Loss aversion for quality in consumer choice.

Abstract:

A reference price is an internal price that consumers are believed to use to compare actual prices. Reference effects for price have been demonstrated in many settings. Reference effects for quality also have been demonstrated using scanner data. Here we present experimental evidence. First[y, it is shown that high quality goods will be valued more by consumers who consider trading down in quality than by those who consider trading up in quality. Secondly, we show that when all prices fall, more switching up in quality from the reference brand will occur than switching down in quality when all prices rise, and that when all prices fall, consumers will switch to higher quality up to, but not beyond, the price regularly paid.

Keywords: LOSS A VERSION; CONSUMRE CHOICE; ASYMMETRIC COMPETITION.

1. Introduction

A reference price is an internal price that consumers are believed to use to compare to actual prices (Lilien, Kotler & Moorthy 1992). In making pricing decisions, reference prices as well as actual prices should be considered--assuming reference prices are important. Reference effects for price have been demonstrated experimentally in many settings (e.g. Kalyanaram & Little 1994; Kalwani & Yim 1992; Mayhew & Winer 1992), while reference effects for quality were demonstrated by Hardie, Johnson, and Fader (1993) with scanner data from the refrigerated orange juice market. In this paper, we present experimental evidence of reference effects for quality. Furthermore, we show the impact of reference effects on global price changes, which leads to new insights for mental accounting in consumer choice.

>From a managerial perspective, understanding a phenomenon at the individual level can be invaluable when developing strategies to influence consumer response. Winer, Bucklin, Deighton, Erdem, Fader, Inman, Katahira, Lemon, and Mitchell (1994) suggested the use of experiments to complement panel-data findings to assure internal and external validity.

Hardie, Johnson and Fader (1993, hereafter HJF) inferred the existence of reference effects based on the relative size of model coefficients for gains and losses. However, they did not provide a direct test for the presence of reference effects. Typically, reference effects have been demonstrated in controlled experimental settings rather than inferred from correlational data. In this paper, we report the results of two experiments that: (1) support the existence of loss aversion for quality in a controlled choice setting with real products; and (2) show that both the reference price (or reference budget) and reference level of quality have different effects on choice in the face of changing prices.

In section two, explanations for reference effects and asymmetric price competition are reviewed. In section three, assuming the existence of asymmetric price responses at the market level, we present two experiments. The first demonstrates loss aversion for quality and we explain how these data can lead to asymmetric price responses. The second experiment, rather than shifting one product's price, globally shifts prices for an entire product category. This experiment provides further evidence for loss aversion for quality and the existence of non-fungible budgets for individual product categories. The paper concludes with a general discussion and suggestions for future research.

2. Reference Effects and Asymmetric Pricing

2.1 Reference Effects

In the marketing literature, the internal reference price has been used as an explanatory variable in models of brand choice, estimated with scanner data. Typically, a reference price is assumed to exist and is based either on the last price paid or some average of recent prices. Psychological concepts are frequently invoked to explain the existence of reference prices and related phenomena. For example, Weber's Law can be applied to the perception of price changes. Assimilation-contrast theory (Sherif 1963) suggests a 'latitude of acceptance' around a reference price, much like Gabor and Granger's (1964) notion of acceptable price range, in which consumers are thought to be relatively insensitive to price changes close to a good's normal price. Adaptation level theory (Helson 1948) has been used to explain the existence of internal reference prices as a function of previously observed prices, although it does not address issues of stimulus pattern. The psychological theory that may have had the most impact on recent work in pricing is prospect theory (Kahneman & Tversky 1979), which stresses the importance of reference points and the direction of change from the referent in the evaluation of outcomes. In addition, losses from the reference point loom larger than gains of the same amount; a phenomenon known as loss aversion.

The translation of loss aversion into the domain of pricing is straightforward--price increases will hurt sales more than a price decrease of the same amount helps. Kalwani, Yim, Rinne, and Sugita (1990), Mayhew and Winer (1992), and Kalyanaram and Little (1994) are among the researchers who have found that modelling the effects of price increases and decreases separately improves the fit of brand choice models. In an extension of classic economic theory, Putler (1992) found that price increases affected demand for shell eggs significantly more than price decreases.

While prospect theory was originally limited to risky choices involving a single dimension (money), an updated version known as reference theory applies in riskless settings, with alternatives comprising values on multiple dimensions (Tversky & Kahneman 1991). Its key features are: (1) reference dependence; the idea that value is judged in terms of gains and losses from a multidimensional reference point; (2) loss aversion; and (3) diminishing sensitivity; the notion that the marginal value of both gains and losses diminishes with their size. The utility of an alternative comprises separate monotonic functions on each attribute, which are combined within another monotonic function. Tversky and Kahneman developed the particular case of an additive reference structure, which assumes a constant degree of loss aversion for each attribute.

Tversky and Kahneman reported data from experiments involving hypothetical choices that were consistent with loss aversion for non-monetary dimensions. Hitherto, demonstrations of loss aversion in controlled real choice settings have been limited to uni-dimensional reference points. One contribution of this paper is to demonstrate loss aversion for multiple dimensions in controlled choice setting with real products.

