Oaxaca Blinder Decomposition: Intercept Differences Vary Widely Between Years

[HannahE Olson]

Hello,

I think this is more of a statistical question than a stata question. I am running a oaxaca blinder decomposition of race differences in birth weight at two different time points, separately by delivery method (code below):

 oaxaca bwt    gest    age somecoll married malechild breech (hypertension: phyper chyper) diab eclamp tobacco monpre normalize(wtg1 b.wtg2 wtg3) if year==`y' & del3==`d', by(black) w(0)   adjust(malechild breech) detail

My question is about the intercept difference and how drastically it shifts between years. In 1990, for induced births, the intercept difference is -93 grams. By 2018, the intercept difference is -544 grams. At the same time, the difference that is attributable to gestational age coefficients (i.e., race differences in GA-specific birth weight) changes from -135grams in 1990 to +302 grams in 2018. This means that for induced births the difference in the black-white gap in mean birth weight that is attributable to race differences in GA-specific birth weight increases by 438 grams between 1990 and 2018, but at the same time the difference that is attributable to intercept differences decreases by 452 grams...Does anyone have any ideas about what is going on here? Any insight or ideas for further investigation are appreciated.

Thanks,
Hannah

[HannahE Olson]

update: I rescaled gestational age so that "full term" (i.e., 40 weeks) is equal to zero rather than 0 weeks being equal to zero and that evened things out.

[FernandoRios]

Hi Hanna
Something else you can do, if you re still looking into explaining your model.
I find it useful to estimate all the underlying models manually (or using noisily option). This way you can see how the difference coefficients are before the decomposition is obtained.
That being said. Scales. as you saw, can make an effect on the estimation of the gaps. So one should be careful in justifying a particular transformation.
HTH
F