Is there a rule for including/excluding fixed effects?

[Ruth Azlor]

Is there a good argumentation to drop fixed effects indicators? I am investigating the effect of local unemployment on individual labour market outcomes (1=employed, 0=not). My model is as follows:

xtreg employed unemp_rate $id $municipality i.year i.nation i.municipality, re i(pnr) vce(robust)

The global variables include individual and municipality characteristics. If I include year and nationality fixed effects (i.year i.nation), local unemployment (unemp_rate) is significant and has a negative effect (which is what I expected), but if I include municipality fixed effects (i.municipality), local unemployment is insignificant and slightly positive.

My argument would be that the within municipality variation in unemployment is low, i.e. within municipality standard deviation is 1% and across municipalities 1.7%, the average unemployment rate is 4.4%. How does that sound? Does anybody know any published paper/literature with similar argumentation?

[Carlo Lazzaro]

Ruth:
I fail to get whether your dependent variable is binary (1=employed, 0=not) or else.
It it were binary, you should go -xtlogit-.

[Phil Bromiley]

To expand on Carlo's points, you'll increase your chances of a useful answer if you follow the FAQ on asking questions - provide Stata code in code delmiters, Stata output, and sample data using dateex. Try to simplify the code and example to what you need to illustrate your problem.

You have a lot of fixed effects and a set of random effects. If you don't want to interpret them, you don't need higher level fixed effects if you include lower - fixed effects for municipalities will control for national differences as well. If pnr is always within the same municipality, you could try pnr as a fixed effect and drop municipalities. A random effects estimator includes both within-panel and cross-panel variation so it is not the cause the it requires within-panel variation (as a fixed effects estimator does).