measuring school enrollment gap using blinder-oaxaca decomposition

[chania dera]

Hi,
I am using linear twofold blinder-oaxaca to measure school enrollment gap by disability status
dependent variable is binary school enrollment, 1 if enrolled and 0 not enrolled
independent variable is disability status, 1 if disable and 0 if not disable
control variable: age, age square, gender (dummy), siblings (dummy), father educ (5 category), mother educ (5 category), working mother status (dummy), social economic status (5 category), area (dummy), KIP(dummy), PIP(dummy)
using disable as reference group i run command in stata:

oaxaca school age agesq male siblings father_educ mother_educ working_mother_status social_economic_status urban KIP PIP by(disability) weight() noisily

result of the decomposition is bellow:

School	Coefficient	SE	P>z	[95%
overall
group_1 (non disable)	0,8726211	.0021931	397.89	0.000
group_2 (disable)	0,7460733	.022563	33.07	0.000
difference	0,1265478	.0226693	5.58	0.000
explained	0,0300243	.0113003	2.66	0.008
unexplained	0,0965235	.0210202	4.59	0.000
explained
age	-0,0106228	.0532505	-0.20	0.842
age sq	0,0237524	.0561009	0.42	0.672
male	0,0009473	.0014548	0.65	0.515
siblings	-0,0014554	.0018508	-0.79	0.432
father educ	0,0016144	.0027575	0.59	0.558
mother educ	0,0008308	.0024314	0.34	0.733
working status	0,000062	.0006613	0.09	0.925
economic status	0,0070862	.0042934	1.65	0.099
urban	0,0009698	.0015023	0.65	0.519
KIP	0,0014575	.0046005	0.32	0.751
PIP	0,0053821	.0038322	1.40	0.160
	0,0300243
unexplained
age	0,0051625	.4945026	0.01	0.992
age sq	0,0085134	.2732435	0.03	0.975
male	0,0285709	.0207227	1.38	0.168
siblings	0,0433314	.0307077	1.41	0.158
father educ	-0,0964097	.0638921	-1.51	0.131
mother educ	-0,0041287	.0626446	-0.07	0.947
working status	0,0027917	.021307	0.13	0.896
economic status	-0,0815583	.0519058	-1.57	0.116
urban	0,0209397	.018868	1.11	0.267
KIP	0,0014226	.0121007	0.12	0.906
PIP	-0,0153601	.0117019	-1.31	0.189
_cons	0,1832479	0,2344106	0.78	0.434

Please help me to interpret the result. thank you in advance

[George Ford]

The first two lines gives the conditional means of the two groups.
The third line is the difference between them.
The fourth line says how much of that difference is explained by the model.
The fifth line says how much of that difference is unexplained by the model (here, the bulk of it). This is considered the "discrimination" effect.

The explained/unexplained parts describe the contributions of the variables to the fourth and fifth lines. They will sum to the total effects for each.
So, the most influential variable for the explained part is age^2 (the 0.0300 part is just the sum, ignore it), and for the unexplained part is father educ, economic status, and the constant.

[George Ford]

https://repec.ethz.ch/ets/papers/jann_oaxaca.pdf

[chania dera]

Thank you for helping me to interpret the result.
My supervisor ask for the deeper explanation and interpretation
in overall part
group 1 means the average of school enrollment for non disable is 87,26%
group 2 means the average of school enrollment for disable is 74,60,26%
difference is the gap of the school enrollment between tho groups 12,65%
explained part is the difference of school enrollment because of the characteristic (age etc) can be explained 0,0300 percent point or 23%
the unexplained part is the difference because of the coefficient that means if the non disable have the same characteristic as disable the school enrollment would be lower by 0,0965 percent point
is my answer for the unexplained right?

[George Ford]

The explained part is related to differences in the Xs based on the pooled coefficient.

The unexplained part is the difference based on differences in coefficients from the pooled results.

clear all
sysuse auto, clear

oaxaca mpg weight, by(foreign) pooled noisily

tabstat weight , by(foreign)

di "group_1" _col(20) 41.6797-0.0065879*3317-1.6500029*0
di "group_2" _col(20) 41.6797-0.0065879*2316-1.6500029*1
di "difference" _col(20) (41.6797-0.0065879*3317-1.6500029*0)-(41.6797-0.0065879*2316-1.6500029*1) 
di "explained" _col(20) -0.0065879*(3317-2316)
di "unexplained" _col(20) -1.6500029*1
di "explained"
di "weight" _col(20) -0.0065879*(3317-2316)
di "unexplained"
di "weight" _col(20) (-0.0065879- -0.010426 )*2316 + (-0.0059751- -0.0065879 )*3317 
di "_cons" _col(20) (39.64696)-(48.9183)