Context Price Sensitivity

JOEL HUBER, MORRIS B. HOLBROOK, and BARBARA KAHN*

Previous marketing research on the effects of price has tended to ignore the roleof such moderating variables as competitive context and the availability of addi-tional information. The authors presented subjects with forced-choice decisions inseveral product categories. Using a multinomial logit formulation to measure pricesensitivity, they find relatively lower price sensitivity (1) when a brand is placed atthe upper price-quality boundary of a choice set rather than in the middle, (2) whenonly brand names are provided as opposed to only quality ratings, and (3) whenquality information is added to available brand-name information. They discuss the

theoretical and managerial implications of these findings.

Effects of Competitive Context and ofAdditional Information on Price Sensitivity

As a key variable in the marketing mix, price has beengiven considerable attention by marketing researchersstudying the effects of pricing on consumer responses.However, excellent reviews by Gabor, Granger, andSowter (1970), Monroe (1973, 1976), Monroe andKrishnan (1985). Olson (1977). and Parsons and Schultz(1976) suggest little effort has been made to determinethose variables that moderate the effect of price on brandchoices.

If such moderating effects are conceived as differ-ences in price sensitivity (represented by the slope ratherthan the level of the demand curve), relatively few re-searchers have attempted to account for significant mod-erators of price effects. Early studies focused on the ef-fect of price on perceived quality and showed that itincreases when price serves as the only available cue incomparison with the inclusion of such additional infor-mation as brand name or store name (Olson 1977; Staf-ford and Enis 1969). In an elaborate factorial study,Monroe (1976) showed that sensitivity to with in-subjectprice manipulations decreased when the price of the pre-ferred brand was lowered rather than raised. Further work

•Joel Huber is Associate Professor, Fuqua School of Business, DukeUniversity. Monis B. Hoibrook is Associate Professor. Graduate Schoolof Business. Columbia University. Barbara Kahn is Assistant Profes-sor, Graduate School of Managemetit, University of California, LosAngeles.

The second author gratefully acknowledges the support of the Co-lumbia Business School's Faculty Research Fund.

has examined the effects of different ways to expressprice discounts (e.g., cents off or percentage reductions)and generally has found that greater self-reported will-ingness to buy derives from relative as opposed to ab-solute expressions of price (Barnes 1975; Berkowitz andWalton 1980; Delia Bitta, Monroe, and McGinnis 1981).These studies have used such multiple indicators of theeffects of price as (I) judgments of the quality or effec-tiveness-in-use of the brand, (2) estimates of how gooda value the brand is for the money spent, and (3) indi-cators of willingness to purchase the brand. This mul-tiple-indicator approach has the advantage of capturingvarious psychological pricing effects, but has generallystopped short of measuring effects on actual choice be-havior (as opposed to purchase intentions).

In our experimental study, we extend the previous workusing multiple indicators of price-effect moderators to acontext of forced choice. Price sensitivity is operation-alized as an elasticity that reflects the percentage changein choice share divided by the percentage change in price.Two potential moderators for this measure are investi-gated, (1) the competitive context of the brand and (2)the nature of additional available Information. As theseare somewhat unfamiliar variables, we first describe themand then explain how each is expected to moderate pricesensitivity.

Competitive Context

Competitive context is defined here as the relative po-sition of a brand in comparison with those with which

250

Journal of Marketing ResearchVoi. XXIII (August 1986). 250-60

EFFECTS ON PRICE SENSITIVITY 251

it competes in a choice set. To simplify the concept ofrelative position, suppose all brands can be categorizedmeaningfully by the price charged and a measure of per-ceived quality. If dominated brands (higher in price butlower in perceived quality) are omitted from consider-ation, the remaining brands of interest form a price-qual-ity continuum with the lowest quality, lowest price brandsat one end and the highest quality, highest price brandsat the other. We define bracketed brands in a choice setas those in the middle of this continuum, and the low-boundary and high-boundary brands as the two brandsat the extremes.

The research question is:

—Does the addition or deletion of a competitor that altersthe competitive context (hracketlng) of a brand also changethe hrand's price sensitivity?

The practical importance of this question is readily ap-parent. It reflects the extent to which well-known high-boundary brands such as Chivas Regal scotch. Rolls Royceautomobiles, and Joy perfume derive their profitabilityfrom their respective positions at the upper ends of price-quality continua. In particular, would the price sensitiv-ity of such brands (and thus their optimal prices) changesignificantly if they became bracketed by competitors withpositions at still higher levels of price and perceivedquality? Of comparable strategic concem are the effectsof bracketing low-end brands—for example, by the en-try of generic offerings. We empirically test the effectof bracketing oniy at the upper end of the price-qualitycontinuum. However, the hypothesized effect would oc-cur regardless of whether the experimental brand is thelow- or the high-boundary brand.

Specifically, we hypotfiesize that modification of achoice set that changes a brand from being bracketed tobeing at the boundary also decreases the brand's pricesensitivity for two reasons. First, the relative price canbe gauged more easily in the bracketed condition. Eval-uating the price in a boundary condition is more difficultbecause, instead of being able to interpolate betweencompetitive alternatives, one must extrapolate beyondthem to evaluate the price of the boundary altemative.Besides the difficulty of such an extrapolation as a featof cognitive algebra, it is not clear statistically or psy-chologically that the price-quality tradeoff between twoitems ought to extend beyond their range. For example,equal increments of perceived quality might decrease invalue as one moves to higher price alternatives. Thus,any simple projection has more intuitive justificationwithin rather than beyond a set.

Second, as developed more formally by Hauser andShugan (1983), bracketing tends to place a brand in di-rect competition with a greater number of neighboringbrands. Hence, if a brand is at a boundary of the choiceset, a switch to a competitor is less likely simply becausethere are fewer adjacent competitors to which to switch.

