[FernandoRios]

Hi Belen
I would suggest you to also run the individual rif regression to have a better understanding of how the geographical composition affects wages before you try to interpret the OB decomposition.
It usually helps me understand what is going on.
Fernando

[Belen Fontenez]

Hi FernandoRios. Thank you so much for your suggestion. I ran two rif regressions (one for mothers and one for fathers) and I found that region coefficients are negative, as you can see below:

For mothers

rifhdreg lnrwage exper expersqr yr_school married dgba dnoa dnea dcuyo dpampeana in_employee self_empl if parent==1 & fem==1, rif(q(10))  robust

For fathers

rifhdreg lnrwage exper expersqr yr_school married dgba dnoa dnea dcuyo dpampeana in_employee self_empl if parent==1 & fem==0, rif(q(10))  robust

Results

 
Dummy
Coefficient mothers
Coefficient fathers

dgba
-0.2807077
-0.1918838

dnoa
-0.1927269
-0.2261934

dnea
-0.1580311
-0.3719334

dcuyo
-0.0281222
-0.1705591

dpampeana
-0.0673897
-0.1264541

The signs of these coefficients make sense, as Patagonia is the region with the highest wages in Argentina. So then, is the interpretation of my previous post correct? That mothers in gba are paid less than fathers in comparison to what mothers and fathers are paid in Patagonia and the other way around for cuyo?

Thanks again for your help!!
Best wishes,
Belen

[FernandoRios]

Ok this is more informative. But the conclusion is not as simple.
For a second, forget the quantile (that complicates things a bit).
IF this were coefficients for a simple regression, you would say that. Women living in GBA earn lower wages compared to those in Patagonia. While men in Patagonia, also experience a penalty living in GBA, the penalty is not as large (for women 28% for men 19%). However, you do not know if women are paid less GBA compared to Men, because you do not know how much they earn in Patagonia.
What I have seen for these cases, I try a narrative tying the wage structure to regional wage differences, but without pointing at specific regions.

Now, the quantile.
you cannot analyze the effects on a single quantile as an isolated event. You need to look or consider the whole distribution to say something about the effect on the distribution.

if the 10th quantile increases, doesnt mean that wages are increasing for everyone. If all quantiles increase, then you could say something on those lines.

The bottom line, you will need a larger narrative to explain the results. And avoid making explanations assuming only specific quantiles are affected. You need to look at the whole picture to really appreciate the differences in fathers' and mothers' wage distribution.

Fernando

[Belen Fontenez]

Thank you FernandoRios! This is really helpful! Thanks for taking the time to read it and explain to me!

Kind regards,
Belen

[Belen Fontenez]

Hi FernandoRios! Thank you again for your help! I have another doubt. As I mentioned before, I'm looking at the gender wage gap between mothers and fathers, on the one hand, and between non-mothers and non-fathers, on the other. My main objective is to study if discrimination is greater between mothers-fathers (MF) than between non-mothers and non-fathers (NMF). I ran rif regressions and oaxaca rif decompositions, but I'm not sure if I can compare the unexplained part calculated from MF with the unexplained part from NMF. For example, I get the following results:

Wage decomposition:

 

OLS
q10
q50
q90

Mothers vs Fathers 
0.4992
0.93972
0.37717
0.1887

 Explained
0.1349
0.35368
0.09514
-0.00879

 Unexplained
0.3643
0.58604
0.28203
0.19749

Non-mothers vs non-fathers 
0.22077
0.47102
0.20489
0.10342

 Explained
0.01299
0.08661
-0.00994
-0.05178

 Unexplained
0.20778
0.38441
0.21483
0.1552

in %

 

OLS
q10
q50
q90

Mothers vs Fathers
100%
100%
100%
100%

 Explained
27%
38%
25%
-5%

 Unexplained
73%
62%
75%
105%

Non-mothers vs non-fathers
100%
100%
100%
100%

 Explained
6%
18%
-5%
-50%

 Unexplained
94%
82%
105%
150%

As you can see, the unexplained component of the gender gap is greater between MF than between NMF, showing evidence of higher discrimination for mothers. However, when I compute the percentage of the unexplained component relative to the raw gender wage gap, I find that, in relative terms, discrimination is higher for non-mothers. So, my question: can I compare both unexplained components to say something about discrimination for mothers?

