# How should i treat an explanatory variable that is proportional data for simple linear regression odels?

I am trying to analyse home behavioural data from a field study in Madagascar based on a species of lemur. I hoped to investigate different behvaioural predictors of Day Path Length. These include feeding, resting, socialising and locomotion times as proportion of the total behaviours per day (frequency of behaviour over total number of scans for that focal day).

Does my proportional data on my x axis require a logit transformation in order to run linear regression models?

Many thanks for your help.

## 1 Answer

Ordinary least squares regression (i.e., a general linear model with a Gaussian link function) does not make an assumption about the univariate distribution of the predictors. You can use proportions that are bounded between 0 and 1 as predictor variables just fine. The key distributional assumption is that the predicted variables are normally distributed, given the predictors.

If a rate or proportion were your dependent variable, you could consider something like beta regression. But not for independent variables (X, explanatory, covariates, whatever you want to call these variables).

• Thanks Mark, that makes a lot of sense! – Ben Morrison Apr 2 '20 at 9:19