Hetero Treatment Effects using IV

[Bryce Mancino]

Hi all,

I'm trying to determine an equation that I can use to determine if the effect of lagged GDP growth on conflict vary heterogeneously depending on slave exports.

This is what I came up with:

regress any_prio gdp_g_l c.gdp_g_l#c.slave_exports, robust cluster(ccode)

However, I also used lagged rainfall as an IV to determine the effect of GDP growth on conflict:

ivreg2 any_prio (gdp_g_l = GPCP_g_l) $x_covariates i.year i.ccode, robust cluster(ccode)

where,
any_prio = conflict
gdp_g_l = gdp growth lagged
GPCP_g_l = rainfall growth lagged
additional info: I am using panel data of countries in Africa from 1985 - 1999

Is there a way I can keep the IV but then include an interaction to determine if there are heterogeneous treatment effects?

[Jeff Wooldridge]

What kind of heterogeneity? You want to interact gdp_g_l with some of the covariates? If so, I'd first demean those covariates and then use the GPCP_g_l*xj_demeaned as the instrument for gdp_g_l*xj_demeaned.