An alternative way to incorporate spillovers is to use larger geographical units (Borjas
2003). We re-estimate models 5 and 6 from Table 2 using provinces rather than districts.
[Table 3 here]
The spatial lags models, shown in columns 1 and 2 of Table 3, fail to detect any
relationship between wages and distance to the Myanmar border. Most of the coefficient
estimates from the spatial lags models are smaller in absolute value than those of Table 2,
but detecting a relationship with distance to the border is perhaps especially difficult
because distance is highly correlated among neighboring districts. Without the negative
relationship between wages and migration or distance to Myanmar, the coefficients on
the Myanmar border dummy again become negative.
The instrumental variables models based on provinces, shown in columns 3 and 4, give a
completely different result. The coefficients on migrant intensity are about three times as
large as those from Table 2, albeit with much larger standard errors.