induce customers to estimate optimal consumption levels over relatively long time periods. In fact, some consumers refrain from buying certain items unless those items are on sale. Thus, periodic sales may cause the consumer to behave as if an item is not continually available. The ideas in this paper may, therefore, be valuable to marketers who use price discounts as a promotional tool. This paper provides a rationale, beyond inducement of trial or switching, for the offering of periodic price discounts.

What follows is a discussion of several processes which may contribute to bias in the forecasting of consumption quantities. The bulk of this paper discusses a process by which difficulty in forecasting preferences may cause biased quantity forecasts. Additional, complementary processes and biases are also discussed. The major goal of this paper is to apply knowledge concerning cognitive functioning to a relatively ignored problem in consumer research.


Forecasting Preference

Previous Research. It seems that forecasting one's future preferences is a difficult task. March (1978) theorizes that future preferences are fuzzy and capricious. Therefore, he states, decisions concerning future outcomes must reflect guesses, not certainty, about one's preferences for these outcomes. Furthermore, he hypothesizes that current theories of choice, which do not acknowledge irregularities in preference, are inadequate for the description of human decision making.

Kahneman and his colleagues (Kahneman and Snell 1990; Varey and Kahneman In press) offer experimental evidence indicating that individuals cannot accurately predict their own preferences. For instance, subjects in Kahneman and Snell's (1990) experiments were asked to predict, after an initial sampling, their own preference fluctuations during repeated exposures, over several successive days, to food and music. Correlations between actual and predicted changes in tastes were minimal, even for this relatively simple preference forecasting task.

Further, subjects do not consistently incorporate information about fluctuations in the quality of a class of experiences into decisions concerning these experiences. Kahneman and his colleagues demonstrate that subjects choose risk seeking alternatives when deciding among probabilistic strings of increasingly unpleasant experiences. Risk seeking decisions are inconsistent with the increasing marginal disutility reported by the majority of these same subjects. For instance, subjects did not consistently prefer a probabilistic (in the number of treatments necessary) medical treatment which offered a reduction from ten to eight treatments for one outcome in exchange for an (equally probable) increase from two to four treatments for another outcome. However, the same subjects indicated that treatment side effects would worsen with time, implying that the reduction from ten to eight is more "valuable" than the reduction from four to two (Kahneman and Snell 1990; Varey and Kahneman In press). Varey and Kahneman (In press) conclude that "responses to questions eliciting intuitions about global utilities [for a series of experiences] are unaffected by people's beliefs about the changes in the experience over time" (p.19).

Why might people be insensitive to trends in a series of hedonic experiences? Kahneman and Snell (1990) hypothesize that people commonly fail to conduct separate evaluations of each experience. They propose that individuals simplify decisions by using an "instantaneous representation" to evaluate temporally separated sets of outcomes. An instantaneous representation is simply a projection of the relevant time period onto a single moment. Consultation of the instantaneous representation results in neglect of the unique aspects of each experience. For example, instead of integrating the predicted utility of each of a set of experiences, people may use "news utility" as a simplification allowing evaluation of the set. News utility consists of the individual's prediction of his own emotional reaction to learning that he will undergo the experiences in question (Kahneman and Snell 1990). To foreshadow, it seems that news utility will be evaluated against current preferences. For instance, it seems that the news that one will consume three chocolate bars in the future will be less favorably evaluated if one has just consumed two.

Preference Forecasting. The above theoretical development may be directly applied to the quantity forecasting process. To make optimal quantity decisions, the consumer must consider the hedonic quality of several, temporally separated, consumption experiences. Each consumption experience will likely be unique due to preference fluctuations. The empirical work discussed above indicates that the consumer will be unable to accurately incorporate these preference fluctuations into her decision. Instead, she may collapse the forecast period and use one measure of preference. This will most likely be current preference since evidence suggests that people cannot estimate future preferences very well. In other words, current preference may be used to represent preferences throughout the forecast period.

