Oaxaca Decomposition in R [closed]

Question

Closed. This question needs to be more focused. It is not currently accepting answers.

---

Want to improve this question? Update the question so it focuses on one problem only by editing this post.

Closed last year.

Improve this question

I would like to make a Oaxaca Decomposition in R. It is used in e.g. labor economics to distinguish explained variance versus unexplained variance, I believe. I have not been able to find a suitable solution in R, and I am rather reluctant to create one myself (I would probably mess it up).

Anyway, the procedure is briefly explained here:

http://en.wikipedia.org/wiki/Ronald_Oaxaca

Stata is blessed with a rather good package for this, but Stata is not easily available to me.

www.stata.com/meeting/5german/SINNING_stata_presentation.pdf

Please note: I have also posted a message on R-help but it has gotten no reply. I hope it is okay to post on this list as well.

Thanks in advance, Rasmus

Edit: I have made the following function, which seems to yield wrong answers (urgh). I tried to follow the Stata link above but it did not work out as I hoped :)

oaxaca <- function (fsex,frace1,frace2) {
  ## First we make regresions
  data1 <- subset(l2,sex==fsex & race==frace1)
  data2 <- subset(l2,sex==fsex & race==frace2)

  mindata1 <- subset(cbind(grade,exp,I(exp^2)),sex==fsex & race==frace1)
  mindata2 <- subset(cbind(grade,exp,I(exp^2)),sex==fsex & race==frace2)

  reg1 <- lm(log(wage)~grade+exp+I(exp^2), data=data1)
  reg2 <- lm(log(wage)~grade+exp+I(exp^2), data=data2)

  ## DECOMPOSITION
  ################

  ## Variables
  gap <- mean(log(wage[race==frace1 & sex==fsex]))-mean(log(wage[race==frace2 & sex==fsex]))

  mean1 <- colMeans(mindata1)
  mean2 <- colMeans(mindata2)

  beta1 <- summary(reg1)$coefficients[,1]
  beta2 <- summary(reg2)$coefficients[,1]
  beta1incep <- summary(reg1)$coefficients[1,1]
  beta2incep <- summary(reg2)$coefficients[1,1]
  beta1coef <- summary(reg1)$coefficients[c(2,3,4),1]
  beta2coef <- summary(reg2)$coefficients[c(2,3,4),1]
  betastar <- .5*(beta1coef+beta2coef)
  betastar2 <- (beta1+beta2)/2

  expl <- sum((mean1-mean2)*beta1coef)
  uexpl <- sum(mean2*(beta2coef-beta1coef))

  pct=expl/gap
  pct2=uexpl/gap

  ## output
  out <- data.frame(Gap=gap,
         Explained=expl,
         Unexplained=uexpl,
         Pct=pct*100)

  return(out)
 }

Answer 1

I have used Oaxaca type decompositions. Did never find any package for R, so I wrote some functions that do it. It is similar to the corresponding package in Stata.

You can find a copy of it in: https://github.com/eyjo/Oaxaca

Note that at the time I was interested in using Fixed Effects (panel data) models, which are not directly compatible with these decompositions. There is an unfinished handler for FE type models, but it should not be used. I meant to create a package out of it, but never got around to it.

Answer 2

The oaxaca package on CRAN can estimate Blinder-Oaxaca decompositions for linear models, as well as producing bar charts that show the results: http://cran.r-project.org/web/packages/oaxaca/index.html

It can also calculate bootstrapped standard errors to provide a sense of how much estimation uncertainty there is.

The vignette gives a detailed description of the package's capabilities, as well as providing some examples of its use. See here: "oaxaca: Blinder-Oaxaca Decomposition in R"