Checking model assumptions after xtlogit

[Bryony Porter]

Hello,

I am running a random effects logistic regression model using -xtlogit- in Stata 14 and I want to run the necessary checks on the model but I am unsure of the model assumptions that should be checked and how to do so after xtlogit command.

Here is the code below.

Many thanks.
Bryony

 xtlogit anticholinergic_stopp   age2_centred i.gender i.demreview_previous  cciscore i.e2015_imd_5 i.carehome_variable i.totalmeds_poly pracid_re, intpoin
> ts(20)

Fitting comparison model:

Iteration 0:   log likelihood =  -11049.56  
Iteration 1:   log likelihood = -10972.972  
Iteration 2:   log likelihood = -10971.386  
Iteration 3:   log likelihood = -10971.386  

Fitting full model:

tau =  0.0     log likelihood = -10971.386
tau =  0.1     log likelihood = -10950.723
tau =  0.2     log likelihood = -10938.088
tau =  0.3     log likelihood = -10943.064

Iteration 0:   log likelihood = -10938.088  
Iteration 1:   log likelihood = -10911.103  
Iteration 2:   log likelihood = -10865.115  
Iteration 3:   log likelihood = -10862.984  
Iteration 4:   log likelihood = -10862.983  

Random-effects logistic regression              Number of obs     =    145,581
Group variable: patid                           Number of groups  =     22,448

Random effects u_i ~ Gaussian                   Obs per group:
                                                              min =          1
                                                              avg =        6.5
                                                              max =         14

Integration method: mvaghermite                 Integration pts.  =         20

                                                Wald chi2(12)     =     123.97
Log likelihood  = -10862.983                    Prob > chi2       =     0.0000

---------------------------------------------------------------------------------------
anticholinergic_stopp |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
----------------------+----------------------------------------------------------------
         age2_centred |  -.0144668   .0032455    -4.46   0.000    -.0208277   -.0081058
                      |
               gender |
              Female  |   .0018406   .0524427     0.04   0.972    -.1009453    .1046265
 1.demreview_previous |  -.2865569   .1128348    -2.54   0.011     -.507709   -.0654048
             cciscore |   .0453788   .0135608     3.35   0.001     .0188002    .0719574
                      |
          e2015_imd_5 |
                   2  |   .0837064   .0797953     1.05   0.294    -.0726896    .2401024
                   3  |  -.0150258   .0805759    -0.19   0.852    -.1729517       .1429
                   4  |  -.0027931   .0826149    -0.03   0.973    -.1647153    .1591291
      least deprived  |   .0347228   .0809657     0.43   0.668     -.123967    .1934126
                      |
    carehome_variable |
           care home  |   .0982343   .0565456     1.74   0.082     -.012593    .2090616
                      |
       totalmeds_poly |
                 5-9  |   -.028016   .0632179    -0.44   0.658    -.1519208    .0958889
                 10+  |   .1285853   .0682999     1.88   0.060      -.00528    .2624507
                      |
            pracid_re |   1.306764   .1593021     8.20   0.000      .994538    1.618991
                _cons |  -5.122177   .1006212   -50.91   0.000    -5.319391   -4.924963
----------------------+----------------------------------------------------------------
             /lnsig2u |   .0788858     .08467                     -.0870644    .2448359
----------------------+----------------------------------------------------------------
              sigma_u |   1.040231   .0440382                      .9574017    1.130226
                  rho |   .2475053   .0157695                      .2179059    .2796876
---------------------------------------------------------------------------------------
LR test of rho=0: chibar2(01) = 216.80                 Prob >= chibar2 = 0.000

[Carlo Lazzaro]

Bryony:
a first question about your model specification relates to the seeming omission of the linear term for -age-.

[Bryony Porter]

Hi Carlo,

that would be the -age2centerd- variable which is age centred around the mean.

[Carlo Lazzaro]

Bryony;:
thanks for clarifying.
Hence I would propose a revised question: have you ruled out that centered_squared_age play any role in data generating process?

[Bryony Porter]

My apologies for the missclarificaiton. It is not a squared variable but that is the -age2- is simply the name of a second -age- variable that I generated.

[Carlo Lazzaro]

Bryony:
no need to apologize, as your previous clarification was absolutely effective.
My question about centered_squared_age sprang to my mind considering that a linear and a squared term for -age- can identify a turning point in a regression model.
Hence, just out of curiosity, I would have tweaked your code a bit and given the following one a try:

xtlogit anticholinergic_stopp  c.age2_centred##c.age2_centred i.gender i.demreview_previous  cciscore i.e2015_imd_5 i.carehome_variable i.totalmeds_poly pracid_re, intpoints(20)

[Bryony Porter]

Thank you Carlo. I will take a look at that.

In regards to my initial question

I am running a random effects logistic regression model using -xtlogit- in Stata 14 and I want to run the necessary checks on the model but I am unsure of the model assumptions that should be checked and how to do so after xtlogit command.

Do you know how to check model assumptions after xtlogit?

[Carlo Lazzaro]

Bryony:
all the threads on this topic (spotted via googling with the string -goodness of fit after xtlogit-) point out to -xtlogit postestimation-.