Multilevel Model with ESS Data (Survey Data) doesn't converge

[Emely Ulrich]

To whom it may concern,

I try to calculate a multilevel model for the effect of migration integration policy index (MIPEX) on the individual attitudes toward immigrant across 19 European countries. My dependent variable is an index of seven variables on a 10-point likert scale. All seven variables are uniquely coded and I uses Cronbach's alpha (= 0.86) to from this index. Everything worked so far.

alpha imueclt imwbcnt imtcjob imbleco imwbcrm rlgueim imbgeco, gen(attitudes)
label variable attitudes "Attitudes towards immigrants, higher values meaning more pos. attitudes"

But if want to run an empty multilevel model, the second and concurrent last iteration is backed up. What seems really weird to me, because I have no idea why already the empty model failed to reach convergence. Hence, the problem must be my dependent variable or the country variable?

mixed attitudes || country:, var

My question now is, how can I solve this problem? Without convergence I don't get reliable results.

Thanks in advance for every answer.

Best,
Emely

[Clyde Schechter]

Does the attitudes variable vary within countries? If it does not, the model is unidentifiable and will fail to converge.

If it does, vary within countries, then this is hard to explain. First, I would look at the distributions of atittudes within each country and overall to see if anything is anomalous. If not, you can estimate the same model with

xtset country
xtreg attitudes, re

The estimation algorithm is different from that in -mixed- but the model being estimated is the same. -xtreg-'s estimation does not rely on iterations, so convergence is not an issue. Ordinarily the differences between results obtained with -mixed- and -xtreg, re- are ignorably small.