Why do estimation commands sometimes omit variables?

[Johnny Chin]

Hi guys I am stucked: STATA omitted my variable due to collinearity, but I have no idea why. I ran another regresssion to determine the dependency of the variables but it turns out that R^2 is zero? That means one of the variable is independent? how could i determine which one is independent, and independent to what?

This is the table I ran: the issue is with column 5 where Has_Sus_Effort is omitted: attached

. reg WaterDisclosures STI_30 Produce_Full_SR Has_Sus_Efforts C_or_S1, r
note: Has_Sus_Efforts omitted because of collinearity

Linear regression Number of obs = 135
F( 3, 131) = 9.49
Prob > F = 0.0000
R-squared = 0.1598
Root MSE = .46525

---------------------------------------------------------------------------------
| Robust
WaterDisclosu~s | Coef. Std. Err. t P>|t| [95% Conf. Interval]
----------------+----------------------------------------------------------------
STI_30 | .1026111 .1032129 0.99 0.322 -.1015687 .3067909
Produce_Full_SR | .3506471 .0877845 3.99 0.000 .1769885 .5243057
Has_Sus_Efforts | 0 (omitted)
C_or_S1 | .1776471 .1030142 1.72 0.087 -.0261395 .3814336
_cons | .3453112 .0531926 6.49 0.000 .2400836 .4505388
---------------------------------------------------------------------------------

. reg Has_Sus_Efforts STI_30 Produce_Full_SR C_or_S1, r

Linear regression Number of obs = 140
F( 3, 136) = 0.66
Prob > F = 0.5781
R-squared = 0.0093
Root MSE = .11984

---------------------------------------------------------------------------------
| Robust
Has_Sus_Efforts | Coef. Std. Err. t P>|t| [95% Conf. Interval]
----------------+----------------------------------------------------------------
STI_30 | .0114819 .0084168 1.36 0.175 -.0051628 .0281267
Produce_Full_SR | .0159939 .0114146 1.40 0.163 -.0065792 .038567
C_or_S1 | .0129523 .0093962 1.38 0.170 -.0056291 .0315338
_cons | .9765746 .0165918 58.86 0.000 .9437633 1.009386
---------------------------------------------------------------------------------

Thanks for the help ^^

Johnny

[Carlo Lazzaro]

Johnny:
collinearity (and related omission) is quite common. You may want to take a look at -estat vce- output after -regress- to see what's the matter with that variable.
Usually, there's nothing you can do but changing your model specifications (if feasible).
Your first model is clearly different from the ones you reported in the spreadsheet snapshot (for instance, the number of observations differs, possibly due to missing values and/or different specification in the right-hand side of your regression equations) .
Eventually you do not tell the list if you have done some regress postestimation- tests to check whether your regression model is well or ill-specified.
As an aside, please use CODE delimiters to post what you typed and what Stata gave you back. Thanks.

[Phil Bromiley]

To expand on Carlo's excellent advice, another thing to do is regress the omitted variable on the included variables. You should also look at the FAQ on asking questions. I note that you have lost a lot of observations when you added C_or_S1. It could be that Water_Efforts would be dropped even without C_or_S1 if you use the same sample.

[Johnny Chin]

Johnny:
collinearity (and related omission) is quite common. You may want to take a look at -estat vce- output after -regress- to see what's the matter with that variable.
Usually, there's nothing you can do but changing your model specifications (if feasible).
Your first model is clearly different from the ones you reported in the spreadsheet snapshot (for instance, the number of observations differs, possibly due to missing values and/or different specification in the right-hand side of your regression equations) .
Eventually you do not tell the list if you have done some regress postestimation- tests to check whether your regression model is well or ill-specified.
As an aside, please use CODE delimiters to post what you typed and what Stata gave you back. Thanks.

Thanks Carlo,

I got this after -estat vce-: what does it mean?

J

[Johnny Chin]

To expand on Carlo's excellent advice, another thing to do is regress the omitted variable on the included variables. You should also look at the FAQ on asking questions. I note that you have lost a lot of observations when you added C_or_S1. It could be that Water_Efforts would be dropped even without C_or_S1 if you use the same sample.

Phil, can u explain more? In col 5, I did not ad Water_Efforts in the firsst place

[Carlo Lazzaro]

Johnny:
I forgot to mention that, to get the correlation of the parameters included in -regress-, you should type:

estat vce, corr

otherwise, you'll get the covariance
As far as your results are concerned, since -Has_Sus_Efforts- has been omitted due to collinearity, it is not reported in the -estat vce- matrix.
Try -pwcorr- before running -regress- and investigate the collinearity pattern between -Has_Sus_Efforts- and other predictor(s).