xtpoisson on unbalanced panel data output endless iterations and fails to reach to actual estimations (stata2014)

[Flory Kamgaing]

About 13,266 women (15-49 years) are observed (identified by id), from the date of the survey (exact date of interview) till 3 years preceding the survey.. or when they reach 15 years if before.
The period of interest (3years preceding the survey) is split into sub-periods during which the age groups to which women belong are constant. The exposure variable captures the number of years of exposure in each age group, and the variable births, the number of births in each age group.

year is just the whole part of the exact years
(the total period of exposure is less than 3 year for those women who reach 15 years before it end, does it contribute to the problem?)

I used
xtset id year
xtpoisson births i.agegroup2 i.agegroup3 i.agegroup4 i.agegroup5 i.agegroup6 i.agegroup7, re offset(exposure)
xtpoisson births i.agegroup1 i.agegroup2 i.agegroup3 i.agegroup4 i.agegroup5 i.agegroup6 i.agegroup7, re offset( exposure) noconstant


the age groups were dichotomised before xtpoisson

I organised the data base and it looks like

id	exact date of interview	exact age during the interview	exact year	age group	births	exposure	year
1	2015.917	44.25	2015.917	[40-45[	0	3	2015
2	2015.917	28.08333	2015.917	[25-30[	1	3	2015
3	2015.917	41.25	2015.917	[40-45[	0	1.25	2015
3	2015.917	41.25	2013.917	[35-40[	0	1.75	2013
4	2015.917	35.33333	2015.917	[35-40[	0	0.3333321	2015
4	2015.917	35.33333	2014.917	[30-35[	0	2.666668	2014
5	2015.917	16.08333	2015.917	[15-20[	0	1.083334	2015
6	2015.917	34.5	2015.917	[30-35[	1	3	2015
7	2015.917	36.91667	2015.917	[35-40[	1	1.916668	2015
7	2015.917	36.91667	2013.917	[30-35[	0	1.083332	2013
8	2015.917	45.25	2015.917	[45-50[	0	0.25	2015
8	2015.917	45.25	2014.917	[40-45[	0	2.75	2014
9	2015.917	48.08333	2015.917	[45-50[	1	3	2015
10	2015.917	32.08333	2015.917	[30-35[	1	2.083332	2015
10	2015.917	32.08333	2012.917	[25-30[	0	0.9166679	2012
11	2015.917	41.66667	2015.917	[40-45[	0	1.666668	2015
11	2015.917	41.66667	2013.917	[35-40[	0	1.333332	2013
12	2015.917	38.83333	2015.917	[35-40[	1	3	2015
13	2015.917	18	2015.917	[15-20[	0	3	2015
14	2015.917	15.75	2015.917	[15-20[	0	0.75	2015
15	2015.917	19.58333	2015.917	[15-20[	1	3	2015
16	2015.917	30.75	2015.917	[30-35[	1	0.75	2015
16	2015.917	30.75	2014.917	[25-30[	0	2.25	2014
17	2015.917	23.75	2015.917	[20-25[	1	3	2015
18	2015.917	17.33333	2015.917	[15-20[	0	2.333334	2015
19	2015.917	22.75	2015.917	[20-25[	1	2.75	2015
19	2015.917	22.75	2012.917	[15-20[	0	0.25	2012
20	2015.917	35.83333	2015.917	[35-40[	0	0.8333321	2015
20	2015.917	35.83333	2014.917	[30-35[	1	2.166668	2014
21	2015.917	19.58333	2015.917	[15-20[	2	3	2015
22	2015.917	28.08333	2015.917	[25-30[	1	3	2015
23	2015.917	31.33333	2015.917	[30-35[	0	1.333334	2015
23	2015.917	31.33333	2013.917	[25-30[	1	1.666666	2013
24	2015.917	29.16667	2015.917	[25-30[	2	3	2015
25	2015.917	45.58333	2015.917	[45-50[	0	0.5833321	2015
25	2015.917	45.58333	2014.917	[40-45[	0	2.416668	2014
26	2015.917	26.91667	2015.917	[25-30[	0	1.916666	2015
26	2015.917	26.91667	2013.917	[20-25[	0	1.083334	2013
27	2015.917	37.75	2015.917	[35-40[	0	2.75	2015
27	2015.917	37.75	2012.917	[30-35[	0	0.25	2012
28	2015.917	30.91667	2015.917	[30-35[	0	0.916666	2015
28	2015.917	30.91667	2014.917	[25-30[	0	2.083334	2014
29	2015.917	17.33333	2015.917	[15-20[	1	2.333334	2015
30	2015.917	17.91667	2015.917	[15-20[	0	2.916666	2015
31	2015.917	38.41667	2015.917	[35-40[	1	3	2015
32	2015.917	28.91667	2015.917	[25-30[	1	3	2015
33	2015.917	44.41667	2015.917	[40-45[	1	3	2015
34	2015.917	22	2015.917	[20-25[	1	2	2015
34	2015.917	22	2012.917	[15-20[	0	1	2012
35	2015.917	44.58333	2015.917	[40-45[	0	3	2015
36	2015.917	36.33333	2015.917	[35-40[	0	1.333332	2015
36	2015.917	36.33333	2013.917	[30-35[	0	1.666668	2013
37	2015.917	40.41667	2015.917	[40-45[	0	0.4166679	2015
37	2015.917	40.41667	2014.917	[35-40[	0	2.583332	2014
38	2015.917	28.5	2015.917	[25-30[	1	3	2015
39	2015.917	44.58333	2015.917	[40-45[	0	3	2015
40	2015.917	22.66667	2015.917	[20-25[	1	2.666666	2015
40	2015.917	22.66667	2012.917	[15-20[	0	0.333334	2012
41	2015.917	22.75	2015.917	[20-25[	0	2.75	2015
41	2015.917	22.75	2012.917	[15-20[	0	0.25	2012
42	2015.917	26.83333	2015.917	[25-30[	2	1.833334	2015
42	2015.917	26.83333	2013.917	[20-25[	0	1.166666	2013
43	2015.917	15.5	2015.917	[15-20[	0	0.5	2015
44	2015.917	29	2015.917	[25-30[	1	3	2015
45	2015.917	28	2015.917	[25-30[	2	3	2015
46	2015.917	39.83333	2015.917	[35-40[	1	3	2015
47	2015.917	30.41667	2015.917	[30-35[	1	0.416666	2015
47	2015.917	30.41667	2014.917	[25-30[	0	2.583334	2014
48	2015.917	35.5	2015.917	[35-40[	0	0.5	2015
48	2015.917	35.5	2014.917	[30-35[	0	2.5	2014
49	2015.917	43.33333	2015.917	[40-45[	0	3	2015
50	2015.917	18.66667	2015.917	[15-20[	0	3	2015

