Decomposing Erreygers CI - Need results to show Categorical Variables

[Nthato Ras]

Dear All,

I am using Stata 16 and trying to decompose the Erreygers CI. My dependent variable is binary, obese. My independent variables are categorical - quintile, feedingscheme (participating or not), race(African, White, Indian, Coloured), geography(rural, urban) , gender, empl_m (yes or not), educ_m (primary, secondary, tertiary) and wgt_m (underweight, normal, over, obese)

I am able to decompose and get results using the main independent variables - quintile, feedingscheme, gender, race, geo, empl_m, educ_m and wgt_m.

I would like the results to show each category of the independent variables. I tried using i.quintile i.feedingscheme i.gender and so on. It gives an error "factor-variable operators not allowed".

My commands are s follows;

[/CODE]conindex obese [aweight=wt], rankvar(quintile) erreygers bounded limits (0 1)
sca CI=r(CI)
global X i.quintile i.feedingscheme i.gender i.race i.geo i.empl_m i.educ_m wgt_m
qui sum obese [aw=wt]
sca m_obese=r(mean)
qui glm obese $X [aw=wt], family(binomial) link(logit)
qui margins , dydx(*) post

foreach x of varlist $X {
sca b_`x'=_b[`x']
}
foreach x of varlist $X {
qui{
conindex `x' [aw=wt], rankvar(quintile) truezero
sca CI_`x' = r(CI)
sum `x' [aw=wt]
sca m_`x'=r(mean)
sca elas_`x' = b_`x'*m_`x'
sca con_`x' = 4*elas_`x'*CI_`x'
sca prcnt_`x' = (con_`x'/CI)*100
}
di "`x' elasticity:", elas_`x'
di "`x' concentration index:", CI_`x'
di "`x' contribution:", con_`x'
di "`x' percentage contribution:", prcnt_`x'
}
matrix Aaa = nullmat(Aaa) \ (elas_`x', CI_`x', con_`x', prcnt_`x')
}
matrix rownames Caa= $X
matrix colnames Caa = "Elasticity""CI""Absolute""%"
matrix list Caa, format(%8.4f) [/CODE]

The results are as follows:

sca CI=r(CI)

. global X i.quintile i.feedingscheme i.gender i.race i.geo i.empl_m i.educ_m wgt_m

. qui sum obese [aw=wt]

. sca m_obese=r(mean)

. qui glm obese $X [aw=wt], family(binomial) link(logit)

. qui margins , dydx(*) post

.
. foreach x of varlist $X {
2. sca b_`x'=_b[`x']
3. }
factor-variable operators not allowed


My data is as follows:

* Example generated by -dataex-. To install: ssc install dataex
clear
input float obese byte quintile float(feedingscheme gender race geo empl_m educ_m wgt_m)
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . 0 0 1
0 1 0 1 1 0 0 0 1
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . 0 0 1
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . 1 0 2
. . . . . . 1 0 2
. . . . . . 1 0 2
. . . . . . . . .
0 2 1 1 1 0 1 0 2
. . . . . . 1 0 2
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
0 3 0 1 1 0 . . .
0 3 0 0 1 0 . . .
. . . . . . . . .
. . . . . . . . .
. 3 0 1 1 0 . . .
. . . . . . 0 0 3
. . . . . . . . .
. . . . . . . . .
0 3 0 0 1 0 1 0 1
0 3 0 1 1 0 1 0 1
. . . . . . 1 0 1
. . . . . . 1 0 1
0 1 0 0 1 0 . . .
0 1 0 0 1 0 . . .
. 1 . 0 1 0 . . .
0 1 0 0 1 0 . . .
. . . . . . . . .
. . . . . . . . .
0 1 . 1 1 0 . . .
0 1 0 1 1 0 0 0 3
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
0 1 0 0 1 0 0 0 3
0 1 0 1 1 0 0 0 1
0 1 0 1 1 0 0 0 3
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
0 1 0 1 1 0 . . .
. . . . . . . . .
. . . . . . . . .
0 1 0 1 1 0 . . .
. 1 0 0 1 0 . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . 0 1 1
. . . . . . . . .
. . . . . . . . .
. . . . . . 0 0 1
. . . . . . . . .
. . . . . . 0 1 1
0 2 0 1 1 0 0 1 1
. . . . . . 0 1 1
0 1 . 0 1 0 0 0 0
. . . . . . . . .
. . . . . . . . .
0 2 0 0 1 0 1 0 3
0 2 1 1 1 0 1 0 3
. 2 . 1 1 0 1 0 3
. . . . . . . . .
1 2 0 1 1 0 1 0 3
. . . . . . 0 2 2
0 2 0 1 1 0 1 0 3
. . . . . . 0 0 3
. . . . . . 0 0 3
. . . . . . . . .
. . . . . . 0 1 2
. . . . . . 0 1 2
. . . . . . 0 1 3
. . . . . . 0 1 3
. . . . . . 0 0 3
. . . . . . . . .
0 1 1 1 1 0 0 0 3
. . . . . . . . .
. . . . . . 0 1 2
. . . . . . 0 1 2
0 1 0 0 1 0 0 0 3
1 1 0 1 1 0 0 0 3
. . . . . . 0 0 3
. . . . . . 0 0 3
. . . . . . . . .
. 1 1 0 1 0 0 1 2
end
label values obese obese5
label def obese5 0 "Not_Obese", modify
label def obese5 1 "Obese", modify
label values feedingscheme feedingscheme5
label def feedingscheme5 0 "nfscheme", modify
label def feedingscheme5 1 "fscheme", modify
label values gender gender5
label def gender5 0 "Male", modify
label def gender5 1 "Female", modify
label values race race5
label def race5 1 "African", modify
label values geo geo5
label def geo5 0 "Rural", modify

