How to find upward variable bias?

[Andreas Baltin]

Dear Statalisters,

I am currently running some analysis on a few different dependent variables in a 2-period panel with a fixed effects estimator. I get good results for each dependent variables, however, I feel like I have an upward bias in my predictor.
I have strong literature backing supporting that 2 of my dependent variables should be positively prediced by the predictor, whereas 2 should be negative. What I get is that the 2 positive ones are positive with a high magnitude, the 2 that are meant to be negative are also positive with a really small magnitude. I strongly believe that my regression suffers from omitted variable bias.

I have constructed my panel from a survey with hundreds and hundreds of observation. I have consulted literature to find the appropriate control variables, but I still think I have an upward bias in there (as I am doing something slightly different than literature).

Is there any way to find out what causes this bias? Or what should I do in order to adress this?

Many thanks in advance,
Andreas

[Phil Bromiley]

You'll increase your chances of a helpful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

You mention bias in predictor then you talk about bias in parameter estimates. These are very different things. User written ivreset and reset both attempt to identify specification error. I don't know how good these techniques are. I've also seen reference to "the impact threshold of a confounding variable (ITCV) for our interaction terms (Frank, 2000; Larcker and Rusticus, 2010)" which might be relevant.

The most obvious answer is that you got the results you got and shouldn't be manipulating things to produce the results you think you should have obtained.