Country-specific time trend

[Clara Weber]

Hi everybody,

I am doing a DiD analysis including country and time fixed effects as well as a country-specific time trend (c.year#i.country). Not including the country-specific time trend, the R squared is relatively low at only 0.1. However, when including the time trend, it rises to 0.7. At the same time, both the coefficients for my DiD period dummy and interaction of treatment and period dummy change greatly in magnitude and sign. They are always insignifcant, though.

What exactly does that suggest? My guess would be that there is a high degree of heterogeneity across the countries included in the sample as well as potential issues with the classification of the treatment and control group. Could that be it?

In general, what does the country-specific time trend control for? How is it different to the fixed effects?

Thanks in advance!

Best regards
Clara

[George Ford]

1. You have no treatment effect.
2. There are differences in trends among the countries, so parallel paths is dubious (but corrected for, it is claimed, by the trend variables).

See Angrist/Pischke, Mastering Metrics, p. 196- for a discussion.