Quantity forecasts, therefore, will be biased to the extent that current preference is not representative of future preference. Hence, one must investigate the likely nature of preference fluctuations in order to determine the nature of the bias, if any, induced by reliance upon current preferences. McAlister (1982) offers a useful model of preference fluctuation. She proposes that individuals hold inventories of recently consumed attributes (such as "fruit flavor") and that preference is inversely related to inventory levels. In her model, attribute inventories continually deplete, with individual- and attribute-specific parameters determining the depletion rate. Preference is at its highest when attribute inventories are at their lowest. Preferences continually fluctuate, lowering with consumption and then gradually rising. Satiation, or the degree to which a preference deviates from its highest level, is included in current preferences to the extent that the relevant attributes have been recently consumed.

The work of Kahneman and his colleagues (Kahneman and Snell 1990; Varey and Kahneman In press) suggests that satiation will only affect the preference forecast process to the extent that current preferences are moderated by satiation. Absence of recent consumption, and therefore absence of satiation, is almost guaranteed at the beginning of a restricted availability period. Obviously, an individual cannot consume a good which is not available. Preferences throughout the forecast period will, however, be moderated by consumption. Therefore, with limited availability, preferences during purchase are very likely to be higher than average preferences over the forecast period. If these abnormally high preferences are used as a proxy for preferences during the forecast period, upwardly biased quantity forecasts should result.

The likely level of preferences at the time of purchase is less certain with sale items. However, if the individual buys more than he normally would because an item is on sale, and consumption elevates as a result, preferences over the period should be more moderated by satiation than are preferences at the time of purchase. This implies that the manager should allow time for inventories (real and, hence, attribute inventories) to become depleted between sales. Further, it implies that a strategy of restricting availability will work best for distinctive goods. Otherwise, attribute inventories may be maintained by consumption of similar goods.

In general, abnormally high preferences at the beginning of an availability period should result in overestimation of the average hedonic quality of consumption of the good in question. Thus, the following is proposed:

P1a: Consumers will rely on current preference as a proxy for preferences throughout the consumption period when determining purchase quantity.

P1b: Forecasts of optimal consumption quantities will become upwardly biased as the time horizon of forecast is extended.

The above analysis implies that the individual generally cannot adequately forecast satiation or incorporate anticipated satiation into his forecasts. This hypothesis follows from the empirical work of Kahneman, Snell, and Varey. Of course, this does not imply that satiation can never be anticipated. An individual may eventually learn to recognize that she typically overestimates preference for a certain item; this seems most likely with frequently purchased items. However, the difficulty individuals display in predicting preferences should preclude the general finding of accuracy in predicting satiation. This is, of course, an empirical question:

P2a: Individuals will not adequately consider satiation when making consumption forecasts.

Although it may be impossible for the individual to incorporate satiation into her quantity forecast, manipulation of attribute inventory levels should alter preferences, affecting quantity forecasts:

P2b: Forecasted quantity will decrease with recent consumption of the item under consideration.

The Role of Certainty. Subjective certainty in one's current preference may contribute to bias in quantity forecasts. Certainty is not typically correlated with accuracy in decision making, with overconfidence being the norm (e.g., Lichtenstein, Fischhoff, and Phillips 1982). Recent evidence suggests that overconfidence is caused by people's tendencies to overweight the strength or extremeness of evidence relative to its weight or credence (Tversky and Griffin In press). Individuals tend towards overconfidence in decisions when the evidence upon which they are deciding has high strength but low weight. For instance, an extreme proportion (high strength) based upon a small sample (low weight) may produce overconfidence. Extreme preferences are high in strength but low in weight; they are noticeable, like the extreme proportion, but unreliable, like the small sample. Thus, a high preference level may actually cause overconfidence. Subjective certainty may, therefore, operate to exacerbate bias from reliance upon abnormally high current preferences. (Of course, this overconfidence may also operate on abnormally low preferences, but in this case the individual will simply not buy the good in question.)

