Interaction term is omitted because of collinearity

[Evans Adjei]

Good day everyone.

I am running an experimental analysis (difference-in-differences) and I have a little but a big problem. The interaction term (treatment year and treatment group in this case ty##tg) is omitted because of collinearity. I have tried every possible means to address the problem but to no avail (note: there are enough observations for the interaction term and even after implementing the filters (ifs)).

Please below is the command line the note from stata.

reg arbink ty##tg man age i.sni2. i.FA. i.year. i.educ if inlist(qun, 1,5) & bet2050==1, vce (cluster PID)

note: ob.ty#1.tg identifies no observations in the sample

note:1.ty#tg omitted because of collinearity

Please, any help as to how to deal with this?

Thank you in advance.

Evans.

[Rich Goldstein]

try putting "i." in front of each of ty and tg if they are categorical; if either is quantitative, put "c." in front of it; here are a couple of examples:

c.ty##i.tg
i.ty##i.tg

[Evans Adjei]

Hej Rich,

They are both categorical (dummies) variables, hence I have tried

i.ty##i.tg

But I got the same problem.

By the way thanks

[Rich Goldstein]

you have not supplied example data using -dataex- which would reduce me to guessing; rather than that, please show example data; also show the results of

d ty tg
ta ty tg

[Evans Adjei]

HEJ Rich,

I am sorry to say I'm working with a restricted register data, which I cannot copy any part.

Hopefully this image will help.

Thank you

[Rich Goldstein]

1. the FAQ explains about using fake data similar to yours if your data is confidential

2. I see no reason why the results of describe and of tabulate cannot be given

3. what you supplied is not a replacement and does not help me; maybe someone else can help

[Evans Adjei]

Hej Rich,

I am sorry. I probably misunderstood your statement, about supplying data, of course, I can supply the results of describe and tabulate. I will post them...

[Evans Adjei]

So, I have this data sample from my data.

dataex Ty TgNonEU if ar==2000

 
clear



input
byte(Ty
TgNonEU)

1 1



1 1



1 1



1 1



1 1



1 1



1 1



1 0



1 0



1 0



1 0



1 0



1 1



1 1



1 1



1 1



1 1



1 1



1 0



1 0



1 1



1 0



1 0



1 0



1 0



1 0



1 0



1 0



1 0



1 0



1 0



1 0



1 1



1 0



1 0



1 0



1 0



1 1



1 0



1 0



1 0



1 0



1 1



1 0



1 0



1 0



1 0



1 1



1 0



1 0



1 0



1 0



1 1



1 0



1 1



1 0



1 0



1 1



1 0



1 1



1 1



1 1



1 1



1 1



1 1



1 1



1 0



1 1



1 1



1 1



1 0



1 0



1 0



1 0



1 0



1 1



1 0



1 0



1 1



1 0



1 0



1 0



1 1



1 0



1 0



1 0



1 1



1 1



1 1



1 1



1 1



1 0



1 1



1 0



1 0



1 0



1 0



1 0



1 0



1 0

d Ty TgNonEU

 
variable name
storage    type
display format

Ty
       byte
%8.0g

TgNonEU
       byte
%9.0g

ta Ty TgNonEU

 


TgNonEU


Ty
0
1
Total

0
12,102,436
0
12,102,436

1
37,191,054
 7,693,900
44,884,954

Total
49,293,490
  7,693,900
56,987,390

Any help in relation to #1

[Rich Goldstein]

what does the blank directly under "Ty" in the output (it would be preferable to copy and paste the exact output as shown by Stata in the results window) mean? if it means "missing" that is your answer; my guess, however, is that the "0" that should have been there is attatched to the 12 million shown to its right; in that case the answer is the "0" in the same row but the second column

[Evans Adjei]

It is not 'missing', it is the control year (the 0 was right aligned).

[Evans Adjei]

Ok, so my question now is, does no observation between control year (Ty=0) and treated group (Tg=1) have is implications for the interaction between the treated year (Ty=1) and the treated group (Tg=1)?

That seems to be the problem.