HJF's model of brand choice was based on Tversky and Kahneman's (1991) reference theory. In their model, the last brand purchased served as a multidimensional reference point against which alternatives were evaluated. Consumers were assumed to be loss averse for multiple dimensions, for example, price and quality. Loss aversion for money has been supported in previous reference price models that separated the effects of price increases and decreases (Kalyanaram & Little 1994; Kalwani et al. 1990; Mayhew & Winer 1992; Putler 1992).

HJF added a reference point for quality and found support for loss aversion for quality as demonstrated by the relative size of the coefficients of loss aversion for quality and money. They modelled scanner data from the refrigerated orange juice market. In this model they used the Consumer Reports' rating of the brand last purchased as a proxy for each household's quality reference point. Of the several models they tested, the best fit was given by the model that incorporated dual reference points for price and quality.

2.2 Explanations of Asymmetric Price Competition

Blattberg and Wisniewski (1989) explained asymmetric competition between goods in different price tiers by invoking a bimodal distribution of consumers' relative preferences for national brands and private label brands. Consumers were assumed to be indifferent between goods when the utility of the price difference equalled the utility of the perceived quality difference, and brand switching was assumed to occur when a price shift changed the magnitude of the price difference between brands. Their model fits actual sales histories for five product-categories quite accurately, although the true distribution of preferences in the population was unknown.

Allenby and Rossi's (1991) explanation of asymmetric brand switching was based on a combination of income and substitution effects, given the assumption of non-homothetic preferences. They argued that when the price of a high quality brand dropped, both an income effect, which put the consumer at a higher level of utility, and a substitution effect, in which the good became cheaper relative to other goods, work in favour of a switch up to a higher quality brand. The income from the price cut was assumed to result in a change in preferences for product quality. On the other hand, when the price of a lower quality brand dropped, the substitution effect induced a switch down to a lower quality brand, while the income effect induced a switch up; the net result was very little switching. Allenby and Rossi found support for their model with data from the margarine market.

A reference effect explanation of asymmetric price competition suggests that price changes will affect buyers of national brands and buyers of private label goods in different ways. Regular purchasers of a product will adapt to the product's price and quality level. Thus, assuming a positive correlation between price and perceived quality, national brand buyers will develop a high quality, high price reference point, whereas private label buyers will develop a lower quality, lower price reference point. Because regular purchasers of products in different price tiers develop distinct reference points for product attributes, price changes affect them differently. Throughout this study we refer to the regularly purchased good as the reference brand, and assume that it comprises a reference point for both perceived quality and price.

3. Experiments

3.1 Study 1, Loss Aversion for Quality

Given the existence of asymmetric price responses at the market level, the purpose of this experiment is to determine whether individual level data are consistent with the reference effect model. As previously noted, loss aversion in the domain of money has been demonstrated in numerous studies. The studies that were most relevant to the current one are 'endowment effect' experiments, in which a discrepancy between buying and selling prices resulted from different endowments; a violation of the Coase theorem (Kahneman, Knetsch & Thaler 1990). In these experiments, one group of participants called 'sellers' was given a good of fairly low retail value, such as a mug, and asked for the price they would be willing to return it. A second group called 'choosers'--seated among the members of the first group--was asked to specify the price (money) they would prefer to receive rather than the good. A third group called 'buyers' was asked how much they would be willing to pay to obtain the good. Sellers typically valued the mug about twice as much as the choosers and buyers respectively. It is important to note the only difference between the groups was the initial endowment of the good. The effect was quite robust, and appeared to be virtually instantaneous. Follow-up research (Loewenstein & Kahneman 1991) showed that subjects in the role of sellers did not find the object more attractive than those in the role of choosers; they simply disliked giving it up. In this particular study, subjects rated the attractiveness of each good in a set, which included the item with which they were endowed. Half the subjects were endowed with a pen and half with a travel mug. Attractiveness ratings did not differ significantly between groups, but when offered the opportunity to trade, far fewer subjects traded than would have been predicted by the neoclassical economic model of choice.

If subjects are loss averse for quality, they should demand more compensation to accept a decrement in quality when endowed with a high quality good than they would be willing to pay to acquire the same increment in quality, as measured by willingness-to-accept and willingness-to-pay measures of value. Choice prices serve as a diagnostic to establish whether any discrepancy was due to loss aversion for quality or loss aversion for money. We show this graphically in figure 1.

Following HJF and Tversky and Kahneman (1991), the graph is constructed in a two-attribute space (price and quality), where price is increasing from right to left. (1) Figure 1 depicts the choices faced by two groups of participants, one group called 'traders-down' and another called 'traders-up'. The first group is endowed with a high-quality product, H, while the latter group is endowed with a low-quality product, L. I[D.sub.2] and I[D.sub.2] refer to the indifference curves of the traders-down, while I[U.sub.1] and I[U.sub.2] refer to the indifference curves of the traders-up.

Suppose now that the trader-down is given a choice between keeping good H, or trading it in for good L, plus some cash amount. What would that cash amount, or trading-down price, be to make her indifferent between the two choices? First note that the product L sits on a lower indifference curve because at the regular prices product H is preferred to L. To make the trader-down indifferent between the two products, the low-quality product L must sit on the same indifference curve as the high-quality product by decreasing in price.

The price-quality bundle at point B indicates the price of the low-quality product that makes the trader-down indifferent between both products. Letting Pa be the trading-down price, the graph shows that:

(1) [P.sub.d] = [P.sub.H] - [P.sub.L]'

where

(1a) [P.sub.L]' = [P.sub.L] - [eta], and [eta]>0 is some discount amount.

