MARKETING
A. Montgomery Ward Professor of Marketing
Lakshman Krishnamurthi is the Montgomery Ward Distinguished Professor of Marketing. He has been a faculty at Kellogg since 1980, having earned degrees in engineering from IIT, Madras, an MBA from LSU, an MS in statistics and a Ph.D. in marketing from Stanford University. He served as the chairman of the marketing department from 1993-2004.
At Kellogg, Professor Krishnamurthi teaches Marketing Strategy & Pricing in a variety of programs. Professor Krishnamurthi is also the Academic Director of the executive program on Pricing Strategies & Tactics, a position he has held since 1995. In addition, he teaches a multivariate statistics course in the Ph.D. program. He was voted “teacher of the year” for core courses in the Kellogg Executive MBA Program (EMP 63), 2006, voted “teacher of the year” by the second graduating class of the joint Kellogg-Hong Kong University of Science & Technology Executive Master’s program in 2000, and was a finalist for the award in 2002. He received the Sidney Levy award for teaching excellence in the MBA program at Kellogg in 1999, 2001, 2003 and 2007, and has been awarded several other teaching commendations.
Professor Krishnamurthi has also won many awards for his research publications including the Paul Green award and the Donald Lehmann award for best paper in the Journal of Marketing Research; the John D.C. Little award for best paper in Marketing Science; and was a finalist for the William O'Dell Award from the American Marketing Association. He serves on the editorial board of Marketing Science and the Journal of Marketing Research. He is a member of the Institute of Management Sciences and the American Marketing Association.
In addition to his teaching and research activity, Professor Krishnamurthi has consulted for numerous companies including Harcourt Publishing, Motorola, the Chicago Tribune, ZS Associates, Eastman Chemical, and several others. He has also conducted executive education seminars for DuPont, Microsoft, Abbott, ExxonMobil, Johnson & Johnson (Ethicon, Ethicon Endo, Ortho Clinical Diagnostics, ASP), ThyssenKrupp Elevators, British Petroleum, Ford Motors, Merck KgaA, Novartis, Wolters Kluwer, Honeywell, Seminarium (Latin America), Peninsula Hotels, Chicago Tribune, Union Camp, Xerox, Coca Cola, Motorola, Procter & Gamble (Latin America), and others.
Business to Business Markets
Consumer Behavior
Data Analysis
International Marketing
Marketing Management
Marketing Research
Marketing Strategy/Planning/Policy
New Product Development
- Recent Media Coverage
BNET: Improve Sales by Monitoring Online Feedback - 6/1/2009
MIT Sloan Management Review: Using online discussions to predict sales - 5/12/2009
Open Forum: How to Use the Predictive Value of Buzz - 5/12/2009
The Mint (Dow Jones publication in India): Asian century: new challenges - 7/14/2008
See all Kellogg in the Media
- Recent Kellogg News
The Asian century - 7/28/2008
See all Kellogg News
The effect of reference price on brand choice decisions has been well documented in the literature. Researchers, however, have differed in their conceptualizations and, therefore, in their modeling of reference price. In this article, we evaluate five alternative models of reference price of which two are stimulus based (i.e, based on information available at the point-of-purchase) and three that are memory based (i.e., based on price history and/or other contextual factors). We calibrate the models using scanner panel data for peanut butter, liquid detergent, ground coffee, and tissue. To account for heterogeneity in model parameters, we employ a latent class approach and select the best segmentation scheme for each model. The best model of reference price is then selected on the basis of fit and prediction, as well as on the basis of parsimony in cases where the fits of the models are not very different. In all four categories, we find that the best reference price model is a memory-based model, namely, one that is based on the brand's own price history. In the liquid detergent category, however, we find that one of the stimulus-based models, namely, the current price of a previously chosen brand, also performs fairly well. We discuss the implications of these findings.
