statistical controls


Fortran must die
We are held to an economic predictor variable in a linear regression model. Certain control variables exist in a fixed effect regression which will control for things which in theory states have no control over (like education when customers apply). More generally to the reality at intake.

I suggested we create programs to address our intake, better counseling for instance to deal with problems tied to the these control variables. So they deal with low literacy or low education cases more extensively than ones not in those groups. But I was wondering if the control variables in the regression would mean that efforts we took to address them would be washed out of the model.


Less is more. Stay pure. Stay poor.
So you have a predictor, then target the predictor and wonder if it will no longer be a predictor. Seems about right, but you would be shifting them into the other group. For example smokers as a predictor of health. I put a tax on tobacco, now I have fewer smokers.

Let me know if I am misunderstanding something in your post.


Fortran must die
[Edit} I see the point you are making now hlsmith. We will have proportionally fewer in one category of the control. I am not sure how that will impact how well we do compared to other states, or if it can.

We have a series of variables that the federal agency will use to control for to level the playing field for states. The response variable is income of our customers. One of the control variables is how much support you get from government at application.

I suggested developing programs to make our response variable (how much income you make) better by dealing with people who get high support (which lowers income ) essentially to moderate the impact of the control variable I think. That may or may not make support not predict, it will probably make it predict differently.

But can that make a difference in income, can we improve income, or will the control simply adjust and defeat the purpose of the program statistically when we are compared to other states. Essentially all states will have a control variable, ours would be somewhat different if the program worked.

The real point I am asking, is if you are being compared to other states and there is a set of control variables and you modify the impact of the control variable in the way I suggested can you still improve how you do on the response variable.