Now suppose that the trader-up is given a choice between keeping good L, or paying some amount in exchange for product H. What cash amount would she be willing to pay to obtain the higher quality good H, or alternatively, what is the trading-up price? In the above graph, the trading-up price, [P.sub.u] is:

(2) [P.sub.u] = [P.sub.H]' - [P.sub.L]

where

(2a) [P.sub.H]' = [P.sub.H] [zeta] and [zeta]>0 is some discount amount.

Now the original price difference, [P.sub.o] between goods H and L is:

(3) [P.sub.o] = [P.sub.H] - [P.sub.L]

>From (1a) we know that [P.sub.L] > [P.sub.L]' which implies that [P.sub.d] > [P.sub.o]. Similarly from (2a) [P.sub.H] > [P.sub.H]' then implies that [P.sub.o] > [P.sub.o]. Therefore [P.sub.d] > [P.sub.u]; in other words the trading-down price is greater than the trading-up price. This reasoning leads to the following hypothesis:

H1: High quality goods will be valued more by consumers who consider trading down in quality than by those who consider trading up in quality, as measured by willingness-to-accept and willingness-to-pay measures of value (i.e. the 'selling' and 'buying' prices in an endowment effect paradigm).

3.1.1 Method and Procedure Seventy-six psychology undergraduate students at a large university participated in this study. Participants were run in small groups and received course credit for their participation. As in studies of endowment effects, participants were assigned to one of three roles--'traders-up', 'choosers' or 'traders-down'.

Chocolate was chosen as the good for this experiment because distinct quality levels existed within the product category and undergraduates have experience in buying it. In addition, because the choice situation was real rather than hypothetical, a fairly inexpensive good was needed. Toblerone Chocolate was the high quality good. It was fairly well known, had a high quality reputation, and commanded a premium price of $1.79 for a 100-gram bar. (2) The low quality chocolate was 'Chocolaty', a chocolate-flavoured bar obtained at an inexpensive variety store. The packaging on this brand looked cheap and the label clearly stated that it was chocolate-flavoured, rather than real chocolate. No price information was on the label of either chocolate bar. Participants who asked about price were told that this question would be answered at the end of the experiment.

To avoid social comparison, only one condition was run within a group. Traders-down received a Toblerone bar and were asked to indicate at what price, [P.sub.d], they would be willing to return their Toblerone bar in exchange for a Chocolaty bar plus that amount in cash. Choosers were informed that they had a choice between a Toblerone bar and a Chocolaty bar plus an amount of money. Choosers were to indicate the amount of money that made them indifferent between the two alternatives, this amount being the choice price, [P.sub.c]. Traders-up were given a Chocolaty bar and asked how much they would be willing to pay to exchange their Chocolaty bar for a Toblerone bar; in other words, they were asked for their trading-up price, [P.sub.c]. In all cases, subjects indicated their switching prices on a form listing amounts in $0.10 intervals from $2.50 to $0.00. To emphasise the importance of indicating their true values for the alternatives and to eliminate strategic trading behaviour, subjects were told that one of the amounts of money listed would be selected and the option they had chosen for that amount would be carried out. They were informed that the amount had already been determined randomly, and that their responses would have no effect on the amount picked. This incentive-compatible procedure eliminated strategic reasons to overstate trading-down prices or understate trading-up prices. In order to avoid income effects, we did not endow subjects with money; a standard procedure in endowment effect studies.

3.1.2 Results and Discussion Median prices were computed for each condition and were as follows: traders-down price, [P.sub.d] = $1.00, choosers' price, [P.sub.c] = $0.70, and traders-up price, [P.sub.u] = $0.50. The pattern of results indicates a 2:1 discrepancy between trading-down and trading-up prices, with choosers' price closer to the traders-up price than to the traders-down price, replicating the pattern found in standard endowment effect experiments for quality. Table 1 reports medians and means for all three conditions.

A large-sample Mann-Whitney test yielded a standard normal z = 0.92, (p = 0.18), indicating that medians for the traders-up and choosers were not significantly different. Traders-down, however, were significantly different from both traders-up, z = 3.82 (p < 0.005), and choosers, z = 2.85 (p < 0.005). These results indicate that, as measured by trading prices, loss aversion for product quality exists.

To demonstrate the presence of loss aversion for quality, we invoke Tversky and Kahneman's (1991) additive reference structure where homogeneity in quality perception is assumed. In this model, [[lambda].sub.p] denotes loss aversion for money, while [[lambda].sub.q] denotes loss aversion for product quality. At the margin, the trading down and trading up prices are given by equations (4) and (5) respectively.

(4) [P.sub.d] - [[lambda].sub.q] Q = 0

(5) Q - [[lambda].sub.p] [P.sub.u] = 0

In the above, Q was the perceived quality difference between the high-quality and low-quality products. In other words, Q = [Q.sub.H] - [Q.sub.L]. Substituting (4) into (5), yields the relationship between trading prices as:

(6) [[lambda].sub.p] [[lambda].sub.q] [P.sub.u] = [P.sub.d]

Recall that the median trading price results indicated that the trading-down price was twice the trading-up price. This implies that:

(7) 2 [P.sub.u] = [P.sub.d]

Thus, using equations (6) and (7) we estimate the relationship between the loss aversion for money, [[lambda].sub.p], and the loss aversion for quality, [[lambda].sub.q] to be

(8) [[lambda].sub.p] [[lambda].sub.q] = 2.