Promotions are being used with increasing frequency by manufacturers facing highly competitive markets, which is causing concern among some marketers who feel that frequent promotions can hurt a brand. Nevertheless, there is little empirical evidence to either support or dispel such fears. To fill this gap in the literature, the authors propose a brand choice model that provides an estimate of the dynamic effects of promotions on loyalty to the brand and customers' sensitivity to the price of the brand, and measures whether promotional purchases reinforce or reduce subsequent response to similar promotions. They estimate a random effects heteroskedastic covariance probit time-varying parameter model on household scanner panel data from the liquid detergent category. Their results indicate that increased purchases using coupons erode brand loyalty and increase price sensitivity. In addition, the authors find that the effect of features and displays on brand choice is reinforced by prior feature and display purchases, respectivly, as well as by feature and display purchases associated with price cuts. Future research directions are also discussed.
There is substantial evidence for variation in price sensitivity of products across stores and chains. Understanding the relationships between price sensitivity and promotional variables (such as price cut, feature advertising, and display), and between price sensitivity and pricing policy (Everyday Low Pricing [EDLP] and High Low Pricing [HLP]) is particularly important to retailers. We develop hypotheses on the relationships between regular price elasticity and retailer promotional variables, and between regular price elasticity and retailer pricing policy. We test these hypotheses by analyzing the variation of regular price elasticity of a frequently purchased consumer packaged brand across stores, both within and across chains, through a multistage regression analysis. In the first stage of our analysis, we use a mixed double-log model to estimate the sales response function for the brand in each store using time series data. In the second stage, we explain the differences in the estimated regular price elasticities across stores within a chain by a process function model. In the final stage, the differences across all stores and chains are explained through an aggregate process function model. We extend the literature by separating regular (long-run) price elasticity from promotional (short-run) elasticity, and by studying the influence of both strategic and tactical retailer variables on regular price elasticity in a single framework within and across chains. Our results for the brand analyzed show that a higher level of display and feature advertising together is associated with a lower level of regular price elasticity in EDLP stores and that an EDLP policy is associated with a higher level of regular price elasticity, whereas an HLP policy is related to a lower level of regular price elasticity.
The authors use a simulation that explores the same factors used by Wildt (1993), but provides results that refute several of the findings reported in that study. The authors maintain that, under conditions of multicollinearity, the Equity estimator provides estimates that are typically closer to the true parameters than the ordinary least squares and Ridge estimates.
How do price changes by one brand affect the choice of competing brands? Such inter-brand effects may depend on the specific strategy followed by a firm. For example, a firm may target a particular brand to exploit its vulnerability or to avoid direct competition with other brands. Or a firm may design its pricing strategy aimed at reducing cannibalization of its own brands. Previous studies have utilized standard logit models to investigate inter-brand ejects. However, these models impose constraints on price elasticities as a consequence of independence of irrelevant alternatives (HA) assumptions. As a result, estimated own- and cross-elasticities reflect the restrictive assumptions of the model and may not provide an accurate description of the hinds of asymmetric competition among brands noted previously. Though existing market share models at the aggregate level do capture such competitive asymmetries, most disaggregate level logit choice models do not capture such asymmetries in a satisfactory manner. In this article a generalized logit (or mother logit) model is used to estimate unique interbrand response parameters to capture asymmetry. This methodology, drawn from the econometrics literature, overcomes the necessity of making a priori assumptions of competitive patterns and instead can be used to identify competitive patterns as they exist in the market place. In analyzing brand choice data from three product classes, ILA is violated in all three cases to varying degrees. The cross-price elasticities are used to draw managerial implications for brand and product line management.
Hill and Cartwright examined the relative performance of ordinary least squares, the Equity estimator, and the Stein estimator on four scanner data sets. Based on this limited simulation, they claimed that the Stein estimator is an attractive alternative to the Equity estimator. In this reply, we demonstrate that their conclusions are not well founded and that a more comprehensive evaluation indicates that the Equity estimator dominates the Stein estimator in wide regions of the parameter space.
