Re: st: My ANOVA and regression results don't agree

[Phil Schumm <pschumm@uchicago.edu>]

On Jan 6, 2014, at 2:15 PM, Pepper, Jessica <jkadis@live.unc.edu> wrote:
> When I run an ANOVA, I see 2 significant main effects. But I when I do a follow-up regression to see the contrasts for the 3-level variable, both effects become non-significant. I'm using Stata 12.1 for Windows.
> 
> Here are the details. I ran an experiment in which I randomly exposed participants to one of 3 ads and then measured their interest. I'm curious whether there's an interaction between participant sex and response to this ad. My DV "interest" is continuous. My IVs are: "sex" (2 levels, coded 0 and 1) and "adcontent" (the manipulation; coded as 0 (control), 1, and 2). Here is my code:
> 
> anova interest adcontent##sex
> regress interest i.adcontent##i.sex
> 
> The ANOVA tells me there's a main effect of adcontent and sex. But when I do a follow up regression, both become non-significant.


In a nutshell, the tests for the two main effects in the ANOVA output do not correspond to the tests you get by default using dummy variables with -regress- followed by -test- (even though, in terms of the fitted values, the two models are identical).  The UCLA Statistical Consulting Group has an FAQ on this:

    http://www.ats.ucla.edu/stat/stata/faq/main_effects.htm

If you still have questions after reading that, let us know.


-- Phil


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/