I also tried decomposing using categorical variables but that is not allowed.

Please advise.

Regards
Nthato


.

[Dung Le]

Hi Nthato,

A simple way to address your problem is to create dummy variables instead of using factor notations. Let's take gender as an example, you can create dummies for this variable by typing: tab gender5, gen(gender_). This command will give you two dummies: gender_1 (=1 if male and 0 otherwise) and gender_2 (=1 if female and 0 otherwise). Do the same thing for other variables (e.g., quintile feedingscheme gender race geo empl_m educ_m) that you want to include into the decomposition. Now, your macro X should be:
global X "gender_1 ... put other variables here" (here I treated gender_1 as a compared group and gender_2 as a reference group, you can change this variable to meet your desire).

Hope this helps.

[Nthato Ras]

Hi Dung,

Thanks for the response. It worked!

Just to be clear, I do not list the reference dummy variable?

Regards
Nthato

[Dung Le]

Hi Nthato,

Glad it worked. You don't need to list the reference group, though there is no harm doing so. If you include the reference group, Stata eventually will drop one, either gender_1 or gender_2 (Stata will drop one with a lower proportion between the two).

[Nthato Ras]

Thanks Dung.

Just a follow up question - The total of my percentage contribution of the decomposed CI does not add up to 100% - it is actually over 100%. I checked a few papers and found that most do not add up to 100%, some are under and others above. This boggles my mind because my mathematical understanding tells me that total contribution should be 100%.

These are my results below: The total contribution adds up to 120%. Is this correct?

Elasticity CI Absolute %
quintile 0.0537 0.1416 0.0304 65.0554
fprogram_1 0.0098 0.2221 0.0087 18.6556
gender_1 -0.0080 -0.0053 0.0002 0.3594
race_1 -0.0036 0.4340 -0.0062 -13.2124
race_2 -0.0147 -0.0776 0.0046 9.7420
race_3 -0.0045 0.0897 -0.0016 -3.4684
geo_2 0.0007 0.1624 0.0004 0.9514
level_2 0.0086 -0.0178 -0.0006 -1.3136
level_3 0.0100 0.3638 0.0146 31.2323
employed_2 0.0018 0.1243 0.0009 1.9603
mbmi_2 0.0142 -0.0375 -0.0021 -4.5606
mbmi_3 0.0205 -0.0199 -0.0016 -3.4944
mbmi_4 0.0910 0.0304 0.0111 23.6876

Please advise
Nthato

[Dung Le]

Hi Nthato,

The percentage contribution does not need to add up to 100% because as you may notice the simple formula of the decomposition analysis. That is, decomposition = explained part (120%) + unexplained part (-20%). That says in your case, the excess 20% could be due to unexplained part. Normally, when reporting the decomposition results, you should report both parts, which should add up to 100%.

[Nthato Ras]

Thank you so much Dung.

So, is the -20% my residual?

[Dung Le]

Thank you so much Dung.

So, is the -20% my residual?

Yes. In other words, your residual should be -20%.

[Nthato Ras]

Thanks Dung