Thank you for your help

[Carlo Lazzaro]

Flory:
welcome to the list.
It may well be that your problem are related to -i.agegroup*- (admittedly, I fail to get why you had to create so many factor variables for age categories; couldn't you simply create a unique -i.agegroup- including all the different age groups?).
That said, the usual recipe to sniff out the culprit (if any) is to start with a more parsimoniuos model, adding one predictor at time and see when Stata starts to iterate indefinitely.

[Flory Kamgaing]

Thank you for your help.
In fact I created a single variable "age group", and tried first with

xtpoisson births iagegroup

, but had the same issue. I tried then to dichotomised the variable......

[Carlo Lazzaro]

Flory:
do you have little/no variation is some of your predictors?

[Flory Kamgaing]

Yes I think so: The predictors (age group) sometimes remain the same for a women, when she stays in the same age group during the 3 years preceding the survey, in this case the "exposure" is 3 years, some times the women goes throught 2 different age groups during the period of interest ( 3 years preceding the survey), I monitor the time in each age group with the variable "exposure".

I was also wondering if it is not correct to have for some women the period of interest less than 3 years preceding the survey if, going backward, they reach 15 years old before the end of the period of observation (because we do not observe them before the age of 15 years old).

Thank you very much for your time.

Flory

[Flory Kamgaing]

