Sangman Han (a)
Sunil Gupta (b,*)
Donald R. Lehmann (b)
We examine consumers' price sensitivity using a new approach that incorporates probabilistic thresholds for price gains and price losses in the reference price models. We model the threshold as a function of company, competitor and consumer specific factors. Model application to scanner panel data for coffee shows that our model is superior in fit compared to ordinary logit and two existing reference price models. Our results indicate that higher own-price volatility makes consumers more sensitive to gains and less sensitive to losses, while intense price promotion by competing brands makes consumers more sensitive to losses but does not influence consumers' sensitivity to gains. Two clear segments that differ in the size of their thresholds emerge. Managerial implications of these results for segmentation and understanding brand power are discussed. [C] 2001 by New York University. All rights reserved.
Keywords: Pricing; Thresholds; Choice model; Reference price
Of all the tools available to marketers, none is more powerful than price. Price has a significant influence on consumers', purchase behavior and consequently on firm sales and profits. It is, therefore, not surprising that price promotion has become an increasingly large fraction of the marketing budget and an almost ubiquitous aspect of consumer choice. In fact, consumers may be both conditioned to expect deals and desensitized to small ones. As a consequence, changes from a base or reference price are likely to have an impact only when the price change is above a threshold.
It is important for a manager to understand price thresholds for at least four reasons. First, it helps a manager decide the minimum price discount needed to have any impact on consumer choice. This may also help retailers negotiate the appropriate level of promotional discount with manufacturers. Second, it provides a useful method of customer segmentation based on how consumers differ in their price thresholds. Consumers with low thresholds for price changes are more sensitive to price changes than consumers with high thresholds. Third, it helps a manager understand and monitor the power of his brand. Brands with low thresholds for price discounts affect consumer purchases with a small cut in their price and, therefore, exhibit greater clout in the market place. Fourth, it provides an opportunity for a manager to identify and manage variables that affect price thresholds and, therefore, power of his brand. For example, it is possible that frequent discounting of a brand may increase its price threshold such that the same level of discount is no longer sufficient to influence consumers' purchases. In other words, even though frequent discounting may benefit the brand in the short run, it may have a negative impact in the long run (Jedidi, Mela & Gupta, 1999). The pattern of change in the brand's price threshold over time can provide a measure of this long run negative impact.
The purpose of this article is to explore the existence and magnitude of price thresholds and the factors that influence these thresholds. We allow price thresholds to be influenced by company, competitor and consumer factors. In addition to the managerial and substantive benefits our approach also makes a methodological contribution by suggesting that price thresholds are probabilistic in nature. These thresholds capture consumer insensitivity to small changes in price, or the zone of price indifference. We also allow the thresholds to be asymmetric for price decreases (or gains) and price increases (or losses).
2. Reference price and price thresholds
Reference price, which is based in part on the past pricing activity of a product, is stored in a consumer's memory and serves as a point of comparison for future purchases. Increased emphasis on promotion and frequent discounting has sparked significant academic and managerial interest in better understanding the role of reference price in consumer decision-making. It is feared that frequent discounting of a brand may lower its reference price in consumers' minds, thereby having a long-term negative impact on the effectiveness of that brand's promotions.
Several studies have examined the issues related to reference price. These studies conclude that reference prices exist and play a significant role in consumers' choice of brands (Briesch, Krishnamurthi, Mazumdar & Raj, 1997, Kalyanaram & Winer, 1995, Kalwani et al., 1990, Krishnamurthi, Mazumdar & Raj, 1992, Lattin & Bucklin, 1989, Monroe, 1990, Winer, 1986). It has also been shown that if a consumer encounters a brand at a price lower than its reference price, it is perceived as a gain. Conversely, a price higher than the reference price is perceived as a loss. Consumers derive transaction utility from this gain or loss over and above the utility obtained from the product itself (Kalwani & Yim, 1992, Mayhew & Winer, 1992, Thaler, 1985).
