Ordinal Logistic Regression and Bimodality


I'm running ordinal logistic regressions with survey weights. I was under the impression that there are no distributional assumptions for this type of analysis. I have 6 covariates in my model, and one of them is continuous. My outcome variable is 3 levels.

Looking more closely at the distribution of the continuous covariate, I can see that it is bimodal - the bimodality is very evident if i log-transform the variable. The two peaks are normal when log-transformed.

What is the best way to handle a bimodal covariate? It doesn't seem right to just ignore it even though there are no distributional requirements in OLR. I tried splitting the variable to isolate the two peaks and when I run my model twice (swapping out the peaks), I now find that the second peak is highly significant while the first peak is not. Previously, when running the model with the variable without accounting for bimodality, significance was borderline.

However most of the observations are contained in the second peak (about 1800) and only 150 are contained in the first peak.

Is there a good way to handle the bimodal aspect of this variable in OLS? And should I run the model with a log-transformed version of the covariate? I was also planning on running multiple linear regression with this covariate - how should bimodality be dealt with for MLS?

I really truly appreciate any and all help!
Thank you