It is not possible to determine the relative degree of loss aversion for money or quality from this single relationship. The key in determining whether loss aversion for quality or money was responsible for the discrepancy in trading prices was in the responses of the choosers. The choice price [P.sub.c] could be thought of as the value placed on the quality difference, that is [P.sub.c] = [Q.sub.H] - [Q.sub.L] = Q. Now, if there was no loss aversion for quality, then [[lambda].sub.q] = 1. Consequently, equation (8) gives [[lambda].sub.p] = 2. If this is true, equations (4) and (5) become:

(4a) [P.sub.d] > [P.sub.c],

and

(5a) [P.sub.c] 2 [P.sub.u].

Thus, in the absence of loss aversion for quality, the choice price would be equal to (or close to) the trading-down price, and would be double the trading-up price. However, our experiment revealed that the median choice price was not significantly different from the median trading-up price but was significantly different from the trading-down price; that is:

(9) [P.sub.d] > [P.sub.c]

and

(10) [P.sub.c] [approximately equal to] [P.sub.u].

The inescapable conclusion is that loss aversion for money alone cannot explain our data; however, loss aversion for quality explains it well. We do not take the position that loss aversion for money has no role in asymmetric switching; rather, we argue that these experimental data support the existence of loss aversion for product quality.

Another possibility would be that the subjects were simply showing the effects of status quo bias. That is, both groups of subjects could have been averse to giving up the chocolate they already had, irrespective of quality. This was not the case however, because all but one of the 26 subjects endowed with Chocolaty indicated that he or she would switch to Toblerone even it no money was involved. If a strong status quo bias was operating, these subjects would not have indicated a willingness to switch. The status quo explanation was further refuted because the subjects endowed with Toblerone did not face the loss of chocolate--only the loss of high quality chocolate.

To address the question of whether the trading-down versus trading-up discrepancy observed in the data was due to loss aversion for product quality and not related to other differences between the products, such as packaging, a separate group of participants was shown the two chocolate bars and asked to generate a single word which summarised the differences between them, aside from price. The modal response was the word 'quality,' which was generated by 22 of 39 subjects.

At this point we note that both Blattberg and Wisniewski's price tier theory and Allenby and Rossi's model predicted no difference between the trading-up, choice, and trading-down prices. According to price tier theory, the price difference between goods is the critical factor in switching between quality tiers. Therefore, the price difference that would lead to switching in our experiment should be the same across all three conditions. Allenby and Rossi's model was based on a combination of income effects and substitution effects. In our study, the alternatives were the same: Toblerone chocolate or Chocolaty plus money. There was nothing across the three conditions to induce differential income or substitution effects. Thus, the reference effects model is the only model consistent with both the market-wide data and the individual-level data.

Demonstrations of loss aversion in experimental settings with real choices have involved small consumer goods where a uni-dimensional trade-off between the good and money was assumed. In this experiment, we demonstrated loss aversion for a non-monetary dimension with a real choice between two goods in the same product category, distinguished by a difference in perceived quality.

Our study has some limitations that might decrease its generalisability. First, we used students as participants. We tried to mitigate the impact of the participant pool by using a product they were familiar with and making the choices real. Second, because quality was a subjective attribute, measuring the exact magnitude of quality differences with the stimuli we used was impossible. However, using stimuli with quantifiable attributes as a proxy for quality (such as per cent premium components of expert ratings) might make magnitude of quality differences even more salient. Finally, for both time and budgetary reasons this experiment was performed with only one good and we do not report interviews with the subjects. It would be useful if this study were followed up with more goods and, possibly, interviews in a natural purchasing environment. However, these results, when considered with those of other researchers such as HJF, provide converging evidence for the importance of multi-dimensional reference points in consumer choice.

3.2 Study 2: Global Price Changes and Mental Accounting

The objectives of this experiment are to explore reference effects in response to global price changes as well as to provide further investigation of the three theories of asymmetric price response. Global price changes occur in situations where there is inflation or deflation, tax changes or simply a change of venue (e.g. a visit to an expensive resort or a developing country). An interesting aspect of global prices changes is that they allow us to examine whether consumers have non-fungible budgets for narrow product categories.

If a regularly purchased good serves as a reference point for the evaluation of alternatives, the balance of gains and losses will differ in the face of a percentage change in all prices, depending on the direction of the change. The options available to the consumer are illustrated in figure 2. When all prices rail, the consumer can maintain the current level of quality and keep the difference between the regular and sale price as a subsidy (position A3). Alternatively, a higher level of quality than usual can be obtained by spending the usual reference price or budget (position A4). Finally, some combination of cost savings and increase in quality can also be obtained.

However, when all prices increase, the situation changes to a negative choice of a loss of money beyond the price usually paid (by maintaining the current level of quality at position A1), a loss of quality from the level usually consumed (and assuming the usual amount spent at position A2) or some combination of an increase in price and a decrease in quality.

In this experiment, we examined the effects of multidimensional reference points on choice in the face of changing prices. The first study suggested that both loss aversion for money and loss aversion for quality played a role in asymmetric responses. Here, we examine reference effects on brand switching over a broader range of prices and qualities than in the previous study.

