Control variables


No cake for spunky
I run a time consuming report for a bunch of variables the federal government controls for. My argument is that it does not matter what their levels are, because the federal government will control for them in evaluating our success (that is their purpose to level the playing field).

But then I thought this. Say high ages lowers income (the DV) and regression shows this. Say we develop a new program to deal with age (we give them new training that leads them to be more successful). Then age would be different for us and the other states even if it is a statistical control variable wouldn't it? And we could do better even if age is the same say between us and other states.

Does that make sense? I have to make a policy recommendation on this soon....


Less is more. Stay pure. Stay poor.
Well perhaps age isn't intervenable on. You can't change age. You can change associated attributes that is the true feature of interest. Age is just a variable associated with the other things.

This is in most scenarios, but age could be relevant if salary was based on a physical attribute that declines with age (e.g., reaction time, endurance, recovery time, etc.).


No cake for spunky
We are looking at the impact of age on income. And in theory there are interventions you might do to increase income that other states would not do. So age would behave differently in your state than other states (sort of like a random intercept in a ML model although we are not running that).

But would any of that matter in the statistical results if age is in the model as a control? Or regardless of what you do the use of a control would mean age would have no impact between the states.

I have not seen this addressed in discussions of statistical control.