Perils of Using OLS to Estimate Multimedia Communications Effects

Abstract

Companies invest millions of dollars in various forms of marketing communications to impact customers' awareness, attitudes, purchases and, ultimately, profitability. An important question for marketers and shareholders alike is: what effects do marketing investments have on market performance? To assess these effects, marketers estimate marketing-mix models by using regression analysis. However, we show that the estimation of marketing-mix models via regression analysis (i.e., ordinary least squares, OLS) yields severely biased estimates of marketing effects. Specifically, for our examples, OLS overstates the impact of media advertising on sales by 34% to 147%. In other words, an estimated media effect can be twice as much as what it really is. The synergy and carryover effects are under-estimated by 28% and 48%, respectively. To mitigate such severe biases, we present an alternative approach, called the Wiener-Kalman filter, which provides reasonable estimates that are much closer to the true parameters than the corresponding OLS estimates. In addition, we analyze Corolla brand's multimedia campaign and furnish results based on marketplace data that corroborate the simulation findings. Finally, we discuss both the implications of these results for brand managers and the opportunities that lie ahead for advertising researchers.

Keywords: advertising, multimedia communication, OLS, Wiener-Kalman filtering