Consideration of both information-processing andcompetitive-substitution effects thus suggests the hy-

pothesis that price sensitivity will be relatively greaterfor bracketed brands than for brands at the boundary.

Availability of Information

A second moderator of price sensitivity is the avail-ability of infonnation in addition to price that can beused to make the choice. We consider two questions indetail:

—Does the availability of brand information alone result inrelatively more or less price sensitivity than is associatedwith the availability of quality infonnation alone?

—Does the addition of perceived quality ratings to brandinfonnation result in relatively more or less price sensi-tivity than is assoeiated with brand information alone?

The first question concems the relative effect of brand-ing versus other information on price sensitivity. Thesecond question concems the relative impact of qualityratings on price sensitivity for known brands. The latterissue applies to consumerist groups or to marketers con-sidering the dissemination of competitive ratings of thetype collected by Consumer Reports. For both of thesequestions, directly relevant research is scant and has pro-duced conflicting predictions.

Effect of Brand Names Versus Quality Ratings

We now consider how relative price sensitivity can beexpected to change as a function of whether additionalinformation includes brand names or perceived qualityratings. (Subsequently we address what happens whenboth types of additional information are included.) Wepresent reasons why familiar brand names and perceivedquality ratings should be associated with relatively lowerand higher price sensitivity, respectively.

In comparison with a response to a quality rating, afamiliar brand name evokes a more elaborate system ofinterconnected beliefs about the product (e.g., Olson 1977,p. 278). In particular a brand name can serve as a "ges-talt" or "chunk" of information that alters the impor-tance of other avaiiable cues (Lutz and Bettman 1976;Simon 1974), For example, in choice studies where sub-jects select cues needed for a simulated purchase deci-sion, the acquisition of brand-name infonnation tends toinhibit subsequent selection of other information (Ja-coby, Szybillo, and Busato-Schach 1977). A consumerwho thus treats the brand name as a "chunk" may ignoreprice or may simply assume that price is reasonable andthereby lower price sensitivity.

We expect perceived quality ratings to have the op-posite effect. Unlike brand names, perceived quality rat-ings make it easier to assess the appropriateness of anaccompanying price. They provide a way to determinevalue f or the money by comparing marginal changes inperceived quality with marginal changes in price. Thus,the availability of perceived quality ratings should en-courage one to balance price against quality with a re-sultant greater price sensitivity than is associated withbrand infonnation.

252 JOURNAL OF MARKETING RESEARCH, AUGUST 1986

Effect of Adding Perceived Quality Ratings to Brand sNames

From a theoretical perspective the effect on price sen-sitivity of making perceived quality information avail-able in addition to brand names is much more ambiguousthan the effect discussed for brand names versus qualityratings. Though there are good reasons to expect theavailability of quality ratings on known brands to in-crease one's capability to evaluate price, the very pres-ence of this new information may lessen the motivationto perform such an evaluation. Because the issues of ca-pability and motivation lead to opposing predictions, weview this aspect of the study as exploratory.

Increased capability. There are two mechanismswhereby the addition of perceived quality to brand in-fomiation is expected to raise relative price sensitivity.The first mechanism derives from the idea that a per-ceived quality rating enables one to evaluate the mar-ginal benefit of given price increments, thereby facili-tating a direct tradeoff of price and quality. The secondmechanism derives from the idea that the new infor-mation makes one more certain about the quality of agiven brand. If so, a smaller price change may be neededto alter the brand's acceptability. The latter account isconsistent with the finding of Peterson and Wilson (1985)of higher price sensitivity under conditions of lower risk.To the extent that perceived quality ratings reduce ex-pected quality risk, a relatively higher level of price sen-sitivity would be expected.

Decreased motivation. Though additional quality in-formation may make the evaluation of a brand's pricemore feasible, the processing cost of the infonnation maymake that evaluation less likely. In particular, the ad-dition of quality to brand information is expected to makea price-versus-quality analysis more difficult. A price-versus-quality analysis involves the assessment of howpositively one values a brand in comparison with thenegative value of its price. To the extent that this taskis sensitive to the amount of information, the addition ofany new data may reduce the likelihood that a price-versus-quality assessment is made. Further, perceivedquality information puts particular demands on process-ing. As perceived quality ratings are added to brandnames, the evaluation requires merging the two types ofinfonnation and then comparing this merged utility againstprice. However, because perceived quality and brand in-formation are in different forms, merging them may bedifficult. Consumers in such a situation may avoid theanalysis altogether rather than engage in a potentiallydifficult and time-consuming task.

Summary. The addition of quality ratings to brandnames may increase the potential for price sensitivity whiledecreasing the extent to which the information actuallyis used. Thus, because considerations of capability ver-sus motivation suggest different outcomes, we view thispart of our study as an exploratory investigation of thenet impact of these counterposed tendencies.

METHODTo date, little theory and even less empirical work has

addressed the moderating effects of competitive contextand of additional information on price sensitivity. Onereason for this gap is the difficulty of measuring eventhe main effects of price with any degree of realism. Forexample, in an effort to gain experimental control overthe relevant independent variables, much research (oursincluded) involves laboratory techniques in which real-world income constraints fail to operate, thereby raisingthe possibility of underestimating price sensitivity andthus reducing extemal validity. We cope with the prob-lem of unrealistic income constraints by focusing not onincreases or decreases in absolute levels of price sensi-tivity within groups, but rather on differences in pricesensitivity among experimental treatment groups ex-posed to contrasting competitive contexts and differenttypes of additional information. In the following sectionswe show that by using experimentally manipulated stim-uli and by capitalizing on the efficiency of a disaggregatetogit model, we can estimate with remarkable reliabilitythe differences in price sensitivity due to experimentalmanipulations.

