The assumptions of multiple regression: An article by TalkStatters!

CowboyBear

Super Moderator
#1
Hi everyone,

A lot of you are aware of this already, but I'm posting this to a public subforum on Quark's suggestion :)

Recently a group of 3 Talkstats members wrote an article about the assumptions of multiple regression (with great input from a number of other members too!) The article is open access, and available from the link below:


Williams, Matt N., Grajales, Carlos Alberto Gómez, & Kurkiewicz, Dason (2013). Assumptions of Multiple Regression: Correcting Two Misconceptions. Practical Assessment, Research & Evaluation, 18(11). Available online: http://pareonline.net/getvn.asp?v=18&n=11


The article identifies two misconceptions about the assumptions of regression that we found in a previous, widely cited article on this topic (and have come across elsewhere). The misconceptions are:
1) That regression requires normally distributed predictor and/or response variables
2) That measurement error can only reduce (rather than increase) simple regression coefficients and correlations

Starting on page 8, the article also provides a corrected summary of the assumptions of regression. It includes brief comments about how to identify and respond to assumption breaches. We've also tried to include explanations of what the assumptions are needed for, which is something that often isn't covered in regression resources aimed at non-statisticians.

The article is nominally about regression, but our summary of assumptions also applies to other general linear models with single dependent variables (e.g., ANOVA, ANCOVA, t-tests).

Hopefully this will be useful to users of the forum!
 

hlsmith

Omega Contributor
#4
A nice article on model building:


Sander Greenland& Neil Pearce. Statistical Foundations for Model-Based Adjustments. Annu. Rev. Public Health 2015. 36:89–108.