The specific issue we examined in this study was whether a dual reference point for price and quality would lead to different switching patterns when prices rose or fell together. In the absence of reference effects, we would expect consumers to simply pick the good that represented the best combination of quality and price, (changing brands if price changes led to changes in perceived value, irrespective of the direction of the price change). In product categories where price and quality co-vary, we would expect switches to lower-quality products when prices rise and switches to higher-quality products when prices fall, assuming consumers prefer a combination of low price and high quality.

A multi-dimensional reference point implies that switching will be asymmetric and 'sticky' in two instances. First, when prices rise, loss aversion for quality would inhibit switching to quality options below the reference point. Second, when prices fall, loss aversion for money would limit the extent to which quality is upgraded. The issue of when a cost becomes a loss is pertinent. In normal purchase situations, loss aversion for money does not play a role. The exchange of money for a good or service is the essence of a market transaction, and money is simply the cost of participation in the transaction. A cost begins to be treated as a loss when a well-established reference price is exceeded.

Following Thaler (1999), we view consumers as having mental budgets for various product categories and that these budgets are not fully fungible. Therefore, consumers are likely to spend savings within the particular budget category but not go beyond the reference price, which would lead to loss aversion for money. As noted above, loss aversion for money will arise only for expenditures beyond the reference budget. If consumers have a non-fungible budget, then they are likely to spend the savings on an increase in quality within the category. Thus, we should expect that the price paid for a good by subjects who switch up in quality would be very close to the price regularly paid. Again we note that Blattberg and Wisniewski (1989) would predict no switching in either case and Allenby and Rossi (1991) would predict switching only when prices fall. The above reasoning leads to the following hypotheses:

H2: When all prices fall, more switching up in quality from the reference brand will occur than switching down in quality when all prices rise, due to loss aversion for quality in the latter condition.

H3: When all prices fall, consumers will switch to higher quality up to, but not beyond the price regularly paid. That is, they would consume up to, but not beyond, their regular or reference budget.

3.2.1 Method and Procedure The approach we took was to choose a consumer good with which our participants were familiar, and which had several levels of correlated price-quality pairs. We then examined choices when the prices of all goods increase or decrease by 30%.

Participants were MBA students who participated in a classroom setting for course credit and students of various majors who participated in exchange for a $1.00 payment. Only those subjects who identified themselves as regular purchasers of beer were included. Beer was chosen because the category included a wide variety of brands with a positive price-quality relationship. Thirty-five participants identified themselves as regular purchasers of beer.

To establish a reference brand, participants were given a list of 25 different beers varying in quality, listed with the current retail price from a large grocery chain, from which they were asked to indicate the beer they were most likely to buy at the prices listed. They were asked to indicate their regular beer and its regular price if it was not listed. Participants were then asked to indicate their choices in two different scenarios. First, they were to imagine that they were in a specialty store where prices were 30% higher than regular grocery store prices, and that it would be terribly inconvenient for them to go to their regular store. In the other scenario, they were asked to imagine that their regular store was having a one-time promotion in which the prices of all beers were lowered by 30%. For both scenarios, participants were instructed to pick the one beer they were most likely to purchase. A percentage change in prices was used to avoid confounding the balance of gains and losses with a change in price ratios.

Participants were asked to rate the quality of each beer on a 1-9 scale, with a 'don't know' option for unfamiliar beers. Hall rated quality after their choices and half before, to check the possibility that choices might have affected ratings, and vice versa. Quality ratings for each beer were relatively consistent across the two groups, and the data were combined since no significant differences were found in quality ratings.

3.2.2 Results and Discussion In the lowered price condition, 28 of the 35 respondents switched to a beer that they had rated as higher quality than their regular beer. Of these, 25 switched to a beer which they considered to be in the highest quality category, and three switched to the second highest. These three respondents rated only one or two beers higher than the one to which they switched. Respondents who did not switch were already regular purchasers of beer that they rated at the highest quality level of those listed.

In the higher price condition, 19 of 35 respondents did not switch, while 16 switched to a cheaper beer. The different number of switchers in each condition supports hypothesis H2--more switching up in quality occurs when prices fall than switching down in quality when prices rise.

The mean price of the beer chosen in each condition was the primary measure of interest, because it demonstrated the importance of the reference level of expenditure. In the regular price condition, the mean price was $4.02; in the lowered price condition, $3.97; and in the raised price condition, $4.90. The difference between the regular and lowered price condition was not statistically significant, (t(68) = 0.18), thus supporting hypotheses H3. However, the difference between regular and raised price conditions was significant, (t(68) = 3.12, p < 0.01).

In the lowered price condition, respondents faced a positive choice between a gain of money and a gain of quality. Virtually all respondents who were not already at a ceiling for quality would spend the regular amount and upgrade quality rather than take a subsidy for their usual quality level. In the raised price condition, subjects faced a negative choice between a loss of money and a loss of quality. In this case, subjects split almost evenly between saving money and maintaining quality. Table 2 summarises the results of this experiment.

The difference between quality ratings of the beer chosen in each condition was calculated for each respondent. When prices were lowered, the mean quality difference across subjects between the regular beer and the beer chosen was 1.26 points on a 1-9 scale, a statistically significant difference (t(68)= 3.25, p < 0.01). When prices were raised, the mean quality rating was 0.40 less than the regular beer, which was not statistically significant (t(68) = 1.04). The first quality difference suggests a willingness to forego a monetary subsidy in order to improve quality, while the second quality difference suggests a reluctance to accept a decrement below the reference level of quality.