Subjects and TaskThe subjects were 72 residents of a suburban middle

class community and 46 students in an urban universitysetting. We expected these contrasting groups to havesomewhat different preference orderings, which wouldenable us to assess the generalizability of our findings.The choice task was conducted via individually admin-istered questionnaires completed independently by sub-jects, typically at their homes or offices.

The primary task involved asking subjects to makechoices in several product classes, with altematives fromeach class listed on separate sheets of paper. The coversheet contained the following instructions.

This survey is intended to assess your preferences for var-ious products in six produci classes. On each of Ihe fol-lowing six pages, please check the product that representsyour first choice.

On some of the product descriptions you will be given aquality rating. This rating is based on a scale of - 3 lo+ 3, where - 3 repre.sents the lowest quality rating and+ 3 represents the highest quality rating. These ratingswere assigned to these products based on an earlier sur-vey of people like yourself.

The six product classes had the same structure as thesample in Figure 1 except, depending on the informationcondition, either the brand name or the quality ratingsmight have been omitted. The subjects took about 10minutes to make the six choices. Then they were askedto indicate their beliefs about and familiarity with eachof the brands in the six product classes.

Products and BrandsThe six product categories, shown in Table 1, were

peanut butter, pancake syrup, dishwashing liquid, type-


Figure 1SAMPLE CHOICE QUESTION

{color TV sets, RCA is experimental brand, unbracketedcondition, low price)

Small-screen color television setsBuilt-in or clip-on VHF and UHF antennasNo vertical or horizontal hold control (none needed)Lighted channel numbers75-ohni cable TV connection

Please check Brandyour first choice name Price

Averagequality rating(-j'lo +3)

R P A

Phi InnI r P^nn*.'

$380$370

Y $330

.5- . 7

- 1 . 3

writers, 10-speed bicycles, and color television sets. Eachproduct category had four brands from which three wereselected to make up each choice set. The altemativeswere designed to span the price-quality continuum fromhigh price/high quality (e.g., Skippy peanut butter orSmith-Corona typewriters) to low price/low quality (e.g.,A&P peanut butter or J.C. Penney typewriters).

The product classes and brands were pretested to meetcertain criteria. All were relatively well known, reflect-ing the kinds of products that subjects might purchase.Further, the brands were chosen so that at least 90% ofrespondents in a pretest agreed with the rank order ofthe given quality ratings. Finally, both prices and per-ceived quality ratings were designed to be consistent withthe subjects' prior beliefs. Specifically, prices were sim-ilar to those actually charged in stores at the time andperceived quality ratings were based on scores given bysimilar respondents in a pretest questionnaire. The pur-

pose of making the information consistent with the sub-jects' likely prior beliefs was to increase the realism ofthe task and to avoid contrast effects (Sherif and Hov-iand 1953).

Manipulation of Competitive Context

To generate choice sets in the boundary condition, thetop brand was deleted from the four possible altemativesin Table 1 so that the exi>erimental brand would havethe highest levels of price and perceived quality in theoffered set. For the bracketed condition the top brandwas retained and the low price/low quality brand wasdeleted. Thus, in all cases, subjects were asked to chooseone item from three offered in the product classes.

Two aspects of this way of manipulating competitivecontext bear comment. First, the bracketing effect is testedonly in the upper boundary condition. The reason for thislimitation is simply to provide as much statistical poweras possible in an already complex experimental design.Second, competitive context is manipulated by replacingthe highest price brand with the lowest price one. Thisdesign avoids confounding the bracketing-versus-bound-ary manipulation with the number of brands included inthe choice set. The penalty is that the bottom price (lowversus moderate) varies between the boundary and thebracketed conditions, as does the high price (experi-mental versus top brand). A test of what would happenif the low price brand were retained in all treatments istherefore of interest, but is left for future research.

Manipulation of Available Information

Availability of additional information was manipu-lated by varying the kinds of cues given to subjects asa basis for each choice. Relative prices were alwaysavailable. In addition, depending on the experimental

Table 1ALTERNATIVES IN PRICING STUDY

(stimuli, perceived quality ratings, and prices)-.1

Top brandQuality ratingPrice

Experimental brandQuality ratingPrice (high)Price (low)

Moderate price brandQuality ratingFYice

Low price brandQuality ratingPrice

Peanutbutter

Skippy2.4

$2.10

Jit"1.7

$2.06$1.78

Nutnade-0.7$1.73

A&P-0.8$1.70

Pancakesyrup

Log Cabin1.7

$1.77

G. Griddle0.5

$1.74$1.56

Embassy-0 .3$1.53

A&P-0.9$0.99

Dishwashingliquid

IvoryI.I

$1.70

Palmolive0.8

$1.69$1.57

Lux-0.2$1.55

A&P-1.7$0.70

Type-writers

S. -Corona0.9

$399

Royal0.3

$388$305

Olympia-0 .3$295

J.C. Penney-0 .9$238

lO-speedbikes

Raieigti1.2

$225

Schwinn0.6

$219$175

Nishiki-0 .3$170


Color TVsets

Sony1.5

$470

RCA0.5

$460$380

Phi Ico-0 .7$370



group, brand names, perceived quality ratings, or bothpieces of infonnation also were included.

It is relevant to ask why the design did not include acell in which only price information was availabie, be-cause such a cell would have completed a full factorialdesign representing the presence and absence of all com-binations of brand and quality information. The reasonis that subjects in pretests found choices with only priceinformation to be highly problematic and unrealistic. Theywanted to know what they were getting for the higherprice—something that was known if brands or qualityratings were provided. Two choice strategies emerged inthe price-only condition. Subjects either trivially chosethe lowest price or tried to match the price they wouldpay. In either case, the concept of price sensitivity wasfound to have a different meaning to subjects than it didwhen either brand names or quality ratings were avail-able. Thus, the price-only cell was dropped from the de-sign.

