Substantively that seems like a reasonable question, but I have not run across a way to determine the impact of a categorical variable this way. ]]>

Apologies in advance for the long post, and let me know if im violating any forum rule. I’m trying to expand Erik Stafford’s(2015) analysis on taken private acquisition decisions, by having an OLS regression on a binary dummy indicating a company was taken private(data is unbalanced panel with acquired companies and non-acquired companies, with dummy being positive on years of acquisition. His main analysis is presented below:

I have created several other performance variables, the descriptive statistics presented below. Variables are already winsorized at 2%.

The variables are grouped into doubts about how I should deal with them. Any support in dealing with any of the groups is very much helpful!

Group 1 - Variables between zero and 1. Regress on the variable like Stafford does on D/V. No questions.

Group 2 – Absolute measures of size. I simply take the log, like Stafford does on ME

Group 3 – Valuation ratios. I also take the log. My question if I should just define the variable when the ratio is positive(since a company cannot be “worth” a negative multiple) and take the log after and regress on it, or define the variables regardless of whether positive or negative and then take the log and regress on the log. I am currently doing the former.

Group 4 – Ratios from 0 to very high values.

Group 5 – Growth and margin figures as well as averages of these across 3 year periods. This is the one that is really bugging me as they would very interesting to analyze, but I don’t know which transformation, if any, to use for these given their extreme values both on the negative and positive side.

Group 6 – Similar to 5, but with much more extreme observations. It’s the ratio of interest expenses to EBT and it would be an interesting variable to analyze leverage.

Thank you very much for any help in dealing with this! ]]>