There are many products which are repeatedly purchased by consumers. In such cases it is likely that choice history, that is the sequence of choices made in the past, as well as marketing variables affect subsequent choice decisions. Attempts to model the effects of choice history have been generally based on the inclusion of variables that represent brand loyalty and/or variety seeking behavior. In this paper we present a model of dynamic choice behavior which is more general and incorporates four characteristics. The first characteristic labeled preference reinforcement and preference reduction represents loyalty and variety seeking. The second is the short-term reluctance of a consumer to move from the current brand (inertia) or the willingness to move to another brand (mobility). The third characteristic captures the effect of repetitive consumption (the long term effect) on inertia and mobility. The fourth characteristic incorporates the similarity or dissimilarity of choice alternatives. This is important in a dynamic model because choice on the current purchase occasion can be affected by whether a similar or dissimilar alternative was chosen on the previous occasion. Similarities of alternatives are represented in terms of distances. The effect of price on choice behavior is also modeled. Individual-level purchase data from a consumer panel are used to estimate a covariance probit and an independent probit specification of the model. From a substantive perspective the model gives interesting insights into the dynamics of choice behavior. The model predicts switches better than a benchmark model which incorporates only loyalty. In addition, it is superior to three benchmark models in overall predictive ability.
The study investigates whether consumers exhibit asymmetry (i.e., different sensitivity) to negative ("loss") and positive ("gain") differences between the reference price and the purchase price in brand choice and purchase quantity decisions. Using panel data for two frequently purchased products with three brands in each product category, we find that consumers loyal to a brand ("loyals") respond to gain and loss with the same sensitivity in brand choice decisions. However, consumers not loyal to any brand ("switchers") respond more strongly to gains than to losses. In purchase quantity decisions, brand-loyal consumers are found to respond asymmetrically to gains and losses, but the direction of the asymmetry depends on whether the decision is made before or after the household inventory reaches a stock-out level (i.e., the level at which the household inventory needs to be replenished). When the decision is made after a stock-out, brand-loyal consumers are more responsive to a gain in the price of their favorite brand than to a loss. In contrast, when the quantity decision is made before a stock-out, loyals are more sensitive to a loss than to a gain. In only two of the six brands examined do we find evidence of asymmetry in switchers' quantity decisions. In both cases, switchers respond more strongly to a price loss than to a gain, regardless of whether the purchase decision is made before or after a stock-out.
This empirical paper explores the relationship between consumer brand preference or loyalty and price elasticity in purchase behavior. This behavior is conceptualized as resulting from two distinct but related decisions, namely a brand choice decision and a purchase quantity decision. We argue that loyal consumers will be less price sensitive in the choice decision than nonloyal consumers. However, this direction is expected to be reversed in the quantity decision with loyal consumers expected to be more price sensitive than nonloyal consumers. We model the choice and quantity decisions jointly using the limited dependent variable framework described in Krishnamurthi and Raj. The data used are diary panel data on a frequently purchased product class from BURKE and caffeinated ground coffee scanner data from IRI. We show that loyals are less price sensitive than nonloyals in the choice decision but more price sensitive in the quantity decision. Managerial implications of the differing elasticities are discussed.
Multicollinearity often hampers the estimation of the "independent" effects of the marketing mix variables in sales response models. In a previous study, the authors recommended the use of the equity estimator for estimating linear models in the presence of multicollinearity. In this article, they evaluate the performance of equity, ridge, OLS, and principal components estimators in estimating response functions for 36 pharmaceutical products. Overall, equity outperforms the other three estimators on criteria such as estimated bias, variance, and face validity of the estimates. The four estimators have similar levels of predictive accuracy. The authors also present some managerial implications for resource allocation in the pharmaceutical industry.
We compare the part-worth model against alternative specifications with linear and nonlinear functions for continuous attributes. We use four criteria: expected error variance of the model, expected mean squared error for average preference predictions, expected mean squared error for individual preference predictions, and a measure of the quality of parameter estimates. Although the part-worth model is generally superior to the alternatives, the expected validity of conjoint results can he improved with idiosyncratic functional forms. Continuous functions also offer an opportunity to the analyst to impose constraints on the parameter estimates.