Manipulation of Price

The price manipulation involved altering the price ofthe experimental brand given to the subject groups. Thegoal was to vary the price as much as possible whilekeeping it below that of the top brand and above that ofthe moderate brand. Accordingly, the price of the ex-perimental brand was manipulated by setting its price ateither 10% or 90% of the distance between the prices ofthe top and moderate brands.

Overall Design

Figure 2 summarizes the full design. There were twoprice conditions, two competitive context conditions, andthree infonnation conditions. These 12 cells were as-signed systematically to the 708 choices {118 subjects x6 choices per individual), subject to various constraints.Each person made only one choice from each productelass, two of which came from eaeh of the three infor-

Figure 2EXPERIMENTAL DESIGN

Price conditions (2 levels)1. Low: Price of experimental brand {/"eip) is close to price of

middle brand {P^

' ' t i p == fmoiS + .10(P|,>p ~ f*moa) ''

2. High: Price of experimental brand (Pt.p) close to price of topbrand (P.,^)

Additional information conditions {3 levels)1. Price and brand name2. Price and perceived quality rating '3. Price and brand name and perceived quality rating

Competitive context conditions (2 levels)1. Bracketed: choice set = {top, experimental, moderate}2. Unbracketed: choice set = {experimental, moderate, low}

mation conditions. Further, though complete within-sub-ject balance was obviously impossible, each subject re-ceived at least one case of each combination of the twoprice and the two competitive context conditions. Fi-nally, the order of the product classes was randomized,subject to the constraint that the durable products (type-writers, 10-speed bikes, TV sets) and nondurable prod-ucts (peanut butter, symp, dishwashing liquid) appearedin separate contiguous sets.

The design enabled us to estimate price sensitivity ef-ficiently for the experimental brand within each of thevarious conditions, This task was not expected to mirrordirectly price sensitivity in the marketplace for two rea-sons. First, no explicit reference was made to either abudget constraint or a need for the product. Second, therewas no real cost—the exercise was clearly hypothetical.Despite its hypothetical nature, however, subjects in theexercise certainly acted as though a real budget con-straint affected their decisions. That is, they spoke ofbrands as being "too expensive" or having "poor valuefor the money." Thus, though there is some artificialityin the task, we believe the level of realism in the taskis sufficient for the pattem of differences in price sen-sitivity across experimental groups to correspond rea-sonably to actual choice behavior.

RESULTS: AGGREGATE ARC ELASTICITY

We first examine the relative effects of the experi-mental manipulations on differences in aggregate arcelasticity. These effects are assessed separately for thetwo competitive conditions (bracketed versus boundary)and for the three information conditions (brand infor-mation alone, quality information alone, or both). Here,the data are pooled across the six prtxiuct classes. Thislimited aggregate analysis gives a feel for the data. Sub-sequently, we apply more rigorous statistical tests to buildproduct- and brand-specific effects into the context of amultinomial logit model.

Table 2 shows the effect of the price manipulation onthe percentage choosing the experimental brand in thebracketed condition. Across all product classes, the ex-perimental brand at its higher price received an averageof 24% of the choices. At its lower price, representingan average 16% (AP/P) reduction, the experimentalbrand's share increased to 43%. Thus the percentage in-crease in share, ^Q/Q, is 57%, resulting in an estimatedarc elasticity, (AQ/Q)/iAP/P), of -3 .52 . Table 2 alsogives the comparable shares for the experimental brandin the boundary condition, 61% versus 65%. The elas-ticity ofthe experimental brand drops across competitiveconditions from -3.52 to - . 3 4 .

The data in Table 2 support the bracketing hypothesisby showing relatively greater price elasticity in thebracketed than in the boundary condition. Table 3 sep-arates this analysis for the three infonnation conditions.The first column indicates that for the quality-only treat-ment group the relative elasticity of the experimental brandchanges from —6.31 to - . 9 2 between the bracketed and


Table 2EFFECT OF PRICE ON CHOICE PROBABILITIES

Price ofexperimental

brand

Bracketed conditionHighLow

Estimated arc elasticity of experimental brand = -3.52

Boundary conditionHighLow .

Estimated arc elasticity of experimental brand = - .34

Topbrand

5949

Experimentalbrand

6165

Percentage choosing

Experimentalbrand

2443

Moderatebrand

2212

Moderatebrand

178

Low pricebrand

1723

n

166157

173173

boundary conditions. Comparable figures for the brand-only treatment are -4 .03 and - . 2 0 . Finally, when bothpieces of information are provided, the estimated arcelasticity is - . 6 1 and .02, respectively. Thus, within thethree infonnation conditions, price sensitivity is rela-tively lower for brands at the upper boundaries of theirprice-quality continua.

In terms of differences in price sensitivity due to themanipulation of additional information, elasticity is leastextreme when both pieces of infonnation are available,moderate when only brand names are given, and mostextreme when only quality ratings are provided. Theseresults support the hypothesis that the availability of brandnames alone results in relatively lower price sensitivity,in comparison with the elasticity associated with theavailability of quality ratings alone. It appears that, inaccord with the motivational effect, the addition of qual-ity infonnation to brand names results in still lower pricesensitivity. We discuss the implications of these resultsafter using a multinomial logit analysis to estimate moreprecisely the relative magnitude and the statistical sig-nificance of these effects.

DISAGGREGATE LOGIT ANALYSIS

Though the preceding discussion of aggregate arcelasticity provides an understanding of the magnitude andmeaning of the results, the pooling of heterogeneoussubsets clouds any statistical inferences. Further, dis-

Table 3ESTIMATED ARC ELASTICITY WITHIN EXPERIMENTAL

CONDITIONS

Competitivecondition

Information available

Quality ratings Brand names Both

BracketedBoundary

-6.31-0.92

-4.03-0.20

-0.610.02

aggregate analysis is needed to estimate relative pricesensitivity within experimental conditions and productclasses in a manner that avoids potential problems dueto small cell sizes.

