There is an article in the Stata journal 8(4), that gives a great introduction to the concept, as well as to the user-writen command for Stata, written by Ben Jann. 

I hope this helps

Ariel

Let me update my previous response and add some detail:

Stata Journal 8(4):480--492 introduces the Blinder-Oaxaca decomposition for nonlinear regression models, using the command: nldecompose

Stata Journal 8(4):453--479 introduces the Blinder-Oaxaca decomposition for linear regression models, using the command: oaxaca

I hope this helps

Ariel

Worldbank has a nice guide to decomposition explaining it and using stata commands (I suspect it is older so they use a command called decompose to get results) Not sure if it is what you need, but it helped me to understand your question a bit better at least :) 

http://siteresources.worldbank.org/INTPAH/Resources/Publications/459843-1195594469249/HealthEquityCh12.pdf

You can follow this book: http://www.worldbank.org/en/topic/health/publication/analyzing-health-equity-using-household-survey-data

At the end you will find the syntax also for Oaxaca decomposition. 

Thank you all! I'll read the documents you shared and try to see if I can sort it out! :)

the material you all shared and the tips have been very helpful. I'm making sense of this. I have a specific question: should I include weights when using the oaxaca command? or does it depend on the type of decomposition i'm using?

It depends on what your analysis entails. If you require weights in your analysis, then you can certainly include them in the model(s). The Stata commands I gave you in the earlier post accept weights.

Another basic question:

- Should I do any tests with the variables I want to use before or after running the oaxaca (e.g. testing for normality of residuals, checking homoscedasticity or multicollinearity)... as in regression analysis? Or any other procedures? I don't see the papers about salary decomposition mentioning these procedures... Thanks again!

Yes, this approach uses linear regression models so all assumptions for those models apply. There are other decomposition models that are less restrictive.

Thank you very much for your help again! Another question...Whose decomposition does the command "pooled" correspond to? I assume that "omega" is Oaxaca and Ransom (1994)... And what about Reimers and Cotton?

:) If you still have the patience to answer my question, I'd be very thankful :)

See the top of page 470 of Ben Jann's paper. It gives the associated references for each type.

Ariel

http://www.stata-journal.com/article.html?article=st0151

so.. is this right?

weight(1) and weight(0) = Oaxaca, 1973

omega = Neumark, 1988

pooled = ??

I think I'd like to use the pooled model, but from Ben Jann's article I can't find out if it's his model... or whose model it is!

Thank you :)

Just cite Ben Jann's paper, since it is his implementation of the decomposition approach.

Another question. ... I've finally had access to the data I want to analyse and I had results that were very incoherent with what I had observed while doing binary Analysis with all the variables. Oaxaca suggests that there is no difference between men and women's salaries, .007 explained and 

006 unexplained.  

I then used oaxaca with hourly wage (instead of the natural logarithm of hourly wage) and it suggested a difference of .28 and .68 was unexplained... this seemed extreme but closer to the truth.

Then I wondered... shouldn't these results be more similar? What could be wrong?  Should I use the ln of wage or does oaxaca automatically compute the natural logarithm and I should use the hourly wage instead of the log? ??? 

Thanks again! !

Given that you are dealing with regression models, any deviation from linearity will affect the results. Wages are not normally distributed, and therefore researchers typically log transform them. So if you compare your results with and without transforming wages, it does not surprise me that you will get different effect sizes (and possibly conflicting).

I can't really say if this is the only issue with your data, given that I have not seen your data, or tested it myself. So this is my best guess...

Ariel

Thanks again for your answer! There must have been a mistake. I did everyhting from the beginning and now there is a difference between log wages and non-transformed ones, but it is much smaller! Now the results make sense. Thank you very much for your help, it was very useful and I'm  very happy that I managed to do these analysis and learn a lot during this process :)

Olá Inês. Usei a decomposição de Oaxaca em estudos sobre relações salariais.

p.e. http://www.tandfonline.com/doi/full/10.1080/09645290802628437

Se estiver interessada posso enviar-lhe o paper. Os cálculos foram feitos em stata. Se estiver interessada em ver essas questões com pormenor diga ([email protected]) e combinamos a melhor forma de o fazer.

Luis

Muito obrigada Luís Delfim Santos pela sua resposta! De facto, já conheço o seu trabalho na área, que li e reli várias vezes. Neste momento, já terminei a tese de doutoramento, mas em situações futuras penso que poderia ser relevante sim rever algumas questões mais ao detalhe e eventualmente colaborar.

Muito obrigada um vez mais!

Inês

Excuse me, Ariel Linden

I checked my regression models for the linear assumptions before running the oaxaca. There are heteroscedasticity and autocorrelation in the model of 1 group. Dose the linear assumptions not satisfy so I can't use the command as Jann's instructions? Can you tell me the other decompositions model that are less restrictive? Or could you give me some advices, please?

Thank you very much!

how we run oaxaca in stata 13 as it is not installed

See the link:

https://www.stata-journal.com/sjpdf.html?articlenum=st0151

You can also use R:

https://cran.r-project.org/web/packages/oaxaca/index.html

I also would like to use the Blinder-Oaxaca decomposition in my thesis. The topic of my thesis concerns the tax knowledge gap among managers of enterprises, decomposing the gender knowledge gap would be interesting for my research. Are there any steps to follow or assumptions that need to be met before running linear regression? and can work for a knowledge gap?

Ines,

I never work with SPSS for this type of analysis, but Stata is pretty much easy to work for this techniques, if you give me the name of your variables, I can write codes for you alternatively we can have a zoom meeting and I can explain that to you?

Thanks,

Dr. Zare

Hello

I want to do Blinder-Oaxaca decomposition analysis in R. I tried to do it many times with the help of pdf tutorial but failed. please can anyone help me with the commands?

I want to analyse the performance gap of milk co-operatives of the two regions where I have taken various socio-economics parameters for explaining the performance gap.

STATA is better for such technique

I suggest Stata is pretty good for such analysis.

https://www.youtube.com/watch?v=NQaA8OAlCag&ab_channel=MelissaTrussell

https://www.youtube.com/watch?v=3sbiagPDlR8&ab_channel=EconomicsinManyLessons

https://www.youtube.com/watch?v=2YlDyWFzxPk&ab_channel=SmrutirekhaSinghari

https://www.youtube.com/watch?v=BgBaqP-hCMY&t=14s&ab_channel=TimbulWidodo

Article Decomposition Methods in Economics

@Alam Galicia-Robles thank you for the suggestions. I have finished my PhD thesis six years ago and solved this problem almost eight years ago. Stata worked well. I think this post keeps being useful for people who are struggling with the same question. Thank you all for replying.