advice on centring and how it effects my regressions

[Bill phippz]

Hi,

In my sample, I have a cutoff value of 0.5 units, with individuals below this threshold being 'treated'. To analyze the data, I created a new variable centered around 0 (i.e., 0 is the midpoint), by calculating gen new_var = units - 0.5.

However, when running my regression, I am unsure whether I should use the original or the centered version of the variable (Problem 1).

Additionally, I’ve been instructed to include only data within a range of -0.5 to +0.5 units (i.e., units >= -0.5 & units <= 0.5 or new_var >= -0.5 & new_var <= 0.5, if I use the re-centred?), since I’m focused on observations near the cutoff. Since the values of my centered variable fall within this range, it seems fine.

But I was also advised to expand the bandwidth by 0.1 to include observations further from the cutoff. However, all my centered data already falls within the -0.5 to +0.5 range, so expanding the bandwidth doesn’t change anything. Should I be doing something differently, they implied that it should?

[Mike Lacy]

I think you'll get better advice here if you explain some of the substance of your data and the question(s) about it you want to answer. I suspect your real difficulty might be somewhat different than what you or your advisers think it is, but it's hard to tell. So, I'd encourage you to repost with some explanation.

[Daniel Feenberg]

Subracting a constant from a RHS variable affects only the constant in a linear regression. Coefficient and standard error remain the same. If you see other things change, it would be because some other action in the code is affecting the sample composition.
.