Appropriateness of Fixed Effects when using a dummy that is only time variant for some panel units.

[Kieran Davies]

I am looking to estimate the impact of the presence of a fiscal council on a country's real growth forecasting error. I have run the following fixed effects model:

xtreg rgfe L.fc L.fri L.rgog L.debt i.pubdate, fe robust

Where rgfe is the real growth forecasting error and L.fc is a lag of a fiscal council dummy variable equal to 1 in the presence of a fiscal council and 0 otherwise. I am aware that time-invariant explanatory variables are removed because of the transformation involved in fixed effects estimation. For some countries, the fiscal council dummy is time-invariant, i.e. it is always equal to either 0 or 1 for all time periods in my sample. For others, the fiscal council varies over time, e.g. so that for the UK, FC=1 in all time periods after 2009 but FC=0 before.

Is it appropriate to use fixed effects estimation for this model if I am primarily concerned with the fiscal council as an explanatory variable?

[Andrew Musau]

From an estimation point of view, there is no issue with including the variable as long as it varies for some units over time. The decision therefore should rest on whether this makes sense in terms of your theoretical model.

[Kieran Davies]

That's good to know. Thanks Andrew.

[Phil Bromiley]

Just remember that the estimate will then depend on just those panels where council varies over time. In some cases, this could be identified but might only be one or two panels. If this never happens in a particular country, then you may want to not generalize from parameters essentially estimated on the UK experience to those other countries.