help with the logitfe command

[Luis Mijares Castaneda]

Hello, I am trying to use the -logitfe- command to run the model shown below. I have dataset with 3 periods of data from a survey, and I want to run a logit fixed effects regression using one lagged time period of the dental_Wave outcome variable, but I get the error that can be seen below. I wonder if my data is not structured properly; I have included a -dataex- of the data, the dataset was originally in wide format, so I transformed it to long data format, it's also possible that the variables I selected don't have enough variation between time periods and produce the problems below. Any guidance would be greatly appreciated, thank you.

.
. logitfe dental_Wave ///
>     L1.dental_Wave ///  
>     c.depressive_symptoms_Wave3  c.depressive_symptoms_Wave4 c.depressive_symptoms_Wave5 /// <- 1 to 4 lags of depression
>     c.age_Wave3 c.age_Wave4 c.age_Wave5 i.smoked_marijuana_alt_Wave3 i.smoked_cigarettes_Wave4 ////
>         i.smoked_cigarettes_Wave5 c.smoked_cigarettes_Wave3 c.smoked_cigarettes_Wave4 c.smoked_cigarettes_Wave5 ///
>     i.alcohol_Wave3 i.alcohol_Wave4 i.alcohol_Wave5 i.education_Wave3 i.education_Wave4 i.education_Wave5, ///
>     analytical ieffects(yes) teffects(yes)

Computing uncorrected fixed effects estimator

note: 11.smoked_cigarettes_Wave5 != 0 predicts failure perfectly;
      11.smoked_cigarettes_Wave5 omitted and 2 obs not used.

note: 19.smoked_cigarettes_Wave5 != 0 predicts failure perfectly;
      19.smoked_cigarettes_Wave5 omitted and 4 obs not used.

note: smoked_cigarettes_Wave4 omitted because of collinearity.
note: smoked_cigarettes_Wave5 omitted because of collinearity.
note: multiple positive outcomes within groups encountered
note: multiple positive outcomes within time periods encountered
note: 5725 groups (11393 obs) dropped because of all positive or
      all zero outcomes
note: depressive_symptoms_Wave3 omitted because of no within-group variance
note: depressive_symptoms_Wave4 omitted because of no within-group variance
note: depressive_symptoms_Wave5 omitted because of no within-group variance
note: age_Wave3 omitted because of no within-group variance
note: age_Wave4 omitted because of no within-group variance
note: age_Wave5 omitted because of no within-group variance
note: smoked_cigarettes_Wave3 omitted because of no within-group variance
Iteration 0:  f(p) = -3941.7763  
Iteration 1:  f(p) = -2425.3123  
Iteration 2:  f(p) = -2281.2968  
Iteration 3:  f(p) = -2258.8606  
Iteration 4:  f(p) = -2254.2237  
Iteration 5:  f(p) = -2253.2433  
Iteration 6:  f(p) =  -2253.023  
Iteration 7:  f(p) = -2252.9684  
Iteration 8:  f(p) = -2252.9575  
Iteration 9:  f(p) = -2252.9551  
Iteration 10: f(p) = -2252.9546  
Iteration 11: f(p) = -2252.9545  
Iteration 12: f(p) = -2252.9544  

Computing analytical correction
initial values not feasible
r(1400);