Thank you again! I really appreaciate your comments!!!
Best wishes,

Belen

[FernandoRios]

Hi Belen
Interesting results. I think , however, some of the wage differentials may be coming from differences between parents and non parents.
For instance, it is often found in simple mincer regression that children have a positive effect on earnings. But this effect could be related to the life stage of parents (older than non parents), so the effect of parenthood could be related to them being older/more educated.
did you restricted your sample so demographics of parents and nonparents are similar (except for parenthood)?
Best

[Belen Fontenez]

Hi FernandoRios ! Thanks for your answer. My sample consists of women and men between 18 and 49 years old, with similar sample means of years of education (12.5 yrs for parents and 11.4 for non-parents). I've confirmed that both groups have similar sample means of education by wage decile. Following your suggestion, I restricted, even more, my sample to people aged between 30 and 49 years old, to get rid of any kind of imbalance. I get the following results:

Wage decomposition:

 

OLS
q10
q50
q90

Mothers vs Fathers
0.48776
0.93500
0.33035
0.19129

 Explained
0.13049
0.32542
0.08839
-0.00746

 Unexplained
0.35727
0.60958
0.24197
0.19875

Non-mothers vs non-fathers 
0.18984
0.37436
0.11767
0.12151

 Explained
0.00344
0.09521
-0.02301
-0.05174

 Unexplained
0.18640
0.27915
0.14068
0.17325

in %:

 

OLS
q10
q50
q90

Mothers vs Fathers
100%
100%
100%
100%

 Explained
27%
35%
27%
-4%

 Unexplained
73%
65%
73%
104%

Non-mothers vs non-fathers 
100%
100%
100%
100%

 Explained
2%
25%
-20%
-43%

 Unexplained
98%
75%
120%
143%

As you can see, my results have slightly changed from those posted before, but I still get the same conclusions: the unexplained component of the gender gap is greater between MF than between NMF, while the percentage of the unexplained component relative to the raw gender wage gap shows higher discrimination for non-mothers. Then, do you think that it would be incorrect to compare "absolute" unexplained components to show that mothers experience greater discrimination?

[Fabiola]

Hi,

I’m using the oaxaca command, in Stata 17, to make a twofold decomposition. I read about the use of the normalize option to avoid the omitted category bias.

I have a binary dependent variable and six independent variables, with two categories each. The other two independent variables are continuous. I found that the aggregate estimates for each independent variable using the normalized effects are the same as the estimates for each category when decomposing without using the normalize option.

I also changed the reference category in the model without "normalize", but got the same results. I would like to know if it is correct or if my codes are wrong.

Here is the code I used to get the “normalized” effects

oaxaca sb_auto idade_18 (fem: normalize(fem_d?)) (renda_qui_oaxaca:normalize(renda_qui_oaxaca_d?)) (abvd:normalize(abvd_d?)) (con_dente2:normalize(con_dente2_d?)) (dif_comer2:normalize(dif_comer2_d?)) (prot:normalize(prot_d?)) ndentes, by(escol2_oaxaca) logit weight(0) svy noisily, if subpop_sbauto_oaxaca==1 & ndentes>0

Here is the code without the normalize option (reference category is 0)

oaxaca sb_auto idade_18 fem_d2 renda_qui_oaxaca_d2 abvd_d2 con_dente2_d2 dif_comer2_d2 prot_d2 ndentes, by(escol2_oaxaca) logit weight(0) svy noisily, if subpop_sbauto_oaxaca==1 & ndentes>0

Code without the normalize option (reference category is 1)

oaxaca sb_auto idade_18 fem_d1 renda_qui_oaxaca_d1 abvd_d1 con_dente2_d1 dif_comer2_d1 prot_d1 ndentes, by(escol2_oaxaca) logit weight(0) svy noisily, if subpop_sbauto_oaxaca==1 & ndentes>0

Any support will be very much appreciated.