<- file stat 97modmed.html ->

Moderation/mediation

=======================Paul Bernhardt, 12 Apr 1997==========ssc Message-ID: <199704121841.MAA25199@gos.oz.cc.utah.edu> From: Paul Bernhardt <Paul.Bernhardt@m.cc.utah.edu> Subject: Re: Moderating effect on relationship >In the vernacular that seems to have become adopted since the Baron and >Kenney paper, a moderator is indeed modeled as the interaction between the >two predictors in question. However, the diagram shown in the original >post would actually represent mediation rather than moderation. Mediation >can be assessed with traditional regression models or with path analytic >techniques. For the rationale and how-to's, I'd point the original poster >to the Baron and Kenney paper that appeared in the psychology literature >in the mid-eighties titled something like, The Mediator-Moderator >Distinction (sorry but I don't have the cite right on hand). The cite is: Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51 (6), 1173-1182.

"moderator" and "mediator"

=======================Michael Babyak, 04 Apr 1997==========ssc,sse,ssm From: mbabyak@acpub.duke.edu (Michael Babyak) Subject: Re: Moderating effect on relationship Message-ID: <5i1kma$snv@newsgate.duke.edu> Sylvia J. Hysong (shysong@ruf.rice.edu) wrote: : In article <3341A808.E0C@psy.cuhk.edu.hk>, hhymok@psy.cuhk.edu.hk wrote: : > I'm analysing the relationship between an independent x and a dependent : > y. Besides, I also have a variable z. I want to test whether there is : > any moderating effect of z on the relationship of x predicting y. : > z : > | : > V : > x --------> y : > : > Is anyone can tell me how to do that ? by using SPSS ? : > : > Thanks, : > Helen : A moderator is simply another word for interaction. Given that, any : program that can do ANOVA can give you what you want. simply add z as an : independent variable, and you will get an ANOVA table with 3 effects: the : main effect of x on y, the main effect of z on y, and the x by z : interaction. You test for significance just like you do for any other : main effect. If your independent variables are continuous instead of : categorical, you can simply make an "interaction variable" by multiplying : the values of x by their corresponding values of z (so you get an x*z : term), and regress that onto y. : -- : Sylvia J. Hysong In the vernacular that seems to have become adopted since the Baron and Kenney paper, a moderator is indeed modeled as the interaction between the two predictors in question. However, the diagram shown in the original post would actually represent mediation rather than moderation. Mediation can be assessed with traditional regression models or with path analytic techniques. For the rationale and how-to's, I'd point the original poster to the Baron and Kenney paper that appeared in the psychology literature in the mid-eighties titled something like, The Mediator-Moderator Distinction (sorry but I don't have the cite right on hand). --

Moderated regression

=======================Jeremy Miles, 21 Feb 1997==========spss Message-ID: <3.0.1.32.19970221112204.00aa6968@unix1> From: Jeremy Miles <J.N.V.Miles@DERBY.AC.UK> Subject: [long] Re: Test of moderator effect At 09:38 21/02/97 +1000, B.Ong wrote: > >I agree that both procedures are equivalent >tests if we are only interested in AB. If we are also interested in the >main effects of A and B (as in factorial designs), then which procedure >should we use (or am I revisiting the choice between nested and 'unnested' >factorial ANOVA designs)? > The issues in moderated regresson get a bit complex to cover in an email, here are some references which (might? should? could?) help: Anderson, L.E., Stone-Romero, E.F. and Tisak, J. 1996 A comparison of bias and mean squared error in parameter estimates of interaction effects: moderated multiple regression versus errors-in-variables regression Multivariate Behavioural Research 31 (1): 69-94 Talks about power, doesn't look at standard sort of regression. Arnold, HJ 1982 Moderator Variables: a classification of conceptual, analytic and psychometric issues Organisational Behaviour and human performance 29:143-174 General discussion of moderated regression. MacCallum, R.C. and Mar, C.M. 1995 Distinguishing between moderator andquadratic effects in multiple regression Psychological Bulletin 116: 405-421 Does what it says in the title. More complex. McLelland, G.H. and Judd, C.M. 1993 Statistical difficulties in detecting interactions and moderator effects Psychological Bulletin 114 (2): 376 Talks about power, and ability to detect moderator effects. Stone, E.F. and Hollenbeck, J.R. 1989 Clarifying some controversial issues surrounding statistical procedures for detecting moderator variables: empirical evidence and related matters Journal of Personality and Social Psychology 74 (1): 3 - 10 Fairly self explanatory. Stone, EF & Hollenbeck, JR 1984 Some issues with the use of moderated regression Organisational Behaviour and Human Performance 34: 195-213 Fairly self explanatory. Maxwell, S. E., and Delaney, H. D. 1993 Bivariate median splits and spurious statistical significance, Psychological Bulletin, (I seem to have mislaid the rest of the details.) Warns against the temptation of the median split followed by ANOVA, not just because of loss of power (the usual problem), but because of increase in type I error rate. Jaccard, J. and Wan, C.K. 1995 Measurement errors in the analysis of interaction effects between continuous predictors using multiple regression: multiple indicator and structural equation approaches Psychological Bulletin 117(2): 348-357 Follows on from McLelland and Judd, but says power can be increased using SEM. Describes a method using non-linear equality constraints available in LISREL 8. Joreskog and Yang (1996) have a chapter in the book Advanced Issues in Structural Equation Modelling, in which they propose an alternative method for detectinjg moderator effects with SEM. Other books which cover some of the issues are: Cohen and Cohen (1983). Applied Multiple Regression/Correlation analysis for the behavioural sciences. Erlbaum. Judd and McLelland (1989). Data analysis: a model comparison approach. HBJ. And books which focus on the issues are: Two in the Sage little green book series (Quantitative Applications in the social sciences) Jaccard, Turrisi and Wan - Interaction effects in multiple regression Then they change their mind Jaccard and Wan - Lisrel approaches to interaction effects in multiple regression. Finally: Aiken and West - Multiple regression : testing and interpreting interactions. Sage. * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

Document by Rich Ulrich. E-mail to wpilib+@pitt.edu

FAQ top.

Ulrich home page.

Ulrich FAQ. http://www.pitt.edu/~wpilib/stats99.html