Oaxaca Decomposition Explanied % of Each Variable

[Baris Sozen]

I have 500 observation and 4 treatment (x,y,z,t) and 15 variable. I would like Oaxaca Decomposition to obtain explanied percent of variables. How I can do that?

[FernandoRios]

Hi Baris
Your question is not very clear.
Usually you can use the community contributed command -oaxaca- (ssc install oaxaca) to obtain the Oaxaca-Blinder decomposition assuming two groups of interest, which is often also related to the treatment. However, you indicate you have 4 treatments, so it isn clear if you want to do the decomposition for each "treatment" or not.
Anyways, take a look at the oaxaca command, and the companion paper. That should provide you a good starting point to do what you need to do.
HTH
Fernando

[Zeynep Baktir]

Hi Fernando,

I am actually looking into something similar w/ Baris, and got stuck at some point. Maybe you can provide me w/ some advice.

Btw, a total newbie w/ Stata here so I would really appreciate if you could please tolerate my limited knowledge. Just recently started to learn it because I was suggested to use use Stata for showing " % explained" of some variables for a few hypothesized relationships via Oaxaca- Blinder Decomposition. However, I couldn't figure out how exactly Stata does it for you. I needed to make manual calculations at some point and it just doesn't feel right. I suppose there must me some functunality of Stata for it.

Here is my data in a nutshell:

I have 1000 observations, 4 conditions (A,B,C,D) and 6 variables. I would like to see the effects of these variables on below relationships, shoud there be any significance:

1. A>B
2. C>D


I need to do decomposition for the two relationships above.


Could you please guide me through the process, or provide me with any documents, videos etc that would explain how it is done?

Thank you very much for your invaluable guidance.

Best,

Zeynep

[Marcos Almeida]

because I was suggested to use use Stata for showing " % explained" of some variables for a few hypothesized relationships via Oaxaca- Blinder Decomposition.

Just as a starting point, you may wish to take a look at the user-written programs - oaxaca - and - nldecompose - for that matter. After installing both, just type - help nldecompose - and - help oaxaca - to check the examples.

[FernandoRios]

Hi Zeynep
I have never used the nldecompose suggested by Marcos, however, the -oaxaca- by Ben Jan is pretty comprehensive, and may do what you want.
The paper to go, and start just playing with options and replicating examples is to follow the companion paper :https://www.stata-journal.com/articl...article=st0151
It has a very comprehensive description of what the command does, and what type of output it produces.
Now, in terms of %explained. As far as I know, that section needs to be done manually. And some times, using shares of the explain component and unexplained component may produce results that are difficult to explain if, the explained component has a negative contribution to the total gap, or if the total gap is close to zero.
For better advice, it would be wise to provide a bit more information about what you are trying to do.
1. what is your outcome (dependent variable)
2. what do "conditions" are (A B C D)? what do they measure?
3. What do the explanatory variables measure?
4. What type of decomposition do you want to implement (Oaxaca command implements many types of decompositions)

HTH
Fernando

[Zeynep Baktir]

Hi Fernando,

Thank you for your kind reply. Let me answer your questions as follows:

1. Dependent Variable: Hourly wage offered for a job in informal sector
2. A: Immigrants dressed well to an interview B: Immigrants not dressed well to an interview
C: Locals dressed well to an interview D: Locals not dressed well to an interview
3. Confounding variables: gender, age, ethnic background, religious background, technical training level, former work experience
4. Can you please provide a resource that explains those different types of decompositions?

Thanks again,

Zeynep

[FernandoRios]

The paper I cited https://www.stata-journal.com/articl...article=st0151
has the explanations of the multiple decompositions -oaxaca- does. You can also look into the help file.
Now, you have 4 different groups, so keep in mind that decompositions only compare two groups at the same time. You have to decide which groups will be compared each time.
HTH

[Zeynep Baktir]

Thanks Fernando. I will compare A and B, and then C and D. So two groups each time.

[Zeynep Baktir]

Btw, I am surprised how complicated it seems to do such a basic analysis with Stata. A very straightforward design actually but not a clear, straightforward answer yet

[FernandoRios]

Hi Zeynep,
I think you have all the material you need. You just need to read through it and start replicating the examples in the help files.
Im assuming, however, that you have the basic understanding of OLS regressions, and what a Oaxaca Decomposition does.
If Ben Jan paper seems too complicated, you can also read http://siteresources.worldbank.org/I...EquityCh12.pdf
And if that is not sufficient, you will be better off asking someone who can provide a 1-to-1 tutoring on the method.
HTH
Fernando