P3a: Individuals should show a significant, positive correlation between preference level and certainty.

P3b: Individuals should show an insignificant correlation between certainty and accuracy when forecasting quantities.

Thus far, we have discussed cognitive processes which may cause individuals to inadvertently buy more than their optimal amount of a good. Other cognitive mechanisms may cause individuals to intentionally buy more than they believe is optimal. This possibility is discussed below.

Purchases Above Forecasted Optimality: Safety Stock

One could conceptualize the task of forecasting optimal quantity as an inventory problem. A possible contributor to large quantity forecasts, evident from the inventory framework, is the purchase of excess inventory or safety stock. These are simply items purchased, in addition to the quantity forecasted as optimal, in order to lower the risk of a stockout. Both normative solutions to inventory problems (i.e., the newsboy problem) and intuition dictate that safety stock depends upon the costs of overestimation and underestimation of optimal quantity. The individual should buy exactly the forecasted amount if both costs are equal. As underestimation of optimal quantity becomes relatively more costly, the individual should purchase increasing amounts of safety stock.

The human value function is such that the perceived cost of underestimating consumption may consistently dominate that of overestimating. People tend to value the disutility associated with goods of which they are deprived more highly (i.e., against a steeper value function) than they value the disutility of forfeiting the purchase price of those same goods. In general, goods of which one is deprived are valued as a loss while money spent on goods is not, and value functions are typically steeper for losses than for gains (e.g., Tversky and Kahneman In press). Even anticipated consumption is thought to shift the individual's reference point, so that not having an anticipated item will be valued as a loss (e.g., Hoch and Loewenstein 1990). Thus, people may evaluate the cost of having one too few of the item under consideration by a steeper disutility function than that by which they will evaluate the cost of buying one unnecessary item. If so, the cost of underestimating consumption will routinely dominate the cost of overestimating consumption. Hence, the consumer will be prone to consistently buy at a level above his forecasted optimal quantity.

Forecasting Consumption Opportunities

This section will briefly discuss a second estimate which must be made before purchase quantity can be determined. Specifically, the individual must forecast the number of opportunities he will have to use the product class in question. Note that this forecast is very likely made simultaneously with one's preference forecast; the processes are separated for clarity. The general form of the heuristic used to forecast consumption opportunities may influence the amount of bias in one's final quantity forecast. An estimate of consumption opportunity must, in general, be based upon either a rate or a whole number. When using a rate-based strategy, the individual may be less likely to consider satiation because the focus is upon "one-at-a-time" (i.e., once a day) consumption. Therefore, bias should generally be at its worst when this form of heuristic is utilized. Alternatively, a holistic (i.e., based upon determination of a total number of opportunities) heuristic seems likely to cause the decision maker to focus on a large amount (i.e., 30 items monthly) of consumption. Therefore, satiation seems more likely to become salient. Use of a holistic heuristic may therefore lead to moderation of the proposed bias.

P4: Forecasted consumption quantity will be larger with a rate-based heuristic than with a holistic heuristic.

It would be helpful to determine factors influencing the type of forecasting heuristic which is used. Some related research, involving estimates of past behavior, exists (e.g., Blair and Burton 1987). For instance, there is some suggestion that a (holistic) "recall and count" strategy is less likely, and a rate-based strategy is more likely, as the estimation period is extended (Blair and Burton 1987). Thus, rate-based strategies may be more likely once the forecast period is extended. This may be yet another contributor to the expected upward bias in quantity forecasts with lengthened forecast periods.

However, several factors unrelated to the length of the forecast period may influence what sort of heuristic is adopted. The heuristic may be influenced by a firm. For instance, some forms of packaging (i.e., soda in separate cans) may promote a rate-based heuristic, while others (i.e., two liter bottles) may promote a holistic heuristic. Other factors, such as experience with the product class, may influence which heuristic is chosen and may also directly influence the magnitude of the bias observed. For instance, experienced users of a product class may be more likely to use a rate-based heuristic, but, due to their experience, they may also be less likely to arrive upon a biased forecast. Thus, the relationship between rate-based heuristics and forecasting bias may be attenuated by the effects of factors such as experience.


