Is there a way to get overall effects for a KHB decomposition when I have a categorical key variable?

[patricia oreilly]

Dear all,
I am currently running a KHB mediation analysis with a continuous DV (cognitive development), a categorical key variable (education) and a continuous mediating variable (centered_HLA). The model provides results for each level of education with the highest level of education (4) specified as the reference category. However, is it possible to get the overall direct and indirect effect for education without it being decomposed by level? If I put education in without the factor ( i.) I get a result for education (without it being split into categories) but I assume that the KHB then treats education as a continuous variable.

Here is the command I have used:

khb reg b2_nvtscore ib4.education4 || centered_HLA if sample==1, conc(female nonenglish) summary disentangle (see output below)

I will really appreciate any help on this issue. Many thanks.

	(1)
	b2_nvtscore
main
Reduced	-7.239***
	(-14.47)
Full	-6.786***
	(-13.54)
Diff	-0.453***
	(-4.19)
2.education4
Reduced	-3.530***
	(-11.03)
Full	-3.230***
	(-10.07)
Diff	-0.300**
	(-2.85)
3.education4
Reduced	-1.618***
	(-4.72)
Full	-1.452***
	(-4.23)
Diff	-0.166
	(-1.60)
4b.education4
Reduced	0
	(.)
Full	0
	(.)
Diff	0
	(0.00)

[daniel klein]

Karlson (2013) describes the proper way of doing this. He has Stata code in the appendix; that code, however, probably needs lots of adjustments to your specific situation. So, unfortunately, it is not something that you get by simply pushing a button.

Best
Daniel


Karlson, K. 2013. Summarizing primary and secondary effects. Research in Social Stratification and Mobility, 33, 72--82.