Hi there,

I am attempting to run an xtlogit model for panel data. There are 8,937 observations with a binary dependent variable (0,1) where 0 has n=802 participants and 1 has n=114 participants. There are 6 independent variables, one of which is an integer (gender; coded as 0,1) and the rest are continuous. My outcome of interest is standardised test z-scores (std_Comm_) over 3 years.

My script is as follows:

xtlogit var1 c.age_ i.gender c.birthweight c.familyincome_ c.maternalhighestqualification c.std_Comm_, i(id_new) or

When I run the model, I get this error message:

Fitting comparison model:

Iteration 0: log likelihood = -1371.7743
Iteration 1: log likelihood = -1326.8958
Iteration 2: log likelihood = -1325.9652
Iteration 3: log likelihood = -1325.9647
Iteration 4: log likelihood = -1325.9647

Fitting full model:

tau = 0.0 log likelihood = -1325.9647
tau = 0.1 log likelihood = -1287.0885
tau = 0.2 log likelihood = -1245.3778
tau = 0.3 log likelihood = -1200.9184
tau = 0.4 log likelihood = -1153.703
tau = 0.5 log likelihood = -1103.5098
tau = 0.6 log likelihood = -1049.7339
tau = 0.7 log likelihood = -991.08284
tau = 0.8 log likelihood = -925.17559

Iteration 0: log likelihood = -991.07902
cannot compute an improvement -- discontinuous region encountered
r(430);

.I have tried the following approaches to get the model to run:

The assists
Checked the gradient of the variables
Check the collinearity metrics
Tried a number of maximization algorithms (all coming back with the same answer).
Changed the tolerance levels for acceptable levels of concave.
Edited the var_1 variable type.
Tried a one at a time analysis for each predictor.
Changed the z scores to t scores (small values appear to affect location of the maximum likelihood but I have never had this issue)

When I run the model with both logit and xtprobit I can get results, however I would like to use xtlogit in-line with my other analyses,

Any advice would be appreciated,
Eleanor