Difference in Difference Time-Variant Fixed Effects

[Thomas Wake]

Hi,

I am currently running analysis of the effect of a defunding policy on local level crime rates. I only have 3 time periods for my data (due to lack of data availability), and I have 203 data points for each. I have been struggling to find significance with my DID results, however, I believe I now have achieved this.

My question is, when including fixed effects, is it always necessary to include time-variant fixed effects for such panel data when running xtreg? When I include this (i.year , fe), my results become insignificant, whereas if I do not (,fe), my results are significant.

Any advice would be much appreciated

[Maxence Morlet]

If I were you I would always include time FE. What this tells you is that after accounting for time FE, your predictors do not significantly affect your outcome. You would then have to modify your specifications, change scales, use different predictors, etc.

[Thomas Wake]

Thank you!