Identifying Causal Marketing Mix Effects Using a Regression Discontinuity Design

Abstract

We discuss how regression discontinuity designs arise naturally in settings where firms target marketing activity at consumers, and illustrate how this aspect may be exploited for econometric inference of causal effects of marketing effort. Our main insight is to use commonly observed discontinuities and kinks in the heuristics by which firms target such marketing activity to consumers for nonparametric identification. Such kinks, along with continuity restrictions that are typically satisfied in Marketing and Industrial Organization applications, are sufficient for identification of local treatment effects. We review the theory of Regression Discontinuity estimation in the context of targeting, and explore its applicability to several Marketing settings. We discuss identifiability of causal marketing effects using the design and show that consideration of an underlying model of strategic consumer behavior reveals how identification hinges on model features such as the specification and value of structural parameters as well as belief structures. We emphasize the role of selection for identification. We present two empirical applications: the first, to measuring the effect of casino e-mail promotions targeted to customers based on ranges of their expected profitability; and the second, to measuring the effect of direct mail targeted by a B2C company to zip-codes based on cutoffs of expected response. In both cases, we illustrate that exploiting the regression discontinuity design reveals negative effects of the marketing campaigns that would not have been uncovered using other approaches. Our results are nonparametric, easy to compute, and control for the endogeneity induced by the targeting rule.

Structural Workshop Paper—Discrete-Choice Models of Consumer Demand in Marketing

“…As examples, consider a common marketing problem of measuring the joint lift, or incremental effect, of prices and promotions. An example of an experimental approach to this problem is Fong et al (2010), an example of a nonparametric approach to estimating these effects in the context of nontargeted promotions is Briesch et al (2010), and an example of a nonparametric approach in the context of targeted promotions is Hartmann et al (2011). This literature is often referred to as the "reduced-form, causaleffects" class because of its emphasis on measuring Downloaded from informs.org by [141.241.43.86] on 09 February 2015, at 05:09 .…”

“…This kind of test has no power if restrictions from the supply model are used to estimate demand parameters. One exception is a situation when the exact marketing-mix allocation rule is known-then, there is no misspecification, and incorporating supply-side restrictions improves efficiency while adding little bias (e.g., Hartmann et al 2011 exploit the exact rule firms use to target promotions to consumers).…”

“…Other applications include nonparametric controls for selection and endogeneity concerns. Two examples include Ellickson and Misra (2011), who outline methods for controlling for selectivity nonparametrically when observed data are outcomes of discrete games (e.g., demand in a social interactions setting), and Hartmann et al (2011), who use kinks in firm's targeting rules to nonparametrically control for the endogeneity of targeted marketing in demand under database marketing. This is a fast-growing, frontier area of research in the field.…”

Television Advertising and Online Shopping

“…Systematic differences between the pre-and post-windows are attributed to the ad insertion. The identification strategy is similar to the regression discontinuity approach of Hartmann et al (2011).…”

Sparse Regularization in Marketing and Economics

“…Quantitative variables of interests include lift of non-targeted promotions (see [8]), targeted promotions (see [11]), and joint lift of price and promotions predictions (see [9] and [16]).…”