FAQ/OLSvML - CBU statistics Wiki

Revision 16 as of 2019-08-01 13:08:30

Clear message
location: FAQ / OLSvML

Using linear regression as an alternative to logistic regression

Hellevik (2009) suggests that linear regression can be used with a binary outcome as opposed to logistic regression particularly for large samples. P-values between the two types of regression were found to be very close despite the ordinary least squares approach not incorporating heterogeneity of variance (where the variance of the proportions depends on the proportion) which is taken into account by logistic regression.Mood (2010) shows almost identical results using linear regression and logistic regression in Table 5 of the paper and also suggests linear regressions may be used on a binary outcome e.g. to make the results more interpretable.

On a related theme this research report by Jaeger suggests using mixed random effect models as opposed to the arcsine transformation when analyzing proportions in repeated measures ANOVAs when proportions are less than 0.20 or greater than 0.80.

In addition to the use of OLS and logistic regressions Mood (2010) also raises some caveats about comparing logistic regression odds ratios between different groups but Kuha and Mills (2018) argue that these concerns are usually misplaced.

Reference

Hellevik, O. (2009) Linear versus logistic regression when the dependent variable is a dichotomy. Qual Quant 43 59-74.

Kuha, J. and Mills, C. (2018) On group comparisons with logistic regression models. Sociological Methods & Research 1-28.

Mood, C. (2010) Logistic regression: why we cannot do what we think we can do, and what we can do about it. European Sociological Review 26(1) 67-82.