Three Essential Analytical Techniques for the Behavioral Marketing Researcher: Median Splits, Mean-Centering, and Mediation Analysis

Published on Dec 1, 2015
Dawn Iacobucci51
Estimated H-index: 51
(Vandy: Vanderbilt University),
Deidre Popovich6
Estimated H-index: 6
(TTU: Texas Tech University)
+ 2 AuthorsFrank R. Kardes50
Estimated H-index: 50
(UC: University of Cincinnati)
For the behavioral marketing scholar, experimentation and the analysis of variance are among the most important and frequently relied upon tools of the trade, and many useful texts exist to guide researchers on these topics. This monograph is intended to be a supplemental resource and a helpful guide for conducting three essential analytical techniques that are also frequently useful to the behavioral researcher: (1) we discuss the practice of conducting a median split on a continuous variable to facilitate communication clarity. (2) We demonstrate the practice of centering variables about their means prior to creating product terms to reflect interaction effects in a moderated multiple regression model. (3) We discuss the practice of a mediation analysis to test for the relative impact of direct and indirect effects of predictors on dependent variables.
📖 Papers frequently viewed together
2010ICEE: International Conference on E-Business and E-Government
1 Citations
1 Citations
#1Dawn Iacobucci (Vandy: Vanderbilt University)H-Index: 51
#2Matthew J. Schneider (NU: Northwestern University)H-Index: 9
Last. Georgios A. Bakamitsos (Stetson University)H-Index: 6
view all 4 authors...
There seems to be confusion among researchers regarding whether it is good practice to center variables at their means prior to calculating a product term to estimate an interaction in a multiple regression model. Many researchers use mean centered variables because they believe it’s the thing to do or because reviewers ask them to, without quite understanding why. Adding to the confusion is the fact that there is also a perspective in the literature that mean centering does not reduce multicoll...
84 CitationsSource
#1Dawn Iacobucci (Vandy: Vanderbilt University)H-Index: 51
#2Steven S. Posavac (Vandy: Vanderbilt University)H-Index: 27
Last. Deidre Popovich (Emory University)H-Index: 6
view all 5 authors...
Abstract Some behavioral researchers occasionally wish to conduct a median split on a continuous variable and use the result in subsequent modeling to facilitate analytic ease and communication clarity. Traditionally, this practice of dichotomization has been criticized for the resulting loss of information and reduction in power. More recently, this practice has been criticized for sometimes producing Type I errors for effects regarding other terms in a model, resulting in a recommendation of t...
180 CitationsSource
#1Dawn Iacobucci (Vandy: Vanderbilt University)H-Index: 51
#2Steven S. Posavac (Vandy: Vanderbilt University)H-Index: 27
Last. Deidre Popovich (TTU: Texas Tech University)H-Index: 6
view all 5 authors...
In this rebuttal, we discuss the comments of Rucker, McShane, and Preacher (2015) and McClelland, Lynch, Irwin, Spiller, and Fitzsimons (2015). Both commentaries raise interesting points, and although both teams clearly put a lot of work into their papers, the bottom line is this: our research sets the record straight that median splits are perfectly acceptable to use when independent variables are uncorrelated. The commentaries do a good job of furthering the discussion to help readers better d...
147 CitationsSource
#1Derek D. Rucker (NU: Northwestern University)H-Index: 52
#2Blakeley B. McShane (NU: Northwestern University)H-Index: 18
Last. Kristopher J. Preacher (Vandy: Vanderbilt University)H-Index: 62
view all 3 authors...
We comment on Iacobucci, Posavac, Kardes, Schneider, and Popovich (2015) by evaluating the practice of discretizing continuous variables. We show that dichotomizing a continuous variable via the median split procedure or otherwise and analyzing the resulting data via ANOVA involves a large number of costs that can be avoided by preserving the continuous nature of the variable and analyzing the data via linear regression. As a consequence, we recommend that regression remain the normative procedu...
85 CitationsSource
#1Gary H. McClelland (CU: University of Colorado Boulder)H-Index: 28
#2John G. Lynch (CU: University of Colorado Boulder)H-Index: 50
Last. Gavan J. Fitzsimons (Duke University)H-Index: 43
view all 5 authors...
Considerable prior statistical work has criticized replacing a continuously measured variable in a general linear model with a dichotomy based on a median split of that variable. Iacobucci, Posovac, Kardes, Schneider, and Popovich (2015-in this issue) defend the practice of “median splits” using both conceptual arguments and simulations. We dispute their conceptual arguments, and we have identified technical errors in their simulations that dramatically change the conclusions that follow from th...