Accordingly, we use a multinomial logit model to ac-count parametric ally for the various experimental con-ditions and product classes. This model takes the form

(I)

where Q^ is the proportion of choices altemative / re-ceives and V, is a covert measure of the utihty of brand/. which is a function ofthe brand's characteristics {.v,i's,k= 1. . . . , A-),

(2) = exp

The x,i's reflect characteristies of the individual alter-natives in the various choice sets and the B 's are theircoefficients. For example, one x,* is an indicator variablefor the top brand of color TV sets. These variables pri-marily serve as covariates that facilitate more precise es-timates of the price effects.

A group of the dependent variables, denoted Xp^ (withsubscript p to denote a price term), is defmed as PjP^-,where P^ is the experimental brand's actual price and Pîs the average of the experimental brand's high and lowprices in condition k. The price variable takes nonzerovalues only for the experimental brand because the pricesof the other brands do not vary. Given this coding, themarginal effect of a price change on the probability {Qôf choosing the experimental brand can be shown bycombining equations 1 and 2 and differentiating to ob-tain

(3)

where B j is the eoefficient for price in the logistic


regression model and Q,, is the estimated choice share ofthe experimental brand given the parameters. The esti-mate of elasticity then has a very simple form.

(4)

This enables us to examine elasticity levels (for relativecomparisons among treatments) by considering the Bp^coefficients.

We estimated the multinomial logit model by meansof a computer routine using an adaptive search procedureto find parameters that maximize the likelihood of theactual choices given the model specification. Chi squaretests on the likelihood ratio enable one to test the sig-nificance of composite nested hypotheses. In addition,r-tests for specific variables are possible because the rou-tine supplies asymptotic estimates of the variances andcovariances of the parameters. These estimates are de-rived from the inverse of the matrix of second deriva-tives of the likelihood function with respect to the pa-rameters (Wilks 1962, p. 418).

RESULTS OF MULTINOMIAL LOGIT ANALYSIS

The multinomial logit model was applied to the choicesof 118 respondents each making six selections. Thisanalysis used a total of 28 parameters, and the resultingfit was significantly better than that of the null model(X^(28) = 308, p < .0001). Twenty-four of the param-eters were needed as covariates to model the 708 choicesin one equation. These 24 covariates account for heter-ogeneity across subjects, products, and experimentalconditions, thereby mitigating the problem of poolingcross- and within-subject data.

Role of the Covariates

Tables 4 and 5 show results for the covariates in themultinomial logit analysis. At the top of Table 4 are theaverage coefficients for the four competitive positions.AU four coefficients in each group are given for clarity;however, in the model one is redundant and was not in-cluded in the estimation procedure. To derive probabil-ities these coefficients must be modified by an exponen-tial transformation and then entered into equation 1. Forexample, the top brand has a coefficient of .83, whereasthe exf>erimental brand (whose price is defined here ashalf way between its high and low experimental condi-tions) has a coefficient of .48. These coefficients trans-late into an odds ratio of exp(.83)/exp(.48) = 1.5. Themodel therefore indicates that the top brand has aboutone and one-half times as many choices overall as the ex-perimental brand. The prediction of shares within a choiceset follows similar logic. Thus, the probability of choos-ing the experimental brand in a set that includes the topand the moderate brand as competitors is exp(.48)/lexp(.48) + exp(.83) + exp(-.82)] or .37.

Substantively, these results show that, across productclasses, the top brand was generally most preferred, fol-

Table 4LOGIT ANALYSIS Of CHOICE EXPERIMENT: COVARIATES

Coefficient

Average utilities of competitive positionsTop brandExperimental brandModerate brandLow price brand

Deviations due to quality ratingsTop brandExperimental brandModerate brandLow price brand

Deviations due to brand muneTop brandExperimental brandModerate brandLow price brand

Deviations due to both togetherTop brandExperimental brandModerate brandLow price brand

.83

.48-.82-.49

,09'-.06-.11^jm

- . 3 9- . 2 9- . 2 4

.92

.30.35.35

-LOO

S.E.

.10

.07

.10

.10

.13

.09

.13

.13

.13

.09

.12

.14

.12

.09

.13

.14

The effects are additive so (hat the coefficienl for the top brand inthe quality condition is (.83) + (.09) = .92, This means that, whenaccompanied by quality ratings, the top brand had a utility score thatwas .92 above the utility of the other competitive positions. Note alsothat the fourth parameter in each set is redundant because they arecoded to stim to zero. For example, in the case of the average utilities(.83) + (.48) + (- .82) + (- .49) = 0.

lowed by the experimental brand, then the low price brand,and finally the moderate brand. Recall that these coef-ficients include the price impact and thus reflect the pur-chase utility of each altemative.

Table 4 also shows how the coefficients for the var-ious competitive positions change under the various in-formation conditions. For example, the coefficient forthe top brand increases by ,09 over its normal coefficientwhen only quality ratings are available.

Table 5 lists the coefficients representing utility de-viations for brands within each product category, ln thepresent analysis these coefficients serve primarily to ac-count for expected heterogeneity and, as covariates, topermit more powerful tests of the hypothesized contraststo which we now tum.

Test of Competitive Context and Addition ofInformation Effects

Table 6 summarizes the analysis of the price-changeterms. As only the experimental brand's price was ma-nipulated, all results refer to its estimated elasticity. Itsaverage price coefficient is -3 .63. According to the logitmodel the elasticity depends on the market share of theexperimental brand (e = Bpi,i\ - Q^)). This market share,Qk, varies across the product classes, and among thecompetitive and the information conditions. Because the


choice set always comprised three items, a reasonableway to set Q in summarizing the results is at the pointof equal shares where Q = .33. Other transformationswould result in estimates that differ by only a multipli-cative constant from those given.