Studies have also shown that consumers respond differentially to prices above or below the reference price. Generally consumers react more negatively to losses than they do positively to gains (Kahneman & Tverskey, 1979, Mayhew & Winer, 1992). However, some studies find that consumers are not always more responsive to a loss than to a gain (Krishnamurthi, Mazumdar & Raj, 1992, Greenleaf, 1995). Finally, several studies show that there is a latitude of acceptance or zone of indifference around the reference price, such that minor changes in price do not have any significant impact on consumer choice. Put differently, consumers have thresholds for gains and losses. Unless the difference between actual and reference price is higher than these thresholds, consumers do not experience any positive or negative transaction utility (Gupta & Cooper, 1992, Kalyanaram & Little, 1994, Helson, 1964, Kalwani & Yim, 1992).
Our study builds on these previous results in four important ways. First, unlike the previous studies, we propose that thresholds for gains and losses are probabilistic rather than deterministic. All marketing studies that use the notion of threshold assume that these thresholds are deterministic (e.g., Gupta & Cooper, 1992, Kalyanaram & Little, 1994, Kalwani & Yim, 1992). Since thresholds are not observed, it is unlikely that we capture all the idiosyncracies of consumers and the environment in our estimation of these thresholds. Therefore, it is more reasonable to assume that they are probabilistic. Second, from theoretical and managerial perspective it is important to know what factors affect these thresholds. Our study examines these factors and their influence on thresholds. Third, we distinguish the asymmetric impact of explanatory variables on the thresholds for gain or loss. Fourth, our study provides two important implications for managers -threshold-based segmentation and threshold-based brand power .
In the following section we describe our model. Next, we provide an illustration of our model on scanner panel data for coffee. This is followed by managerial implications and conclusions.
3. The model
We start with a brief description of the brand choice model. We then describe how reference price effects and probabilistic thresholds are modeled and subsequently incorporated within the brand choice model.
3.1. Brand choice model
Consistent with considerable past work in marketing (e.g., Guadagni & Little, 1983) we use a random utility framework where consumers buy the brand that maximizes their utility. Assuming a linear utility function and errors with a double exponential distribution, we get the multinomial logit model where the probability that household h selects brand i at purchase occasion t is given as
[p.sup.h.sub.it] = exp ([U.sup.h.sub.it])/[[SIGMA].sub.k]exp ([U.sup.h.sub.kt]) (1)
where the deterministic component of household h's utility for purchasing brand i at time t is
[U.sup.h.sub.it] = [u.sub.i] + [beta][X.sup.h.sub.it].
The parameter [u.sub.i] is the brand-specific intercept for brand i and [beta] is the vector of response coefficients for the explanatory variables [X.sup.h.sub.it]. The explanatory variables include householdspecific variables (such as brand loyalty), and marketing variables (such as price and promotion).
The advantage of the logit model is its ease of estimation and its wide application among industry practitioners (Bucklin & Gupta, 1999). One of its disadvantages is its property of independence of irrelevant alternatives or IIA. However, two factors mitigate this limitation. First, even if IIA holds at the individual level, it does not hold at the aggregate level (Ben-Akiva & Lerman, 1985). Second, IIA is usually not a problem if the set of brands is carefully defined (e.g., ground caffeinated coffee).
3.2. Reference price effects
The concept of reference price has a strong theoretical and empirical support (see Kalyanaram & Winer, 1995 for details). Theoretical arguments for reference price derive from Adaptation-Level theory (Helson, 1964). This theory states that the perceived magnitude and effect of a stimulus depends on the relation of that stimulus to preceding stimuli. Many researchers have provided empirical support for this concept of reference price (Winer, 1989, Lattin & Bucklin, 1989, Kalwani et al., 1990).
Prospect Theory, as advocated by Kahneman and Tversky (1979), further advanced the argument of reference price. This theory also suggested that consumers' response to gains and losses is asymmetric. In general, consumers react more negatively to losses than their positive reaction to gains. Many studies have modeled and shown this asymmetric effect around reference price (e.g., Mayhew & Winer, 1992). However, some studies have found that consumers are not always more responsive to a loss than to a gain (e.g., Greenleaf, 1995). More recently, some studies have shown that the reference price effect is attenuated by heterogeneity (Bell & Lattin, 2000).
Following this vast body of theoretical and empirical evidence we incorporate the reference price effects (i.e., transaction utility due to gain or loss with respect to reference price) by including a variable to Eq. (2). Specifically, for a gain situation, shelf price [P.sub.it] is less than the reference price [RP.sup.h.sub.it]. Therefore, Eq. (2) is modified as
[U.sup.h.sub.it] = [u.sub.i] + …