Oaxaca decomposition:interpretation of insignificant contributions

[Aleksey Oshchepkov]

Dear all,

I am doing a two-fold Oaxaca-Blinder decomposition in Stata using -oaxaca- command written by Ben Jann (2008, Stata Journal). I am interested in 'detailed' results. I find that controbutions of some varaibles are statistically insignificant. This happens both in the explained part and in the explained part. Should I take those variables into account when interpreting decomposition results? In some cases the contributions are large relative to the gap, but they are insignificant.


Thank you!
Aleksey

[FernandoRios]

Hi Aleksey
My personal advice (what i would do if i were you), is to concentrate on the significant differences, and summarize at the end of your analysis that other variables had large but not statistically significant contributions to the overall gap. (or which ever gap you want to use as reference)
Fernando

[Marcos Almeida]

Well, I’ faced a similar situation in a recent research. My decision was to put all coefficients, CIs and p-values for both ‘explained’ and ‘unexplained’ parcels in a table. I believe it’s up to the reader to check the model, hence I decided to avoid selecting predictors based on p-values or criteria. Hopefully that helps.