The question that arises here is why participants who preferred to protect money when prices increased chose not take a subsidy when prices decreased. It would be reasonable to assume that consumers who were generally price sensitive would prefer to save money and obtain the regular level of quality when prices fall, rather than switching up to higher quality. The answer suggested here is that because a reference amount has already been allocated to the purchase, there is no disutility in spending the full amount when prices fall--the purchase is still within the budget, however, when prices rise, participants are loss averse for both quality and money, and simply choose the option that hurts least.

Individual differences with respect to the relative weight given to money and product quality within the range of stimuli presented resulted in the split between subjects who switched down and those who stayed with the regular beer. Note that hall the respondents would pay more for their regular beer, indicating that the amount usually paid was below the reservation price.

It appears that subjects do have a 'budget' that they are willing to spend regardless of price decreases. Of course, this study is quite stylised--all prices rise or fall and there is only one product category. The latter limitation is of greater concern since there is no specific savings account. It would be interesting to see just how narrow and stable budget categories are when there are multiple possible expenditures, and savings is explicitly mentioned. It this phenomenon is generalisable it has strong implications for marketing as well as tax and savings policies.

4. General Discussion and Conclusions

A growing body of evidence indicates the importance of reference points in choice, whether through a simple bias to stay with the status quo (Samuelson & Zeckhauser 1988) or the more complex phenomenon of loss aversion (Kahneman, Knetsch & Thaler 1990; Simonson & Tversky 1992; Tversky & Kahneman 1991). One manifestation of loss aversion for quality in consumer choice was a reluctance to trade down from higher quality to lower quality goods, even at substantial monetary savings. In this study, loss aversion for quality was crucial in explaining asymmetric competition (HJF).

Other explanations offered for asymmetric competition make no allowances for past consumption in determining response to price promotions. It was not clear whether the change in preferences induced by a price change predicted by the Allenby and Rossi (1991) treatment would influence subsequent purchases by a household. If a price decrease of a high quality brand resulted in a preference for higher quality, would preferences revert to the original lower quality brands when prices returned to normal levels, or would the household continue to prefer the now more expensive higher quality brand to which it switched? In any event, neither of the alternative models were able to explain the behaviour of subjects in these experiments.

Reference point formation presented a difficult empirical problem. Internal reference prices are unobservable and, as yet, vaguely define& Operationalisations such as last price paid or exponential smoothing of the prices recently paid have been used with some success in models of brand choice (Kalyanaram & Little 1994; Mayhew & Winer 1992), but research into the temporal aspects of reference price formation is sorely needed. The issue of internal reference prices was further complicated by data that indicated that consumers often do not have good knowledge of prices (Dickson & Sawyer 1990; Urbany & Dickson 1991), although shoppers tended to be more accurate about the prices of goods they purchase frequently than about other prices. We suggest that consumers are likely to be aware of an ordering of prices for different goods, with the most accurate price knowledge about the most frequently purchased good. In the domain of frequently purchased goods, we believe that the natural reference price is the price routinely paid by the consumer for a good from the given category. When the consumer is brand loyal and there is little price fluctuation, the consumer may be extremely accurate in estimating prices. However, if the consumer switches frequently or prices are highly volatile, the reference price is likely to be a distributed across a range. HJF used the last brand purchased as the reference brand. While this measure is certainly appropriate for brand loyal consumers, it is questionable for consumers who switch brands frequently. We suggest that the brand most frequently purchased over several occasions may be a more reliable measure.

A problem presented by the notion of loss aversion for quality was that in some cases consumers could find themselves trading up to higher and higher levels of quality as the result of price promotions. Switching down to previous lower levels of quality when faced with subsequent normal prices could prove unpleasant if a consumer was extremely loss averse for quality. The question of reference point formation was relevant--a single consumption experience was unlikely to change one's reference level of quality in any long-lasting way. However, the temporal aspects of reference point formation need to be more thoroughly studied.

Whether consumers do in fact 'ratchet up' their reference points for quality over a lifetime of consumption is an intriguing question. Duesenberry (1949) suggested that consumption paths did not track income perfectly. In particular, he noted that consumption during periods of income decline was greater than during periods of income growth, even though the level of income was the same. While Duesenberry's primary interest was in understanding saving behaviour, his finding was consistent with the establishment of a series of increasingly higher reference levels for consumption.

Naturally, the question arises as to what conditions are likely to lead to loss aversion for quality. The obvious first condition is that consumers must perceive a difference in quality between goods, and that the current good must be perceived as higher quality than less expensive alternatives. If a consumer believes that a less expensive item is of higher quality than a more expensive one, he or she may be loss averse for quality, but will know that paying more to maintain quality is unnecessary. This would result in a situation resembling status quo bias, where the consumer simply appears to be unwilling to give up the current good.

We suspect that loss aversion will be much more prevalent for those attributes that have a direct impact on the consumer's use or enjoyment of the good: taste attributes of foods, efficiency of appliances, and effectiveness of drugs are all likely to lead to loss aversion because they involve the consumer's personal comfort in a direct way. Furthermore, loss aversion will be more likely when consumer knowledge and involvement is high, since knowledgeable and involved consumers will be more aware of the benefits certain attributes confer. In addition, brand name may serve as a proxy for quality when consumers are only vaguely informed about products.