The elasticity estimates in the experimental conditionsshown in Table 6 parallel the aggregate arc elasticity lev-els of Tables 2 and 3. In the bracketed condition, esti-mated elasticity is —4.1, whereas in the boundary con-dition it drops to —.7. The difference between these twoeffects is statistically significant (t = 2.8, p ^ .01). Thisresult supports the bracketing hypothesis.

In the three information conditions, the quality ratingsalone resulted in the most extreme estimated elasticity(e •= -4 .3) , followed by brand names alone (e = —2.5)and then by both pieces of information together {e =— .5). As hypothesized, the availability of brand namesalone produced significantly lower estimated price elas-ticity than did quality ratings alone (/ - 1.7, p ^ .05).The addition of quality information to brand names pro-duced lower estimated elasticity (t = 2.6, p < .01) incomparison with brand names alone.

Table 6RESULTS FOR LOGIT ANALYSIS OF PRICE-CHANGE

TERMS

TableDEVIATIONS IN UTILITIES

Product class

Peanut butter

Pancake syrup

Dishwashingliquid

Typewriters

10-speed bikes

Color TV sets

5FOR BRANDS WITHIN

PRODUCT CATEGORIES

Product

SkippyJifNumadeA&P

Log CabinGolden GriddleEmbassyA&P

IvoryPalmoliveLuxA&P

Smith-CoronaRoyalOlympiaJ.C. Penney

RaleighSchwinnNishikiJ.C. Penney

SonyRCAPhilcoJ.C, Penney

Price ($)

2.102.06-1.78

1,731,70

1,771.74-1.56

1.530.99

1,701.69-1,57

1,550.70

399388-305295238

225219-175170100

470460-380370330

Coefficient

-,20',33

- 5 1+ .38

-.28-.06

,44-.10

,04-,25- 3 1+ .52

-.12-.30-.11+ .53

.05,45.12

-.62

.51-.17

.37-.71

S.E.

.23

.20

.20,29

.20

.21

.18

.23

,17.21,19.23

.18

.17

.17

.18

.18

.21

.19

.21

.20

.22

.20

.22

Average price effect

Effect given competitiveBrand is bracketedBrand is at boundary

Effect given informationQiiality ratingsBrand namesBoth

Coefficient

E.st.

-3,63

condition-6.18-1.09

condition-6.49-3.97-0.79

S.E.

0.9

0.90.9

Ul

Ul

Ul

Elasticity'

Q = .33

-2.4

-4.1-0.7

-4.3-2.5-0.5

S.E.

0.7

1.01.0

LI1.11.2

"This coefficient can be interpreted as the deviation from the av-erage of that brand. Thus the utility of Skippy at $2.10 is (.83) forthe utility of the top brand plus ( -20) for its deviation from the av-erage.

The elasticity estimator is derived by B t(I - Q), Thus for thequality condition this is (-6.49X1 - ,33) = -4.3,

Sensitivity of the Results to Different Specifications

The elasticities from the logit analysis are remarkablysimilar to the arc elasticities. However, by accountingfor the various covariates, the multinomial logit proce-dure affords greater confidence in and more powerfulstatistical tests of the results. In addition, the parametricframework permits further tests of the stability of theresults.

Several runs were made to test the sensitivity of ourresults to different model specifications. The basic con-clusion was that, even when additional covariates werestatistically significant, they had very little effect on thefindings relevant to the three major research questions.For example, one analysis allowed each product class tohave its own idiosyncratic elasticity term. The gain inlikelihood due to these six additional variables was sig-nificant (x^(5) = 18.2, p < .001) but, in part becauseof the orthogonal nature of the design, they had virtuallyno impact on the degree to which the experimental con-ditions moderated the relative price sensitivity.

A second test of stability concemed possible interac-tions between bracketing and the information conditions.We suspected that, once price sensitivity was diminishedin the boundary condition, the differences due to infor-mation availability might disappear. A test of this po-tential interaction on the effect of adding quality infor-mation was in the expected direction, but did not reachconventional levels of significance (/ = 1.1, ;7 >: .10).The analogous test for brand versus quality informationwas in the opposite direction, but again did not reachsignificance (r = - . 1 , p > .50). Thus the competitive-context and infonnation conditions appear to have ex-erted independent effects on the estimates of price elas-ticity.

A final test of the stability of our results concemedwhether they hold across various subgroups within theexperiment. One important contrast compared the urbanstudents with the suburban homeowners. A second com-


pared nondurable products with durable products. Bothof these potential interactions were tested by building theappropriate cross-product terms into the price coeffi-cients. Neither the terms involving competitive-contextinteractions (x^(2) = 1.2, p > .50) nor the terms per-taining to additional information (x^(4) = 5.6, p > .20)were significant. This finding means that the pooling ofheterogeneous groups did not affect results for the crit-ical price contrasts. It might be noted, however, that thegroups differed in terms of their average price sensitiv-ity. For example, homeowners were generally more pricesensitive than students. The statistical tests, however, in-dicate that the change in price sensitivity due to the con-text and information manipulations did not differ acrossgroups. This finding supports the generality of the con-textual fmdings even across groups with different baserates of price sensitivity.

In sum, we found that altering the specification of themultinomial logit model had little effect on the particularhypotheses of interest. This outcome suggests that ourresults are stable over a broad range of conditions.

