I'm running a multilevel logistic regression with SHARE Corona Survey. I would like to find out whether there are differences in mental health between individuals and coutries.
My dependent variables are: nervousness (0=no, 1=yes), sadness (0=no, 1=yes) and sleep problems (0=no, 1=yes).
My independent level-1 variables are: age (50-59, 60-69, 70-79, 80+), gender (0=female, 1=male), education (1=low, 2=intermediate, 3=high), partner (0=no, 1=yes), chronic deseases (0=no, 1=yes), Covid-19-death among relatives or friends (0=no, 1=yes).
My independent level-2 variable is lockdown (created self): 0=no lockdown (Sweden), 1=moderate Lockdown (Germany) and 2=hard lockdown (Italy).
Code for a model with level-1 variables only:
Code:
xtset countr id xtlogit nerv i.age gend i.educ partn chron_d cov_d if sample, or
Code:
Fitting comparison model:
Iteration 0: log likelihood = -4167.7406
Iteration 1: log likelihood = -4059.5243
Iteration 2: log likelihood = -4058.8308
Iteration 3: log likelihood = -4058.8307
Fitting full model:
tau = 0.0 log likelihood = -4058.8307
tau = 0.1 log likelihood = -4031.283
tau = 0.2 log likelihood = -4027.1215
tau = 0.3 log likelihood = -4029.9055
Iteration 0: log likelihood = -4030.9008
Iteration 1: log likelihood = -4020.0158
Iteration 2: log likelihood = -4015.1903
Iteration 3: log likelihood = -4014.9227
Iteration 4: log likelihood = -4014.9034
Iteration 5: log likelihood = -4014.9033
Random-effects logistic regression Number of obs = 6,827
Group variable: countr Number of groups = 3
Random effects u_i ~ Gaussian Obs per group:
min = 1,171
avg = 2,275.7
max = 3,173
Integration method: mvaghermite Integration pts. = 12
Wald chi2(9) = 120.07
Log likelihood = -4014.9033 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
nerv | Odds Ratio Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
age |
60-69 | .8303109 .0776436 -1.99 0.047 .6912635 .9973276
70-79 | .7853784 .0745395 -2.55 0.011 .652067 .9459445
80+ | .8211719 .0865503 -1.87 0.062 .6679111 1.0096
|
gend | .6086208 .0344252 -8.78 0.000 .5447542 .679975
|
educ |
interm | .8127958 .0571963 -2.95 0.003 .7080803 .9329972
high | .8525676 .0716341 -1.90 0.058 .7231185 1.00519
|
partn | .9764865 .0614329 -0.38 0.705 .8632076 1.104631
chron_d | 1.307272 .0777535 4.50 0.000 1.163426 1.468905
cov_d | 1.442366 .1752761 3.01 0.003 1.136678 1.830262
_cons | .5399145 .1109514 -3.00 0.003 .3609148 .807691
-------------+----------------------------------------------------------------
/lnsig2u | -2.405808 .8420484 -4.056192 -.7554233
-------------+----------------------------------------------------------------
sigma_u | .3003208 .1264423 .1315858 .6854281
rho | .0266837 .0218694 .0052355 .1249606
------------------------------------------------------------------------------
Note: Estimates are transformed only in the first equation.
Note: _cons estimates baseline odds (conditional on zero random effects).
LR test of rho=0: chibar2(01) = 87.85 Prob >= chibar2 = 0.000Code for a model with level-1 and level-2 variables:
Code:
xtlogit nerv i.age gend i.educ partn chron_d cov_d i.lockdown if sample, or
Results:
Code:
Random-effects logistic regression Number of obs = 6,827
Group variable: countr Number of groups = 3
Random effects u_i ~ Gaussian Obs per group:
min = 1,171
avg = 2,275.7
max = 3,173
Integration method: mvaghermite Integration pts. = 12
Wald chi2(11) = 303.10
Log likelihood = -4007.9753 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
nerv | Odds Ratio Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
age |
60-69 | .8306602 .0777043 -1.98 0.047 .6915085 .9978133
70-79 | .7867664 .0747093 -2.53 0.012 .6531575 .9477061
80+ | .8230418 .0867843 -1.85 0.065 .6693731 1.011988
|
gend | .6084383 .0344237 -8.78 0.000 .5445753 .6797905
|
educ |
interm | .8202108 .0575785 -2.82 0.005 .7147785 .9411947
high | .8629626 .0722752 -1.76 0.078 .7323215 1.016909
|
partn | .9749875 .0613496 -0.40 0.687 .8618635 1.10296
chron_d | 1.308263 .0778426 4.52 0.000 1.164254 1.470084
cov_d | 1.436974 .1746228 2.98 0.003 1.132427 1.823426
|
lockdown |
1 | .9484019 .0827762 -0.61 0.544 .7992817 1.125343
2 | 1.85537 .157249 7.29 0.000 1.571406 2.190649
|
_cons | .4430057 .0579741 -6.22 0.000 .3427809 .5725351
-------------+----------------------------------------------------------------
/lnsig2u | -45.04146 2.31e+08 -4.53e+08 4.53e+08
-------------+----------------------------------------------------------------
sigma_u | 1.66e-10 .0191342 0 .
rho | 8.35e-21 1.93e-12 0 .
------------------------------------------------------------------------------
Note: Estimates are transformed only in the first equation.
Note: _cons estimates baseline odds (conditional on zero random effects).
LR test of rho=0: chibar2(01) = 0.00 Prob >= chibar2 = 1.000I have difficulty to explain why Prob >= chibar2 became 1.000 after adding a level-2 variable? Have you any suggestions?
Many thanks in advance!
Elena
0 Response to Multilevel logistic regression with cross-sectional data: Why Prob >= chibar2 became 1.000 after adding a level-2 variable?
Post a Comment