Finally, this paper supported the notion of mental budgets for certain categories of goods (Thaler 1999) and showed that spending within these accounts was limited by reference points. The question of just how consumers (and organisations or governments for that matter) form their mental accounts is an intriguing one and certainly the least understood aspect of choice examined in this paper. Further examination of this question could improve consumer decisionmaking, tax policy, and strategic decision-making.

Appendix

Stimuli for Experiment 1

Traders-Down Stimulus

The Toblerone Chocolate in front of you is yours to keep. You may take it with you after the session is over. Later in the session, an amount of money will be announced. You will have the option to return your Toblerone in exchange for 'Chocolaty" Chocolate and that amount of money. The amount has already been randomly determined and will not be affected by your responses. For each amount of money listed below, please indicate whether you prefer to keep your Toblerone or return it in exchange for 'Chocolaty' Chocolate and the money. The choice you make for the amount of money that we announce will be carried out at the end of the session.

$2.50    Keep Toblerone    --    'Return For Chocolaty'    --
$2.40    Keep Toblerone    --    'Return For Chocolaty'    --
:        :                 :     :                         :
S0.10    Keep Toblerone    --    'Return For Chocolaty'    --
$0.00    Keep Toblerone    --    'Return For Chocolaty'    --

Chooser Stimulus

You have a choice between a Toblerone Chocolate Bar and a 'Chocolaty' Chocolate Bar plus an amount of money. The amount of money has already been randomly determined and will not be affected by your responses. For each pair of options below, please indicate your choice. The amount has already been randomly determined and will not be affected by your responses. The choice you make for the amount that we announce will be carried out at the end of session.

$2.50    Choose Toblerone    --    Choose 'Chocolaty' + Money    --
$2.40    Choose Toblerone    --    Choose 'Chocolaty' + Money    --
:        :                   :     :                             :
S0.10    Choose Toblerone    --    Choose 'Chocolaty' + Money    --
$0.00    Choose Toblerone    --    Choose 'Chocolaty' + Money    --

Traders- Up Stimulus

The 'Chocolaty' Chocolate in front of you is yours to keep. You may take it with you after the session is over. Later in the session, an amount of money will be announced. You will have the option to return your 'Chocolaty' Chocolate with a payment of that amount of money in exchange for Toblerone Chocolate. The amount has already been randomly determined and will not be affected by your responses. For each amount of money listed below, please indicate whether you prefer to keep your 'Chocolaty' of Return it with a payment of that amount in exchange for Toberone. The choice you make for the amount of money that we announce will be carried out at the end of the session.

$2.50    Keep 'Chocolaty'    --    Exchange for Toblerone    --
$2.40    Keep 'Chocolaty'    --    Exchange for Toblerone    --
:        :                   :     :                         :
S0.10    Keep 'Chocolaty'    --    Exchange for Toblerone    --
$0.00    Keep 'Chocolaty'    --    Exchange for Toblerone    --

Stimuli for Experiment 2

Regular Price Stimulus

This part of the experiment is for those people who, at least occasionally, purchase beer for their own consumption. If you never purchase beer, please do not complete this questionnaire.

What is the name of the beer that you purchase most frequently? --

How much do you usually pay for a six-pack of this beer? --

On the following list, please circle the beer that you would be most likely to buy. The price listed is the current price for a six-pack at a local store. Please circle only one beer.

1. Miller Genuine Draft    $3.38
2. Samuel Smith's          $9.49
3. Michelob                $3.69

Promotional Price Stimulus

Imagine that your regular store is having a one-time promotion where you can purchase any one six-pack of beer for 30% off the normal price. Please circle the six-pack of beer that you would be most likely to purchase at the promotional price. Circle only one beer.

                              Price     Promotional Price

1. Miller Genuine Draft       $3.38     $4.39
2. Samuel Smith's             $9.49     $12.34
3. Michelob                   $3.69     $4.80

Specialty Price Stimulus

Imagine that you are shopping at a specialty store and that it will be terribly inconvenient for you to make a trip to your regular store. At the specialty store, all of the six-packs of beer are 30% more expensive than normal. Please circle the beer you would be most likely to buy at these prices. Circle only one beer.

                              Price     Specialty Store Price

1. Miller Genuine Draft       $3.38     $4.39
2. Samuel Smith's             $9.49     $12.34
3. Michelob                   $3.69     $4.80

Quality Ratings Stimulus

Please rank the quality of the beers listed below on a 9 point scale, where 1 is the lowest quality and 9 is the highest quality. When you make your ratings, remember to use the whole scale. If you have no idea, circle DK.

                                                                 Don't
                             Rating                              Know

1. Miller Genuine Draft      1   2   3   4   5   6   7   8   9    DK
2. Samuel Smith's            1   2   3   4   5   6   7   8   9    DK
3. Michelob                  1   2   3   4   5   6   7   8   9    DK

Table 1

Median and Mean Prices, Experiment 1

                                 Mean    Median    n

Trading Up Price, [P.sub.u]      0.50     0.60     26
                                         (0.42)
Choice Price, [P.sub.c]          0.70     0.80     25
                                         (0.59)
Trading Down Price, [P.sub.d]    1.00     1.28     24
                                         (0.54)

Note: Standard Deviations are in parentheses.