DISCUSSION

We organize the discussion around the three researchquestions and examine each in an attempt to assemble acoherent account of the data. Throughout, we emphasizea distinction between respondents' treating price as atradeoff against perceived quality (price-verûi'-qualitystrategy) and their treating price as consistent with per-ceived quality (price-w/V/i-quality strategy). Thus, as anexplanatory mechanism, we interpret the experimentalmanipulations as modifying the relative likelihood oftheprice-versus-quality (higher elasticity) and price-with-quality (lower elasticity) orientations.

Result I. Bracketing Produced Relatively GreaterPrice Sensitivity

In the bracketed condition, the experimental brand wassurrounded by competitors. By contrast, in the boundarycondition, the top brand was replaced by the low pricebrand and thus the experimental brand was at the top ofthe price-quality continuum. As expected, this differencein competitive context affected price sensitivity, whichwas relatively higher (lower) for the bracketed (bound-ary) condition. This finding supports the two possibleexplanations mentioned before. First, it supports the ideathat making a price-versus-quality evaluation is moredifficult and has less justification in the bracketed thanin the boundary condition. Second, it supports the ideathat adding an adjacent competitor in the bracketed caseincreases price sensitivity of the experimental brand byincreasing its substitutability in the choice set.

The processing and substitutability explanations affordvery different accounts ofthe result found. Testing thesealtemative explanations requires a study manipulating thenumber of brands offered in a manner that allows sub-stitution effects to be separated from processing effects.

Managerially, our results are intriguing regardless of

the underlying process. They account, in part, for thedesire of firms to be positioned at the high price/highperceived quality boundary (e.g., Chivas Regal or RollsRoyce). The implication for brands on the low price endof the spectrum was not explicitly tested and is thereforeless clear. However, if price sensitivity is greater for alow price brand when a still lower price brand is addedto the field, adding such "decoys" (cf. Huber and Puto1983) may generally increase low-end price sensitivity.This conclusion must remain speculative, however, untilfurther research extends the domain of the bracketing hy-pothesis to lower price altematives.

Result 2. Brand Name Alone Produced RelativelyLower Price Sensitivity Than Quality Ratings Alone

The second result stemmed from contrasting the pricesensitivity of choices based only on additional perceivedquality infonnation with that of choices based only onadditional brand information. The finding of relativelylower price sensitivity in the latter condition was hy-pothesized because the well-known brand names wereexpected to be treated as "chunks" of information. This"chunking" could lead respondents to assume that theprice is reasonable and therefore to pay less attention toit. Indeed, this attentional simplification has a rationalbase beyond the immediate saving in processing time.Specifically, because the items tested were available inthe competitive marketplace, market mechanisms wouldhave provided considerable assurance that items werepriced at appropriate levels.

Our finding of the relatively lower price sensitivity withbrand names alone parallels the development of brandloyalty. For example, McConnell (1968) showed thatsimple usage of identical products could instill differ-ential loyalty and significant price inelasticity. Thus, ourfindings corroborate the effort of many firms to establisha unique brand identity.

To summarize, our finding of relatively lower pricesensitivity due to providing brand names with price in-stead of quality ratings with price corresponds to em-pirical results for brand names in other simulated choicesituations. It is normatively reasonable in a competitivemarketplace. Finally, it supports the practitioner's beliefin the value of established brand identities.

Result 3. Brand Name Plus Quality InformationProduced Relatively Lower Price Sensitivity ThanBrand Name Alone

The third result emerged from contrasting the pricesensitivity of groups given only brand names with thatof groups given both brand names and perceived qualityratings. We found that the latter groups exhibited con-sistently lower price elasticity. Thus the increased po-tential to evaluate price appears to have been overcomein our experiment by the motivation to simplify the de-cision. It is useful, then, to explore why this effect oc-curred and to discuss the degree to which it can be ex-pected to generalize to other choice situations.


The motivational disincentives to evaluate price in theface of additional quality information have several ex-planations. First, it may be that the addition of anv pieceof infonnation has a negative impact on the weight givento the other information. Such an account would positthat price sensitivity depends primarily on the number ofother pieces of information provided. This account couldbe tested by experimentally manipulating both the num-ber and importance of available pieces of information.Alternatively, quality ratings may provide informationthat is difficult to merge with the valuation attached tothe brand name. Finally, a specific factor in the designof the quality ratings may have further lowered elastic-ity. The perceived quality ratings displayed were asso-ciated positively with both subjects' expectations for thebrand and the prices given. This information may havereinforced the positive associations between price andquality and thereby favored a price-with-quality strategy.

Whatever the detailed explanation, the addition ofquality information that could have made a price-ver-sus-quality assessment more accurate appears to havemade it more difficuh. This effect led to the simpler price-with-quality processing wherein the confiiet between priceand quality was effectively ignored. Though this resultis found to hold in the context of our relatively low in-volvement task, the motivational explanation leads to aninteresting prediction in the case of high involvement de-cisions. In that case, the value of additional quality in-formation could overcome the disincentives to processit. Such a hypothesis could be tested by manipulatingthe importance of the decision to the subjects. We thencould ascertain whether the relative lessening of pricesensitivity due to the addition of quality ratings is gen-eral or limited to low involvement decisions.

The managerial implications of a finding that the pro-motion of relative quality information for branded prod-ucts lowers price sensitivity could prove very important,particularly if generalized to high involvement choices.Our results suggest, for example, that providing the cus-tomer with comparative quality information may divertattention from evaluating relative prices. Though this re-sult might merely reinforce the general strategic rec-ommendation that producers should try to get buyers tofocus on quality differences rather than price differ-ences, the data imply that the justification for such a rec-ommendation can have an information-processing as wellas a product-differentiation basis.