It is well known that the range of attribute variation used in a conjoint design influences the inferred attribute importance. However, even if the range is held constant, the addition of intermediate levels can increase this importance. In this paper we show why the problem occurs for rank-order preferences. The results from an experimental study confirm the existence of a systematic influence due to the number of (intermediate) levels. Surprisingly, the problem is equally strong when rating scale preferences are collected. Several possible solutions are suggested.
This paper presents a case study to show how a control group can be used to obtain more accurate estimates of the impact of interventions. Intervention analysis using the ARIMA time series method is applied in an experimental design context using multiple input transfer function analysis. The study combines the analytic rigor of time series analysis with the careful controls provided by an experiment involving a test and control series. The data are from a field experiment with test and control panels connected to a split-cable TV system.
Many consumer decisions involve a discrete choice and a continuous outcome. Examples of such decisions are whether to own a home or rent one and how much to spend, which brand of orange juice to buy and how many ounces to buy. In cases like these, the choice decision is typically modeled separately, say, using a logit model and the continuous outcomes modeled separately using regression analysis. However, the continuous outcomes may not be independent of the discrete choice and vice versa, and modeling the two decisions independently can lead to inefficient choice parameter estimates and biased and inconsistent regression parameter estimates. In this paper, we present a methodology from the limited-dependent variable literature to model the dependence between the choice and quantity decisions. Our substantive interest is in the role of price in the choice and quantity decisions. When choosing among alternatives, we argue that consumers consider prices of all the competitive brands. In the quantity decision on the other hand, only the price of the chosen alternative is expected to impact how much of the alternative is purchased. The analysis of three brands, using disaggregate level panel data, strongly supports our hypothesis about the role of competitive prices in the choice and quantity decisions.
The objective of this research is to develop viable approaches to modeling joint decisions. Using conjoint-analysis-type preference data, three methods are developed to combine individual preferences to approximate joint preferences and predict joint decisions. The first is an equal weighting model, which is a simple average of individual members' part-worth utilities. The second is a relative influence model, which combines individual utility functions using a measure of derived influence. The third is a conflict resolution model, which combines utility functions using a measure of conflict. In addition to these three combination models, individual member models and a joint model based on the joint preferences are available. The application area in which the models are operationalized is family decision making. The decision involves choice of a job by MBA students and spouses at a major private university. The models are first calibrated using preference data on hypothetical jobs from MBAs, spouses, and couples and then evaluated on their ability to predict the actual job chosen.
We introduce a new biased estimator called the Equity estimator to estimate parameters of linear models in the presence of multicollinearity. We then employ a comprehensive simulation methodology to evaluate OLS, the Equity estimator, and a particular form of the Ridge estimator which has received some attention in the statistical literature. We show that both Equity and Ridge perform significantly better than OLS on a variety of performance measures over most of the parameter space. The Equity estimator does much better than the Ridge estimator on the squared error criterion. On other criteria, the Equity estimator performs marginally better than Ridge. In particular, Equity does well when the degree of multicollinearity among the explanatory variables is high and/or the Rsup2/sup of OLS for the sample is less than 0.7, conditions frequently encountered in marketing research. We discuss how the results of this study can be used by marketing researchers in choosing an appropriate estimator for their particular application.
The authors study the buildup effect of increased advertising using time series intervention analysis. The data are from an ADTEL field experiment with test and control panels connected to a split-cable TV system. Use of the control series in the analysis depends on the nature of the relevant external factors. If these factors are purely unmeasured, the control series is included as a covariate, resulting in a single-input transfer function-intervention model. If there are also measured external factors, a multiple-input model results. A careful analysis of the increased advertising shows that the buildup effect is immediate with a duration of the order of the purchase cycle.
The authors investigate how increased advertising affects consumer price sensitivity. First, a conceptual framework integrating the role of advertising content is presented. Next, a methodology for studying the impact of advertising on consumer price sensitivity to brand purchase quantity and consumption is developed. Analyses of diary panel data for an established, frequently purchased brand from an ADTEL advertising field experiment clearly demonstrate that increased advertising lowers price sensitivity. Further, this effect is strong in the high price sensitivity segment for purchase quantity and consumption. In the low price sensitivity segment the effect is marginal. Additional support for these results was obtained by choosing different cutoff points for high sensitivity segmentation.
