I don't have much experience with panel data so I apologize in advance if this sounds ridiculous.

I am currently writing a thesis investigating the impact of political risk on managerial risk-taking.

I control for various variables that are known to affect managerial risk-taking and include dummy variables for industry, country, and year (these are the fixed effects).

What exact unobserved variables are controlled for in a model with multiple fixed effects?

My understanding is that the coefficient estimates show the average effect the explanatory variable has on the dependent variable across groups. How are those groups exactly defined, when industry, country, and year fixed effects are included?

Would the group be industry-country-year such as retail-UK-2019?


1 Answer 1


Your interpretation goes in the right direction, but it does not sound fully correct to me.

It's important to interpret any coefficient for political risk as a conditional estimate: It compares managerial risk-taking between different political risks given a certain level of all other covariables. In your case, this means in a given industry in a given year in a given country. So you are essentially comparing within a country, industry and year (that's why it is also called the within estimator). So note that you are not comparing across these "groups", but you are holding them constant.

Note that your dummies only allow to estimate one effect for each year, industry and country, but not for each combination thereof (which would requrire interaction terms - which are probably not an option given the degrees of freedom they soak up).

As an example, your fixed intercept for the year 2012 will apply to all observations, those from Kenya and those from Fiji. If you would like to know the average managerial risk-taking in the group "retail-UK-2019", given your model and covariables, you'd have to sum up the intercepts for "retail", "UK" and "2019".

For more on this, see also this great beginner econometrics resource.

  • $\begingroup$ Thank you very much for your answer! Just to make sure I have understood your answer correctly, is it correct to state that: including country, industry, and year fixed effects (dummy variables) controls for unobserved group heterogeneity (where a group includes all observations that belong to the same industry group AND country AND year)? I infer this statement from your answer that estimates are calculated by comparing observations within a country, industry, and year. $\endgroup$
    – user311668
    Feb 18, 2021 at 15:23
  • 1
    $\begingroup$ It adjusts for unobserved heterogeneity between industries, countries and years. It does not adjust for heterogeneity between specific country-industry-year "groups". For example, if there was something peculiar going on ONLY in retail in UK in 2019 that confounded the relationship, it could be missed. If you want to adjust for these "groups" instead, you need to specify fixed/random effects for these "groups" (which is equivalent to interacting sector, country and year). But the former approach is way more conventional and will leave more variability to be explained by political risk $\endgroup$
    – stefgehrig
    Feb 18, 2021 at 16:59

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.