bilateral migration push pull factor significance

[william kana]

I am attempting to investigate the significance of push and pull factors in determining medical brain drains from developed and developing countries. I have run a random effects regression of the form

logmbd = a + b1loggdpi + b2loggdpj + b3healthexpi + b4healthexpj + ......

where i is origin country and j is destination county

Following this I wish to test whether H0: b1>b2, in absolute terms, regardless of the sign to determine whether the pull factor is more significant than the push factor and then repeat the process for restricted data of only developed and only developing origin countries respectively but am unsure as to what initial test I can run.

[Clyde Schechter]

My initial reaction was to dismiss this as another hopeless quest to compare the effects of two different predictors in a regression model, but this may actually be one of the rare situations where it makes sense. The variables involved are actually measures of the same construct and the variables are all dimensionless--so this looks pretty good.

The code for doing it is very simple:

regression model involving variables loggdpi and loggdpj
test loggdpi = loggdpj // FOR A HYPOTHESIS TEST

// OR

lincom loggdpi - loggdpj // FOR AN INTERVAL ESTIMATE OF THE DIFFERENCE

The one area in which this seems to be, perhaps, not such an ideal context for contrasting these effects is that I would imagine that the distributions of gdp in the origin and destination countries are rather different. Since you are doing a log-log regression, your coefficients as measures of effects are estimating elasticity: % change in bmd per % difference in gdp. So a 1% difference in origin country gdp may represent a very different magnitude of gdp difference than a 1% difference in destination country gdp. So, I'm not entirely sure that a difference in the coefficients really captures a difference of effect in a substantive way. But at least you could argue that with a straight face.

[william kana]

some co-efficients in my regression are log-log but some co-efficients are log-level such as migrant stock (% of population), therefore it would provide semi elasticity. Subsequently the effect of a unitary rise in % migrant stock at origin can be compared against the effect of a % rise in migrant stock at destination.