This article offers an approach to joint decision making that is an extension of the key informant approach. MBAs and their household partners were either prompted to or told not to take each other into consideration in stating their preferences for MBA jobs. The household partners came significantly closer to representing the joint position as a result of the prompt than did the MBAs; they were also more accurate in their perception of the other group's preferences. It is concluded that a key informant's awareness that a decision is joint does not by itself enhance his/her accuracy in predicting joint preferences; rather, accuracy depends on a key informant's knowledge of the other's preferences.
In a recent paper, Currim, Weinberg, and Wittink (1981) noted that attribute importance weights inferred from conjoint analysis results may be influenced by the number of levels on which an attribute is defined.
For a main-effects part-worth model, attribute importance is typically computed by taking the difference between the "best" and "worst" levels' estimated utilities. Currim, Weinberg, and Wittink argued that the minimum weight obtainable for a three-level attribute is higher than the rain hum weight for an attribute with only two levels, if forced tank order or equivalent preference judgments are collected. However, they recognized that the empirical finding could also be due to a "managerial" reason (attributes perceived as critical are defined on a greater number of levels) or a "psychological" phenomenon (respondents may pay more attention to attributes as the number of levels increases). In this paper, we investigate in more detail the possible relation between derived importances and number of attribute levels.
In non-vertically integrated channels manufacturers rely on incentives to influence the prices offered by channel partners such as wholesalers and retailers. One of the most common channel incentives is a trade promotion. Packaged goods manufacturers spend in excess of $75 billion annually on trade promotions with the goal of offering temporary discounts to end-consumers (Cannondale 2001). A common metric used by practitioners and academics to evaluate trade promotion effectiveness is pass-through. Although the effectiveness of trade promotions has been debated for decades, empirical research on the topic is scarce. Extant research has relied on data from one or two supermarket chains in a single market and managerial surveys. For this study we assemble a unique dataset containing information on prices, quantities, and promotions throughout the entire channel in a category. The data enable us to extend the empirical literature on pass-through in two important ways. First, we investigate how it varies across more than 1000 retailers in over 30 states. Second, we study pass-through at multiple levels of the distribution channel, allowing us to assess how channel intermediaries -- such as wholesalers and brokers -- influence pass-through. We find that the median pass-through elasticities are 0.75, 0.58, and 0.39, for the wholesaler, retailer, and total channel respectively. Thus, a 10% reduction in manufacturer price results in a 3.9% reduction in consumer price. At each of the levels in the channel large variances in the estimates of pass-through elasticities are observed. We therefore argue that average pass-through elasticities are only of limited tactical value to manufacturers. For example, the pass-through for a specific chain in California may be 0.98 while it is equal to 0.42 for the same chain in Nevada. The average of these values holds little meaning for a firm trying to evaluate and improve trade-promotion effectiveness. We investigate the sources of pass-through variation using various measures of cost and competition.
This course counts toward the following majors: Marketing, Marketing Management
This course presents an integrative, dynamic view of competitive brand strategy. It focuses on understanding, developing and evaluating brand strategies over the life of a product market. A framework for developing marketing strategies that yield a distinctive competitive advantage based on customer and competitor analysis will be presented and applied in various situations throughout the course. Topics include strategies for pioneering brands, strategies for late entry, growth strategies, strategies for mature and declining markets, and defensive marketing strategies. Material is presented using a mix of cases, lectures and a computer simulation game called MARKSTRAT.
Prerequisite: MKTG-430.
This seminar confronts students with significant problems, issues and theories at the leading edge of the marketing field. Presentations and discussions are designed to stimulate thinking on important areas of research and the development of new theoretical viewpoints.
Marketing Strategy (MKTGX-466-0)
PHONE: 847-467-1286
FAX: 847-491-2498
Jacobs Center Room 483