2
$\begingroup$

I am running regressions to estimate the impact of health conditions on life satisfaction.

I have read a paper that showed that there is likely to be measurement error in the self-reported health variables in that those who do not work are more likely to report health problems to justify not working.

My question is, is this a problem for my estimates? I include various dummy controls to capture employment status so do I simply control for this potentially problematic bias due to measurement error?

Or, is the bias still there in which case I guess it would be a downward bias? Does my use of panel data and fixed effects make any difference?

$\endgroup$
  • $\begingroup$ I am unsure of whether or not it is a problem if you have adequate control variables, but if it is, using Instrumental Variables regression (IV) is quite fruitful in health econ. I would imagine using number of hospitals would serve well as an instrument as it shouldn’t cause any change in life satisfaction aside from its affect through health conditions. $\endgroup$ – Brennan Aug 4 at 3:04
1
$\begingroup$

Check the correlation between health status variables and employment status. Start with simple pairwise correlations and then perform auxiliary regressions where you regress a health indicator on the employment status variables.

If there is visible explanatory power in employment status for health indicators, you may want to treat health indicators as endogenous variables and employment status as instruments. Depending on the number of variables, this may be the Two-stage Least Squares estimator.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.