57 CitationsSource
#1Jamie DeCoster (UA: University of Alabama)H-Index: 41
#2Anne-Marie R. Iselin (University of Pittsburgh)H-Index: 10
Last. Marcello Gallucci (University of Milano-Bicocca)H-Index: 31
view all 3 authors...
Despite many articles reporting the problems of dichotomizing continuous measures, researchers still commonly use this practice. The authors’ purpose in this article was to understand the reasons that people still dichotomize and to determine whether any of these reasons are valid. They contacted 66 researchers who had published articles using dichotomized variables and obtained their justifications for dichotomization. They also contacted 53 authors of articles published in Psychological Method...
271 CitationsSource
#1David P. MacKinnon (ASU: Arizona State University)H-Index: 77
#2Matthew S. Fritz (ASU: Arizona State University)H-Index: 10
Last. Chondra M. Lockwood (ASU: Arizona State University)H-Index: 8
view all 4 authors...
This article describes a program, PRODCLIN (distribution of the PRODuct Confidence Limits for INdirect effects), written for SAS, SPSS, and R, that computes confidence limits for the product of two normal random variables. The program is important because it can be used to obtain more accurate confidence limits for the indirect effect, as demonstrated in several recent articles (MacKinnon, Lockwood, & Williams, 2004; Pituch, Whittaker, & Stapleton, 2005). Tests of the significance of and confide...
1,064 CitationsSource
#1Raj Echambadi (College of Business Administration)H-Index: 17
#2James D. Hess (UH: University of Houston)H-Index: 25
The cross-product term in moderated regression may be collinear with its constituent parts, making it difficult to detect main, simple, and interaction effects. The literature shows that mean-centering can reduce the covariance between the linear and the interaction terms, thereby suggesting that it reduces collinearity. We analytically prove that mean-centering neither changes the computational precision of parameters, the sampling accuracy of main effects, simple effects, interaction effects, ...
323 CitationsSource
#1Dawn Iacobucci (UPenn: University of Pennsylvania)H-Index: 51
#2Neela Saldanha (UPenn: University of Pennsylvania)H-Index: 2
Last. Xiaoyan Deng (UPenn: University of Pennsylvania)H-Index: 10
view all 3 authors...
In this paper, we suggest ways to improve mediation analysis practice among consumer behavior researchers. We review the current methodology and demonstrate the superiority of structural equations modeling, both for assessing the classic mediation questions and for enabling researchers to extend beyond these basic inquiries. A series of simulations are presented to support the claim that the approach is superior. In addition to statistical demonstrations, logical arguments are presented, particu...
789 CitationsSource
#1Patrick RoystonH-Index: 93
#2Douglas G. Altman (University of Oxford)H-Index: 268
Last. Willi SauerbreiH-Index: 61
view all 3 authors...
In medical research, continuous variables are often converted into categorical variables by grouping values into two or more categories. We consider in detail issues pertaining to creating just two groups, a common approach in clinical research. We argue that the simplicity achieved is gained at a cost; dichotomization may create rather than avoid problems, notably a considerable loss of power and residual confounding. In addition, the use of a data-derived 'optimal' cutpoint leads to serious bi...
1,373 CitationsSource
Cited By2
#1Mark van de Ruit (TU Delft: Delft University of Technology)H-Index: 6
#2Michael James Grey (RMIT: RMIT University)H-Index: 26
Abstract Background A trend in the non-invasive brain stimulation literature is to assess the outcome of an intervention using a responder analysis whereby participants are di- or trichotomised in order that they may be classified as either responders or non-responders. Objective Examine the extent of the Type I error in motor evoked potential (MEP) data subjected to responder analyses. Methods Seven sets of 30 MEPs were recorded from the first dorsal interosseous muscle in 52 healthy volunteers...
9 CitationsSource
#1Jiaxiu He (UWO: University of Western Ontario)H-Index: 9
#2Xin Wang (UWO: University of Western Ontario)H-Index: 7
Last. David J. Curry (UC: University of Cincinnati)H-Index: 16
view all 3 authors...
Abstract Statistical tests for mediation in consumer research typically use a regression coefficients (RegCoeff)-based framework. We present a new test based on likelihood ratio principles to complement the RegCoeff approach. We compare the new test's performance to conventional methods that use PROCESS, MEDIATION, and extensions thereof to categorical measures. Our tests address situations in which the assumptions of current methods tend not to hold; that is when one or more variables may be ca...
3 CitationsSource