Finding out the reason for collinearity

[Samyam Shrestha]

This sounds very dumb actually. But I tried to control for treatment type (T1, T2, and T3) in the regression results below, and all of them are omitted due to collinearity. Actually, there are four treatment types: T1, T2, T3, and CONTROL. I left out CONTROL as a base for comparison.

Having said that, is there a way to figure out why this is happening? I know this is difficult without actually looking at the data though.

Thanks!

[Clyde Schechter]

Well, without more information about the data it is impossible to be sure. But here's the most likely scenario. If T1, T2, and T3 are properties of a hhid that do not change over time, then they are colinear with the fixed effects. So their effects cannot be estimated in a fixed-effects model. Only things that change over time can be estimated in a fixed-effects model.

[Samyam Shrestha]

Clyde, that was awesome. Thanks. I totally forgot to make it time-variant.