Cross-classified multilevel logistic analysis. Please help me!

[CarolineOlsson]

Hi you Statalister!

Im reaching out for some much needed help on how to run a proper and correct cross-classified multilevel logistic analysis using Stata 13.

The model is the following:

Response:

employment status (1=in employment, 0= unemployed)


Unit identifiers:

• Immigrants (level 1) (n=1550)
• Origin regions (level 2) (three categories)
• Destination countries (level 2 alt. 3) (three categories)

The following predictor variables are to be included in the models:

Individual variables (level 1); (1) gender, (2) age, (3) time in destination country, (4) immigrant status, (5) educational level.

Origin-specific variables (level 2); (1) Islam as main religion in origin region,(2) language most spoken at home differ from that of that most spoken in destination country, (3) average educational level/origin spec. communities.

Destination-specific variables (level 2 alt. 3); (1) Symbolic boundaries against immigrants/immigration, (2) employment status amongst natives.

When using the xtmelogit-command in Stata, some of the interations cannot concave. Is there anyone out there with knowledge in cross-classified multilevel logistic analysis that perhaps could help me.

+ I have tried to use MLwiN but I have no earlier experience of the problem and cannot figure out how it works.
+ I use multilevel data over three countries from the European Social Survey, round 3, 4, 5.



Best,
Caroline

[Brendan Halpin]

I can see why you are drawn to multi-level models here (most importantly, you have variables measured at group level), but you have very few categories in the two group levels. Some people will argue that even the 30 or so countries in the full ESS are not enough, so 3 is really pushing it!

[CarolineOlsson]

Thanks for replying Brendan Halpin.
Is there any potential way one could come to solve this problem?

[Nick Cox]

More data....

[Brendan Halpin]

I'd experiment with either ordinary logit or xtlogit clustering on the cross-classification of the origin and destination categories.

egen xclass = group(orig dest)
logit empstat rhsvars
logit empstat rhsvars, cluster(xclass)
xtlogit empstat rhsvars, i(xclass)

The variable xclass is the cross-classification of the three origin and three destination categories, and should have nine values. I haven't thought it through completely but it does give Stata a way to take account of the cross-cutting clustered nature of your data.