Interpret coefficients of Unconditional Quantile Regression

[Jenny Nguyen]

Dear all,

I use OLS and Rifhdreg to have the OLS and UQR as the table below. But I am not sure how to interpret the impact of Age on Log hourly wage for a quantile, for example for the 10^th quantile. How to interpret if both Age and Age squared enter the UQR?

Thank for your help.

[Jenny Nguyen]

Anyone can help me?

[Carlo Lazzaro]

Jenny:
focusing on -Q50- and -Female- predictor omnly, other things being equal switching from -male- to -female- reduces the regressand of 0.1482 units.
As an aside, in both -regress- and -qreg- I would have tested possible non-linear relationships between the regressand and -Age- and -Job tenure-.

[FernandoRios]

Hi Jenny
In addition to Carlo's advice,
for Unconditional quantile regression, you should be interpreting the effect of age refer to "what would happen in Avg age increases in 1 year". Based on your results, if the population grows older in 1 year, the 90th quantile of wages will increase in almost 1%.

If you have age and age sqr, you cannot interpret them individually. I find, however, that the best approach is to use centered age. That way, you can still focus on interpreting age in a regular matter and the effect of c_age^2 becomes something closer to analyzing the impact of an increase in the spread of age in the sample.

For Female coefficient, (or dummies in general) is a bit trickier. The thought experiment is,,,what happens if the proportion of women changes? So my suggestion, in that case, is to say, if the proportion of women increases in say 10% (0.1) wages at the 50th quantile would decrease in 1.15 % = (0.1 * 0.115 *100).

For the OLS you can interpret it both ways. Either as the effect on the conditional mean (12.2% lower wages among women) or unconditional mean (a 1.2% decline in average wages if the proportion of women increase in 10%)

HTH

[Jenny Nguyen]

Thank both of you for your reply,

Fernando, I have two more questions about Oaxaca_rif:

1) How can I group some variables into only one aggregate variable? For example, in the following command, how can I group four education variables (education1-education4) into one variable: Education?
oaxaca_rif loghrwage age female children education1-education4 tenure1-tenure5 experience, by (union) wgt(0) rif(mean) rwlogit(age female children ed1-ed4 tenure1-tenure5 experience) noisily relax

2) How can I get percentage of contribution of each factor? For example, in the following image, beside each column Mean, p10, p25 there is a percentage column. Is there any option of Oaxaca_rif to do that, for example, for the regression in the question 1?

[FernandoRios]

Hi
So quick answers:
1) How can I group some variables into only one aggregate variable? For example, in the following command, how can I group four education variables (education1-education4) into one variable: Education?
oaxaca_rif loghrwage age female children education1-education4 tenure1-tenure5 experience, by (union) wgt(0) rif(mean) rwlogit(age female children ed1-ed4 tenure1-tenure5 experience) noisily relax

You have to use the same syntax as with oaxaca: something like (education: educ1 educ2 educ3 educ4 educ5) (Im assuming there is an Educ0 or educ6 so one category is dropped.

2) How can I get percentage of contribution of each factor? For example, in the following image, beside each column Mean, p10, p25 there is a percentage column. Is there any option of Oaxaca_rif to do that, for example, for the regression in the question 1?

You cant. You simply have to do that manually.
HTH

[Jenny Nguyen]

Hi Fernando,
For the second question. I tried with (geduc: educ1 educ2 educ3 educ4 educ5) and found that it work with Oaxaca, but not work with Oaxaca_rif. The error is: "variable geduc not found".

[FernandoRios]

Can you show what are you typing?
For example, if I use the toy dataset in the helpfile:

use http://fmwww.bc.edu/RePEc/bocode/o/oaxaca.dta
oaxaca_rif lnwage (educx:educ exper) tenure, by(female) wgt(1) rif(mean) rwlogit(educ exper tenure)

I receive no error

[Jenny Nguyen]

Thank you, I found that when I put "noisily" into the regression, the error appears: "variable education not found". For example:
oaxaca_rif lnwage (educx:educ exper) tenure, by(female) wgt(1) rif(mean) rwlogit(educ exper tenure) noisily

The error will appear: "variable educx not found"

[FernandoRios]

Sorry i couldnt replicate the error.
Im currently working with Stata 17, and this error does not appear.
Also, I have the latest versions of RIF and oaxaca
Perhaps that will help