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(id surveyexact yearexact grpage births expo year)
 1 2015.9166 2015.9166 6 0         3 2015
 2 2015.9166 2015.9166 3 1         3 2015
 3 2015.9166 2015.9166 6 0      1.25 2015
 3 2015.9166 2013.9166 5 0      1.75 2013
 4 2015.9166 2015.9166 5 0  .3333321 2015
 4 2015.9166 2014.9166 4 0  2.666668 2014
 5 2015.9166 2015.9166 1 0  1.083334 2015
 6 2015.9166 2015.9166 4 1         3 2015
 7 2015.9166 2015.9166 5 1  1.916668 2015
 7 2015.9166 2013.9166 4 0 1.0833321 2013
 8 2015.9166 2015.9166 7 0       .25 2015
 8 2015.9166 2014.9166 6 0      2.75 2014
 9 2015.9166 2015.9166 7 1         3 2015
10 2015.9166 2015.9166 4 1  2.083332 2015
10 2015.9166 2012.9166 3 0  .9166679 2012
11 2015.9166 2015.9166 6 0  1.666668 2015
11 2015.9166 2013.9166 5 0  1.333332 2013
12 2015.9166 2015.9166 5 1         3 2015
13 2015.9166 2015.9166 1 0         3 2015
14 2015.9166 2015.9166 1 0       .75 2015
15 2015.9166 2015.9166 1 1         3 2015
16 2015.9166 2015.9166 4 1       .75 2015
16 2015.9166 2014.9166 3 0      2.25 2014
17 2015.9166 2015.9166 2 1         3 2015
18 2015.9166 2015.9166 1 0  2.333334 2015
19 2015.9166 2015.9166 2 1      2.75 2015
19 2015.9166 2012.9166 1 0       .25 2012
20 2015.9166 2015.9166 5 0  .8333321 2015
20 2015.9166 2014.9166 4 1  2.166668 2014
21 2015.9166 2015.9166 1 2         3 2015
22 2015.9166 2015.9166 3 1         3 2015
23 2015.9166 2015.9166 4 0  1.333334 2015
23 2015.9166 2013.9166 3 1  1.666666 2013
24 2015.9166 2015.9166 3 2         3 2015
25 2015.9166 2015.9166 7 0 .58333206 2015
25 2015.9166 2014.9166 6 0  2.416668 2014
26 2015.9166 2015.9166 3 0  1.916666 2015
26 2015.9166 2013.9166 2 0  1.083334 2013
27 2015.9166 2015.9166 5 0      2.75 2015
27 2015.9166 2012.9166 4 0       .25 2012
28 2015.9166 2015.9166 4 0   .916666 2015
28 2015.9166 2014.9166 3 0  2.083334 2014
29 2015.9166 2015.9166 1 1  2.333334 2015
30 2015.9166 2015.9166 1 0  2.916666 2015
31 2015.9166 2015.9166 5 1         3 2015
32 2015.9166 2015.9166 3 1         3 2015
33 2015.9166 2015.9166 6 1         3 2015
34 2015.9166 2015.9166 2 1         2 2015
34 2015.9166 2012.9166 1 0         1 2012
35 2015.9166 2015.9166 6 0         3 2015
36 2015.9166 2015.9166 5 0  1.333332 2015
36 2015.9166 2013.9166 4 0  1.666668 2013
37 2015.9166 2015.9166 6 0  .4166679 2015
37 2015.9166 2014.9166 5 0  2.583332 2014
38 2015.9166 2015.9166 3 1         3 2015
39 2015.9166 2015.9166 6 0         3 2015
40 2015.9166 2015.9166 2 1  2.666666 2015
40 2015.9166 2012.9166 1 0   .333334 2012
41 2015.9166 2015.9166 2 0      2.75 2015
41 2015.9166 2012.9166 1 0       .25 2012
42 2015.9166 2015.9166 3 2  1.833334 2015
42 2015.9166 2013.9166 2 0  1.166666 2013
43 2015.9166 2015.9166 1 0        .5 2015
44 2015.9166 2015.9166 3 1         3 2015
45 2015.9166 2015.9166 3 2         3 2015
46 2015.9166 2015.9166 5 1         3 2015
47 2015.9166 2015.9166 4 1   .416666 2015
47 2015.9166 2014.9166 3 0  2.583334 2014
48 2015.9166 2015.9166 5 0        .5 2015
48 2015.9166 2014.9166 4 0       2.5 2014
49 2015.9166 2015.9166 6 0         3 2015
50 2015.9166 2015.9166 1 0         3 2015
51 2015.9166 2015.9166 5 1         3 2015
52 2015.9166 2015.9166 5 0         0 2015
52 2015.9166 2014.9166 4 2         3 2014
53 2015.9166 2015.9166 7 0  .4166679 2015
53 2015.9166 2014.9166 6 0  2.583332 2014
54 2015.9166 2015.9166 1 0   .833333 2015
55 2015.9166 2015.9166 3 0         3 2015
56 2015.9166 2015.9166 2 0       .75 2015
56 2015.9166 2014.9166 1 0      2.25 2014
57 2015.9166 2015.9166 5 0         3 2015
58 2015.9166 2015.9166 7 0 1.0833321 2015
58 2015.9166 2013.9166 6 0  1.916668 2013
59 2015.9166 2015.9166 3 1         3 2015
60 2015.9166 2015.9166 2 0       .25 2015
60 2015.9166 2014.9166 1 0      2.75 2014
61 2015.9166 2015.9166 3 1         3 2015
62 2015.9166 2015.9166 6 0  .9166679 2015
62 2015.9166 2014.9166 5 0  2.083332 2014
63 2015.9166 2015.9166 6 0  2.083332 2015
63 2015.9166 2012.9166 5 0  .9166679 2012
64 2015.9166 2015.9166 2 0  1.166666 2015
64 2015.9166 2013.9166 1 0  1.833334 2013
65 2015.9166 2015.9166 2 0   .333334 2015
65 2015.9166 2014.9166 1 0  2.666666 2014
66 2015.9166 2015.9166 6 1  2.916668 2015
66 2015.9166 2012.9166 5 0 .08333206 2012
67 2015.9166 2015.9166 2 0  2.583334 2015
67 2015.9166 2012.9166 1 0   .416666 2012
end
label values grpage grpage
label def grpage 1 "[15-20[", modify
label def grpage 2 "[20-25[", modify
label def grpage 3 "[25-30[", modify
label def grpage 4 "[30-35[", modify
label def grpage 5 "[35-40[", modify
label def grpage 6 "[40-45[", modify
label def grpage 7 "[45-50[", modify

[Joao Santos Silva]

Flory,

In a regular Poisson model, the ML estimators do not exist if there perfect collinearity for the observations where the dependent variable is positive; I assume a similar thing will happen with RE. Maybe you can start by checking that? For more information on this, please install -ppml- and check the references in the help file.

Best wishes,

Joao

[Flory Kamgaing]

Thank you,
I installed ppml and I am reading the references for ppml.
Please is it an issue if the total lenght of the period of observation is not the same for all the women? Knowing we have an unbalenced panel data, and each women is observed throught years.

Flory

[Joao Santos Silva]

That's not an issue :-)