Good afternoon from Switzerland,
I am running Fixed, Random and PA effects models to analyze the effect of being "frail" on the likelihood of being hospitalized (binary 0/1) using a panel dataset.
As my dataset contains many missing values in variables that are very relevant to my model I used a MICE model to impute missing values and then re-fitted the 3 models using the following:
mi estimate, or: xtlogit hospitalization frailty "other regressors", fe
mi estimate, or: xtlogit hospitalization frailty "other regressors", re
mi estimate, or: xtlogit hospitalization frailty "other regressors", pa
However, the Fixed effects model does not converge (not concave).
The same problem with FE would also occur without imputation where the command would be:
xtlogit hospitalization frailty "other regressors", fe or
But in this case, when using the option "difficult" after the comma I am actually able to estimate the coefficients, getting results which put a "." for the variable education. Indeed, whenever I exclude this variable from the model, the model converges without need to use "difficult". So the variable education seems to be the problem. However, this variable is certainly important for my model and would not feel safe to exclude it.
However, the option "difficult" doesn't seem to exist for "mi estimate" (not ML estimation) and therefore I do not get any convergence unless I exclude education from my model.
Would anyone be able to kindly help me understand whether there's a similar option or else how to deal with this problem? Looks like the "dots" option in "mi estimate" could help but the model keeps on running without finding a solution.
Thank you very much in advance for your attention,
Cecilia Luini
Università della Svizzera Italiana, Lugano, Switzerland.
I am running Fixed, Random and PA effects models to analyze the effect of being "frail" on the likelihood of being hospitalized (binary 0/1) using a panel dataset.
As my dataset contains many missing values in variables that are very relevant to my model I used a MICE model to impute missing values and then re-fitted the 3 models using the following:
mi estimate, or: xtlogit hospitalization frailty "other regressors", fe
mi estimate, or: xtlogit hospitalization frailty "other regressors", re
mi estimate, or: xtlogit hospitalization frailty "other regressors", pa
However, the Fixed effects model does not converge (not concave).
The same problem with FE would also occur without imputation where the command would be:
xtlogit hospitalization frailty "other regressors", fe or
But in this case, when using the option "difficult" after the comma I am actually able to estimate the coefficients, getting results which put a "." for the variable education. Indeed, whenever I exclude this variable from the model, the model converges without need to use "difficult". So the variable education seems to be the problem. However, this variable is certainly important for my model and would not feel safe to exclude it.
However, the option "difficult" doesn't seem to exist for "mi estimate" (not ML estimation) and therefore I do not get any convergence unless I exclude education from my model.
Would anyone be able to kindly help me understand whether there's a similar option or else how to deal with this problem? Looks like the "dots" option in "mi estimate" could help but the model keeps on running without finding a solution.
Thank you very much in advance for your attention,
Cecilia Luini
Università della Svizzera Italiana, Lugano, Switzerland.