* Example generated by -dataex-. For more info, type help dataex
clear
input long aid byte year float dental_Wave
1 3 1
1 4 1
1 5 1
2 3 0
2 4 0
2 5 0
3 3 0
3 4 0
3 5 1
4 3 0
4 4 0
4 5 0
5 3 1
5 4 0
5 5 0
6 3 0
6 4 0
6 5 1
7 3 1
7 4 1
7 5 0
8 3 0
8 4 1
8 5 1
9 3 0
9 4 1
9 5 1
10 3 0
10 4 0
10 5 1
11 3 1
11 4 0
11 5 0
12 3 1
12 4 0
12 5 1
13 3 1
13 4 0
13 5 0
14 3 1
14 4 0
14 5 0
15 3 1
15 4 1
15 5 1
16 3 1
16 4 0
16 5 1
17 3 0
17 4 0
17 5 0
18 3 1
18 4 0
18 5 1
19 3 0
19 4 0
19 5 1
20 3 1
20 4 1
20 5 1
21 3 0
21 4 0
21 5 1
22 3 1
22 4 1
22 5 1
23 3 1
23 4 1
23 5 0
24 3 0
24 4 0
24 5 1
25 3 1
25 4 1
25 5 0
26 3 1
26 4 0
26 5 0
27 3 0
27 4 0
27 5 0
28 3 1
28 4 0
28 5 0
29 3 0
29 4 1
29 5 1
30 3 1
30 4 1
30 5 1
31 3 1
31 4 1
31 5 0
32 3 1
32 4 0
32 5 0
33 3 1
33 4 1
33 5 0
34 3 0
end
label values aid id
label def id 1 "10506342", modify
label def id 2 "11316958", modify
label def id 3 "11574211", modify
label def id 4 "12504295", modify
label def id 5 "12572562", modify
label def id 6 "12578037", modify
label def id 7 "13676851", modify
label def id 8 "13715895", modify
label def id 9 "14316051", modify
label def id 10 "14574194", modify
label def id 11 "14590357", modify
label def id 12 "14676257", modify
label def id 13 "14714942", modify
label def id 14 "15574600", modify
label def id 15 "15578543", modify
label def id 16 "15676328", modify
label def id 17 "17254985", modify
label def id 18 "17316657", modify
label def id 19 "17508726", modify
label def id 20 "17608972", modify
label def id 21 "17676352", modify
label def id 22 "18614276", modify
label def id 23 "18676822", modify
label def id 24 "18710296", modify
label def id 25 "19508388", modify
label def id 26 "19606222", modify
label def id 27 "19614572", modify
label def id 28 "20540757", modify
label def id 29 "20574505", modify
label def id 30 "20574622", modify
label def id 31 "20676929", modify
label def id 32 "22574932", modify
label def id 33 "22712340", modify
label def id 34 "24614574", modify
label values dental_Wave dental
label def dental 0 "0.More than 1 year", modify
label def dental 1 "1.Last Year", modify

[Felix Bittmann]

Your dataex only contains 3 variables but you use many more in your command. I suspect that not all variables are actually in long format. Make sure that all controls are also in long. Variable names like

i.smoked_cigarettes_Wave5 c.smoked_cigarettes_Wave3 c.smoked_cigarettes_Wave4

Give me the impression that these are still in wide. This will not work.

[Luis Mijares Castaneda]

Felix Bittmann

I changed the data to make all the variables in long format, but I still get problems when using the -logitfe- model, I want to include lags of the Dental_Wave outcome variable

. 
. logitfe dental_Wave ///
>     L1.dental_Wave ///  <- 1 to 4 lags of outcome
>     c.depressive_symptoms_Wave c.age_Wave i.smoked_marijuana_alt_Wave i.smoked
> _cigarettes_Wave ///
>     i.alcohol_Wave i.education_Wave, ///
>     analytical ieffects(yes) teffects(yes)

Computing uncorrected fixed effects estimator

note: multiple positive outcomes within groups encountered
note: multiple positive outcomes within time periods encountered
note: 5913 groups (11625 obs) dropped because of all positive or
      all zero outcomes
Iteration 0:  f(p) = -3957.5092  
Iteration 1:  f(p) = -2372.2041  
Iteration 2:  f(p) = -2222.6291  
Iteration 3:  f(p) = -2196.7665  
Iteration 4:  f(p) = -2191.5725  
Iteration 5:  f(p) = -2190.3398  
Iteration 6:  f(p) = -2189.9916  
Iteration 7:  f(p) = -2189.9015  
Iteration 8:  f(p) = -2189.8768  
Iteration 9:  f(p) = -2189.8702  
Iteration 10: f(p) = -2189.8687  
Iteration 11: f(p) = -2189.8685  
Iteration 12: f(p) = -2189.8685  

Computing analytical correction
initial values not feasible
r(1400);

[Felix Bittmann]

This looks already much better. However, where exactly the problem lies is not possible to say currently. I suggest to start with a very simple model. No lags, no controls. Find something that works. Then you can add more controls and build a more complex model. This approach will help you find the issue. How many total individuals and data points are in your dataset?

[Luis Mijares Castaneda]

My dataset consists of 3 time periods and Observations: 27,447. When I include the analytical correction, I get the problem of numerical overflow.