Conclusions

The conclusions drawn from this study must be tem-pered by its limitations. Specifically, many relevant as-pects could not be included in the design. For example,the tests of elasticity occurred in a laboratory setting whereboth the desire to simplify the process and the counter-vailing demand to exhibit rationality were certainly greaterthan for choices in the marketplace. Second, respondentswere required to choose within each product category;elasticities might have been substantially greater in the

presence of an income constraint or with an option toreject all of the altematives. Third, the price manipula-tions always affected the top or the second brand andthus the implications of the study for low price/low qualitybrands reamin uncertain. Finally, all price manipulationsand perceived quality ratings adhered to a strict price-quality ordering to avoid dominated altematives (e.g.,an altemative with higher price and lower perceivedquality) that might have lowered price sensitivity.

The preceding limitations indicate areas where the studycan and should be expanded. Within its present confines,however, the major conclusions seem strong. Specifi-cally, differences due to competitive context ox addi-tional infonnation typically brought about large changesin relative levels of elasticity. First, placing a brand atthe boundary of a choice set lowered its relative pricesensitivity. Second, presenting only brand names as op-posed to only quality ratings also lowered the relativeprice elasticity. Finally, adding quality ratings to brandnames had a strong negative impact on relative pricesensitivity. Though these findings are clear within theirlimited context, the most exciting prospect is the furthertesting of these results in different experimental and fieldsettings to establish the domains in which they apply.

REFERENCES

Barnes, James G. (1975), "Factors Influencing Consumers'Reactions to Retail Newspaper Sales," in Proceedings. FallEducators' Conference. Chicago: American Marketing As-sociation, 471-7.

Berkowitz, Eric N. and John R. Walton (1980), "ContextualInfluences on Consumer Price Responses: An ExperimentalAnalysis," Journal of Marketing Research, 17 (August), 349-58.

I>ella Bitta, J. Albert, Kent B. Monroe, and John M. Mc-Ginnis (1981), "Consumer Perceptions of Comparative PriceAdvertisements." Journal of Marketing Research. 18 (No-vember), 416-27.

Gabor. Andre, Clive W. Granger, and Anthony Sowter (1970),"Real and Hypothetical Shopping Situations in MarketResearch," Journal of Marketing Research. 7 (August),355-9.

Hauser, John R. and Steven H. Shugan (1983). "DefensiveMarketing Strategies," Marketing Science. 2 (Fall), 319-60.

Huber. Joel and Christopher Puto (1983), "Market Boundariesand Product Choice: Illustrating Attraction and SubstitutionEffects," Journal of Consumer Research, 10 (June), 21-31.

Jacoby, Jacob, George Szybillo, and Jacqueline Busato-Schach(1977), "Information Acquisition Behavior in Brand ChoiceSituations," Journal of Consumer Research. 3 (March). 209-16.

Lutz, Richard J. and James R. Bettman (1976), "Multiattri-bute Models in Marketing: A Bicentennial Review," work-ing paper. University of California, Los Angeles (Febmary).

McConnell, Douglas J. (1968), "An Experimental Examina-tion ofthe Price-Quality Relationship," Journal of Business(October), 439-44.

Monroe. Kent B. (1973), "Buyers' Subjective Perceptions ofPrice," Journal of Marketing Research. 10 (February), 7 0 -80.


(1976), "The Influence of Price Differences and BrandFamiliarity on Brand Preferences," Journal of ConsumerResearch. 3 (June). 4-49.

and R. Krishnan (1985), "The Effect of Price on Sub-jective Product Evaluation." in Perceived Quality, J. Jacobyand J. Olson, eds. Lexington, MA: Lexington Press, 212-32.

Olson, Jerry C. (1977), "Price as an Informational Cue: Ef-fects on Product Evaluations." in Consumer and IndustrialBuying Behavior. A. Woodside, J. Sheth, and P. Bennett,eds. New York: North-Holland Publishing Company.

Parsons, Leonard J. and Randall L. Schultz (1976), MarketingModels and Economic Re.search. New York: Eisevier.

Peterson, Robert A. and William R. Wilson (1985), "Per-

ceived Risk and Price-Reliance Schema as Price-Perceived-Quality Mediators." in Perceived Quality. J. Jacoby and J.Olson, eds. Lexington MA: Lexington Press.

Sherif, Muzafer and Carl Hovland (1953), "Judgmental Phe-nomena and Scales of Attitude Measurement: Placement ofItems with Individual Choice of Number of Categories."Journal of Abnormal and Social Psychology. 48, 135-41.

Simon, Herbert A. (1974), "How Big is a Chunk." Science.183, 482-8.

Stafford. James E. and Ben M. Enis (1969), "The Price-Qual-ity Relationship: An Extension," Journal of Marketing Re-search, 6 (November). 456-8.

Wilks. Samuel S. (1962). Mathematical Statistics. New York:John Wiley & Sons. Inc.

of Health Care Marketing

ournalJOURNAL OF HEALTH CARE MARKETING Published in winter, spring,

summer, and fall, this journal offers an in-depth look at the prosperingfield of health care marketing. Features include "Commentary On...," adiscussion forum for health care marketers; "Mtnicase," featuring "case"manuscripts from practitioners; and a "Resource Guide" to a variety ofpublications, meetings/seminars, and organizations. In addition, book andarticle reviews, guest ediforials, abstracts, and letters-to-the-editor appearon a regular basis.

Joumal of Health Care Marketing is available for individual as well ascorporate (library, school, business, etc.) use.

•Rates as follows: (effective 1/86—rates subject to change)AMA Members Nonmembers Corporate

Journal of Health Oare Marketing $25 $35 $45

Copies may be purchased on an individual basis as weli as onan annual subscription basis.

For subscription information, contact:The American Marketing Association • Publication Orders Department

250 S. Wacker Dr., Suite 200 • Chicago, IL 60606-5819or call 312/648-0536.

Documents

Context Price Sensitivity