Table 2

Mean Price and Quality Rating of Beer Chosen

  Condition      Price     Quality Rating
                               n = 5

Mean Price       $4.02          6.51
                 (0.98)        (1.78)
Lowered Price    $3.97          7.77
                 (1.08)        (1.29)
Raised Price     $4.90          6.11
                 (1.23)        (1.84)

Note: Standard Deviations in parentheses.

(1.) The reader is referred to Hardie, Johnson and Fader (1993) and Tversky and Kahneman (1991) for details on the assumptions.

(2.) All prices are expressed in US dollars.

(Date of receipt of final transcript: January, 2004. Accepted by Mark Uncles, Area Editor.)

References

Allenby, G.M. & Rossi, P.E. 1991, 'Quality perceptions and asymmetric switching between brands', Marketing Science, vol. 10, pp. 185-204.

Blattberg, R.C. & Wisniewski, K.J. 1989, 'Price-induced patterns of competition', Marketing Science, vol. 8, pp. 291-309.

Dickson, P.R. & Sawyer, A.G. 1990, 'The price knowledge and search of supermarket shoppers', Journal of Marketing, vol. 54, pp. 42-53.

Duesenberry, J.S. 1949, Income, Saving, and the Theory of Consumer Bbehaviour, Oxford University Press, New York.

Gabor, A. & Granger, C. 1964, 'Price sensitivity of the consumer', Journal of Advertising Research, vol. 4, pp. 404.

Hardie, B.G.S., Johnson, E.J. & Fader, P.S. 1993, 'Modelling loss aversion and reference dependence effects on brand choice', Marketing Science, vol. 12, pp. 378-94.

Helson, H. 1948, 'Adaptation-level as a basis for a quantitative theory of frames of reference', Psychological Review, vol. 55, pp. 297-313.

Kahneman, D. & Tversky, A. 1979, 'Prospect theory: An analysis of decision under risk', Econometrica, vol. 47, pp. 263-91.

Kahneman, D., Knetsch, J.L. & Thaler, R. 1990, 'Experimental tests of the endowment effect and the Coase theorem', Journal of Political Economy, vol. 98, pp. 1325-48.

Kalwani, M.U., Yim, C.K., Rinne, H.J. & Sugita, Y. 1990, 'A price expectations model of customer brand choice', Journal of Marketing Research, vol. 27, pp. 251-62.

Kalwani, M.U. & Yim, C.K. 1992, 'Consumer price and promotion expectations: An experimental study', Journal of Marketing Research, vol. 29, pp. 90-100.

Kalyanaram, G. & Little, J.D.C. 1994, 'An Empirical Analysis of Latitude of Price Acceptance in Consumer Package Goods', Journal of Consumer Research, vol. 21, pp. 408-18.

Lilien, G.L., Kotler, P. & Moorthy, K.S. 1992, Marketing Models, Prentice-Hall, Inc.

Loewenstein, G. & Kahneman, D. 1991, 'Explaining the endowment effect', Working paper, Department of Social and Decision Sciences, Carnegie-Mellon University.

Mayhew, G.E. & Winer, R.S. 1992, 'An empirical analysis of internal and external reference prices using scanner data', Journal of Consumer Research, vol. 19, pp. 62-71.

Putler, D.S. 1992, 'Incorporating reference price effects into a theory of consumer choice', Marketing Science, vol. 11, pp. 287-309.

Samuelson, W. & Zeckhauser, R. 1988, 'Status quo bias in decision making.' Journal of Risk and Uncertainty, vol. 1, pp. 7-59.

Sherif, C.W. 1963, 'Social categorization as a function of latitude of acceptance and series range', Journal of Abnormal Psychology, vol. 67, pp. 148-56.

Simonson, I. & Tversky, A. 1992, 'Choice in context: Trade-off contrast and extremeness aversion', Journal of Marketing Research, vol. 29, pp. 281-95.

Thaler, R. 1999, 'Mental accounting matters', Journal of Behavioral Decision Making, vol. 12, pp. 183-206.

Tversky, A. & Kahneman, D. 1991, 'Loss aversion in riskless choice: A reference dependent model', Quarterly Journal of Economics, vol. 106, pp. 1039-62.

Urbany, J.E. & Dickson, P.R. 1991, 'Consumer normal price estimation: Market versus personal standards', Journal of Consumer Research, vol. 18, pp. 45-51.

Winer, R.S., Bucklin, R.E., Deighton, J., Erdem, T., Fader, P.S., Inman, J.J., Katahira, H., Lemon, K. & Mitchell, A. 1994, 'When worlds collide: The implication of panel data--based choice models for consumer behaviour', Marketing Letters, vol. 5, pp. 83-394.

Suzanne Fogel ([dagger])

Dan Lovallo ([section])

Carmina Caringal ([section])

([dagger]) DePaul University, Department of Marketing, 1E. Jackson Boulevard, Chicago, IL 6064. Email: [e-mail]

([section]) Australian Graduate School of Management, UNSW, Sydney, NSW 2052. Email: [e-mail] [e-mail]

The authors would like to thank Daniel Kahneman for his helpful suggestions on an earlier draft. All errors remain our own.