Collinearity when using Fixed Effects in Panel Data and Insufficient Observations when using Random Effects

[Jose Williams]

Hi,

I am using panel data with rounds in 2004-05 and 2011-12. IDPERSON is the id variable. However, when I specify xtset using the IDPERSON and SURVEY (time-variable), I get a weakly balanced panel and I believe it is causing this error.

The xtdescribe output shows that IDPERSON and SURVEY uniquely identify each observation.

IDPERSON: 1.020e+09, 1.020e+09, ..., 3.402e+11 n = 50958
SURVEY: 0, 1, ..., 1 T = 2
Delta(SURVEY) = 1 unit
Span(SURVEY) = 2 periods
(IDPERSON*SURVEY uniquely identifies each observation)

Distribution of T_i: min 5% 25% 50% 75% 95% max
1 1 1 1 1 1 1

Freq. Percent Cum. | Pattern
---------------------------+---------
25479 50.00 50.00 | .1
25479 50.00 100.00 | 1.
---------------------------+---------
50958 100.00 | XX


Based on similar statalist threads, I tried using: egen id = group(IDPERSON). However, it failed to work as well. I have attached a sample of the dataset using dataex.

My regressions are:
xtreg fertility_decision paid_work_status age SURVEY i.SURVEY##age, re
xtreg fertility_decision paid_work_status age SURVEY i.SURVEY##age, fe

* Example generated by -dataex-. To install: ssc install dataex
clear
input int SURVEY double IDPERSON int(fertility_decision paid_work_status age)
0 1020101002 1 0 45
0 1020102009 1 0 26
0 1020103002 0 0 36
0 1020104002 1 0 40
0 1020105003 1 0 28
0 1020107002 1 0 40
0 1020108002 1 0 30
0 1020109002 1 0 41
0 1020110007 0 0 38
0 1020112002 1 1 26
0 1020113002 1 0 36
0 1020114002 0 0 32
0 1020116004 1 0 28
0 1020117001 1 0 36
0 1020118005 1 0 29
0 1020119002 1 0 28
0 1020120004 0 0 35
0 1020201002 1 0 34
0 1020202002 0 0 44
0 1020203002 1 0 26
0 1020204002 1 0 34
0 1020206004 1 0 40
0 1020207004 1 0 22
0 1020210002 0 0 32
0 1020211002 1 0 46
0 1020214002 1 0 35
0 1020215002 1 0 33
0 1020216002 1 0 42
0 1020217002 1 0 29
0 1020218002 1 0 30
0 1020219002 1 0 36
0 1020220002 1 0 39
0 1020302002 1 0 48
0 1020305002 1 0 41
0 1020306009 1 0 28
0 1020307002 1 0 42
0 1020309001 1 0 49
0 1020310004 1 0 25
0 1020311002 1 0 47
0 1020312002 1 0 42
0 1020313003 1 0 34
0 1020314004 1 0 21
0 1020315003 1 0 26
0 1020316002 1 0 40
0 1020317002 1 0 48
0 1020319004 1 0 39
0 1020401002 0 0 37
0 1020404002 1 0 35
0 1020406002 1 0 40
0 1020407002 1 0 45
0 1020408004 0 0 32
0 1020409002 1 0 38
0 1020412002 1 0 49
0 1020413017 1 0 40
0 1020415002 1 0 37
0 1020417011 0 0 29
0 1020420002 1 0 29
0 1020501002 1 0 25
0 1020502002 1 0 45
0 1020504004 0 0 37
0 1020507004 1 0 43
0 1020509002 1 0 45
0 1020511002 1 0 30
0 1020512003 1 0 29
0 1020513002 1 1 38
0 1020515002 1 0 36
0 1020516002 1 0 43
0 1020517002 1 0 36
0 1020518002 1 0 32
0 1020519002 1 0 44
0 1020520004 0 0 30
0 1020601002 1 1 40
0 1020602002 0 0 47
0 1020603002 1 0 40
0 1020604002 1 0 28
0 1020606002 1 0 38
0 1020607002 1 0 30
0 1020608002 1 0 36
0 1020609002 1 0 27
0 1020610002 1 1 43
0 1020611002 0 0 38
0 1020612004 0 0 30
0 1020614002 1 0 34
0 1020615004 0 0 26
0 1020701002 0 0 40
0 1020702002 1 0 42
0 1020703002 1 1 23
0 1020704002 0 0 48
0 1020705002 0 0 42
0 1020706004 1 0 35
0 1020707002 1 0 35
0 1020708002 0 0 35
0 1020710002 1 0 25
0 1020711002 0 0 38
0 1020712002 1 0 30
0 1020715002 1 0 48
0 1020803003 0 0 16
0 1020804001 0 1 37
0 1020805002 1 0 45
0 1020806002 0 1 30
end
label values SURVEY SURVEY_LABEL
label def SURVEY_LABEL 0 "IHDS 1", modify
label values fertility_decision GR3A
label def GR3A 0 "No 0", modify
label def GR3A 1 "Yes 1", modify
label values paid_work_status WS2Y
label def WS2Y 0 "No 0", modify
label def WS2Y 1 "Yes 1", modify

[Carlo Lazzaro]

Josè:
the issue here is that you have one wave of data only.
Hence, you are dealing with a cross-sectional dataset, not a panel one.
That said, I'm not clear with your interacting -age- with -SURVEY-.

[Jose Williams]

Hi Carlo,

Thank you for your reply. I have two waves of data. It is just that I have converted the time survey to a dummy having values 0 for 2004-05 and 1 for 2011-12.

I also tried removing the interaction term but the problem remained.

[Jose Williams]

Carlo Lazzaro Sorry I'm new to this forum. I should have tagged you for the above comment as well. It is just that I have converted the time survey to a dummy having values 0 for 2004-05 and 1 for 2011-12. The dataex example used only round 1 values and somehow I cannot use randomtag as I keep getting the varlist error.

However, I think so I may have found the error in the model but cannot figure out how to solve it. When running fixed effects, my number of observations is the same as the number of groups. This should not be the case but I cannot solve it.

Hope this helps!

[Carlo Lazzaro]

Josè:
again, as per your screenshot (BTW: please note that, screenshots are discouraged on this forum, for reasonds well explained in the FAQ, that you're kindly recommend to read and act on in your future posts. Thanks) the issue here is that you have one wave of data only.
This is self-apparent from -xtreg- outcome: the average number of observations per panel=1 (and min and max=1 either). Thus clearly means that you have one wave of data only (cross-sectional dataset); thererfore, no wonder that your -xtreg- table is basically empty.