Question

What are the limitations in dichotomising a continuous covariate in simple linear regression?

It's an open question if you have some points please share it!

Answer 1

Dichotomisation of continuous covariate in simple linear regression is a very debatable concept. Although it makes the presentation and interpretation of statistical data easier, it is considered to be statistically non-intuitive because of it's limitations :

1. Dichotomising may work if it creates groups with similar characteristics. But in other hand it results in loss of information.

2. Non-linear relationships such as U-shape associations etc can not be detected through this.

3. Dichotomisation is appropriate only when a threshold effect value truly exists, i.e. if we can assume binary split of the continuous covariate, which creates two relatively distinct but homogeneous groups with respect to a particular statistical output.

4. It is responsible for loss in regression model performance.