This paper begins investigation of a problem which has been neglected by consumer researchers, namely the manner in which a consumer forecasts optimal purchase quantity for a repeatedly consumed good. This paper has further focused upon the special case of limited product availability. However, some of the factors discussed above may impact quantity forecasts of goods which are of unlimited availability. For instance, individuals may have a general propensity to purchase safety stock. Thus, the process description offered above may apply to more than just this special case.

Research promoting understanding of the forecast process could have detrimental as well as beneficial effects on the consumer. It almost certainly is just as easy to "help" the consumer add bias to his estimates as it is to help the consumer remove this bias. However, future research efforts may uncover or develop techniques for avoiding overestimation of purchase quantities. For example, consumers may reduce bias in their forecasting decisions by variety seeking. It may simply be easier to generally restrict oneself from buying too much of one thing than it is to make accurate forecasting decisions. An individual may learn to curtail his tendency to overestimate optimal quantity by developing variety seeking rules.

Finally, note that the specific processes and events described above may operate in a complementary- or a contradictory- manner. Returning to our Warner-Lambert gum example, several processes may have combined to create the observed sales increase. Consumers' preferences may become enhanced with limited availability; perhaps the gum is now seen as a novelty item. Deficits in forecasting satiation with the gums' distinctive flavors may occur, leading to upwardly biased forecasts of preference over the forecast period. The cost of underestimation of consumption is likely to dominate that of overestimation, so that individuals purchase safety stock. Finally, individuals may tend to overestimate the possible opportunities for chewing gum. For instance, one may use a rate-based heuristic and forget to adjust downward for occasions when gum chewing is not appropriate.


Alsop, Ronald (1989), "Gums' Revival Proves Popular and Profitable," Wall Street Journal February 21, B1.

Blair, Edward, and Scott Burton (1987), "Cognitive Processes Used by Survey Respondents to Answer Behavioral Frequency Questions," Journal of Consumer Research, 14 (September), 280-288.

Cialdini, Robert B. (1988), Influence: Science and Practice, Boston: Scott, Foresman, and Company.

Folkes, Valerie, Kamal Gupta, and Ingrid Martin (1991), "When to Say When: Effects of Supply on Usage," working paper, University of Southern California, Los Angeles, CA 90089-1421.

Hoch, Stephen J. and George F. Loewenstein (1991), "Time-Inconsistent Preferences and Consumer Self Control," Journal of Consumer Research, 17 (March), 492-507.

Kahneman, Daniel and Jackie Snell (1990), "Predicting Utility," in Insights in Decision Making: A Tribute to Hillel J. Einhorn, ed. Robin Hogarth, Chicago: University of Chicago Press.

Lichtenstein, Sarah, Baruch Fischhoff, and Lawrence Phillips (1982), "Calibration of Probabilities: State of the Art to 1980," in Judgement Under Uncertainty: Heuristics and Biases. eds. D. Kahneman, P. Slovic, and A. Tversky, New York: Cambridge University Press.

March, James G. (1978), "Bounded Rationality, Ambiguity, and the Engineering of Choice," Bell Journal of Economics, 9, 587-608.

McAlister, Leigh (1982), "A Dynamic Attribute Model of Variety-Seeking Behavior," Journal of Consumer Research, 9 (September), 141-150.

Tversky, Amos, and Daniel Kahneman (In press), "Loss Aversion is Riskless Choice," Quarterly Journal of Economics.

Tversky, Amos, and Dale Griffin (In press), "The Weighting of Evidence and the Determinants of Confidence," Cognitive Psychology.

Varey, Carol and Daniel Kahneman (In Press), "Experiences Extended Across Time: Evaluation of Moments and Episodes," Journal of Behavioral Decision Making.