I was recently asked about whether it’s okay to treat a likert scale as continuous as a predictor in a regression model. Here’s my reply. In the question, the researcher asked about logistic regression, but the same answer applies to all regression models.
1. There is a difference between a likert scale item (a single 1-7 scale, eg.) and a full likert scale , which is composed of multiple items. If it is a full likert scale, with a combination of multiple items, go ahead and treat it as numerical.
2. If it is a a single item, it is probably fine to treat it as numerical. There is more justification for this if it has 7 or more values, but even with 5 you may be okay.
3. There are NO assumptions about the distribution of the predictor (independent) variables in any regression. However, parameter estimates generally are only interpretable for nominal categories or numerical quantities.
The coefficient is interpreted as the difference in the mean of Y, the outcome, for each one-unit difference in X, the predictor. If the predictor is categorical and dummy coded, a one-unit difference simply refers to switching from one category to the other. If the predictor is numerical, a one-unit difference should be meaningful.
Ordinal predictor variables have to be treated as either nominal unordered categories or numerical. In the former case, you are throwing away information about the ordering. In the latter, you’re making assumptions about the differences between the scale items. If those distances can be reasonably considered equal and meaningful, then it is reasonable to treat the predictor as numerical (i.e., if a one-unit change from 1 to 2 is roughly equivalent to a one-unit change from 3 to 4).
For more information and some nice references on using likert scales see my post on “Can Likert Scales Ever be Considered Continuous?”