0
$\begingroup$

I have a panel dataset of 40,000 observations, that is 8,000 parishes and 5 years. When I run a fixed effects regression, using parish as a spatial unit, it shows negative adjusted R2 and rather meaningless coefficients. However when I switch to a much larger unit, county (parishes are grouped into 42 counties) the model works fine. Is this a common problem? How can this finding be explained?

$\endgroup$
2
  • $\begingroup$ What is it you are testing ? What is your question and your model formula ? $\endgroup$
    – CaroZ
    Feb 22 at 21:01
  • $\begingroup$ My dependent variable is a share of workers on long term contracts. I test the effect of 10 variables, including topographic factors, population density, availability of alternative employment opportunities etc. Population density is in a log form, all others are in the original form. $\endgroup$
    – Mikhail
    Feb 22 at 21:27

1 Answer 1

-1
$\begingroup$

You are not able at present to detect a parish effect, but you detect a country effect. This might be because the parish effect is not independent form the country.

If you are not directly interested for the effect of parish and country in themselves, which you do not seem to be according to your answer, you might want to instead "correct" for them.

In this case the best approach would be to tun a mixed effect model, which would include the variables you mention in your comment as fixed effect, and parish nested in country as a random factor : +(1|country/parish).

$\endgroup$
6
  • 1
    $\begingroup$ Thank you. Does this mean that I have fixed effects at a county level, and below that, random effects at a parish level? $\endgroup$
    – Mikhail
    Feb 22 at 21:43
  • 1
    $\begingroup$ Welcome to the site. At present this is more of a comment than an answer. You could edit to expand it, perhaps by giving a summary of the information at the link, or we can convert it into a comment for you. $\endgroup$ Feb 22 at 23:52
  • $\begingroup$ Thank you. I agree that the text above is a comment rather than an answer. In response, I’d like to clarify that I am very much interested in fixed effects. My problem is that I know the theory but have limited practical experience. All my variables are at a parish level, and I expected to find the fixed effects for parishes. However I can only see them at a higher level of a county. The question is whether this is a common occurrence, to be explained by limited deviation of variables by parish, or something odd and suspicious? $\endgroup$
    – Mikhail
    Feb 23 at 9:43
  • 1
    $\begingroup$ This is definitely not odd and suspicious. $\endgroup$
    – CaroZ
    Feb 23 at 11:01
  • $\begingroup$ Great, thanks! If it is not odd, am I correct when I explain this by small variation of my dependent variable by parish, and larger variation at a county level? $\endgroup$
    – Mikhail
    Feb 23 at 11:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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