omitted variable

[Sjaakie de Boer]

For my analysis (random effects panel data), I am using a dummy variable as independent variable along with some control variables. However, when I run my analysis Stata keeps ommiting the variable from the output and I don't understand why. What are possible causes of this ommitance? Or would you need more information for that? Thanks in advance.

[Clyde Schechter]

At a minimum, you need to show the exact regression command you used and the complete output from that command. And if your data was -xtset- or -tsset-, show that exact command as well. While that might be sufficient to answer your question, it might also be necessary to see some example data from your data set, so it probably makes sense to show that as well. Place the command and output between code delimiters so that it will align readably, and use the -dataex- command to show example data. If you are not familiar with code delimiters or -dataex-, please read the Forum FAQ, with special attention to #12 for information.

[George Ford]

Your dummy is perfectly correlated with another variable or set of variables.

[Sjaakie de Boer]

crisisJST is my dummy variable. I want to test long term returns and the impact crises have on them.


**xtscc long_term_return_70 crisisJST rgdpbarro cpi ltrate, re**



Regression with Driscoll-Kraay standard errors Number of obs = 27
Method: Random-effects GLS regression Number of groups = 11
Group variable (i): country_num Wald chi2(3) = 46.91
maximum lag: 2 Prob > chi2 = 0.0000
corr(u_i, Xb) = 0 (assumed) overall R-squared = 0.0358

------------------------------------------------------------------------------
| Drisc/Kraay
long_term~70 | Coefficient std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
crisisJST | 0 (omitted)
rgdpbarro | .0001311 .0002754 0.48 0.654 -.0005768 .000839
cpi | -4.03e-06 .0005704 -0.01 0.995 -.0014703 .0014623
ltrate | -.0032803 .0019396 -1.69 0.152 -.0082661 .0017055
_cons | .122408 .0843691 1.45 0.207 -.0944696 .3392856
-------------+----------------------------------------------------------------
sigma_u | .02347922
sigma_e | .00351608
rho | .978066 (fraction of variance due to u_i)
------------------------------------------------------------------------------
************************************************** ************************************************** ****************************************
. dataex crisisJST

----------------------- copy starting from the next line -----------------------

* Example generated by -dataex-. For more info, type help dataex
clear
input byte crisisJST
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
end

------------------ copy up to and including the previous line ------------------

Listed 100 out of 1100 observations
Use the count() option to list more

[George Ford]

You're estimating with 27 observations, and 11 groups? But you also say you have 1,100 observations. (xtscc wants a large T, by the way).

type mdesc to see if you have a lot of missing on some variables.

[Clyde Schechter]

The variable crisisJST must be colinear with some other variable(s) in the model. Your data example is not useful because you showed only the crisisJST variable. To diagnose what the colinearity is, it is necessary to have all of the model variables included in the example.

[Sjaakie de Boer]

You're estimating with 27 observations, and 11 groups? But you also say you have 1,100 observations. (xtscc wants a large T, by the way).

type mdesc to see if you have a lot of missing on some variables.

I think here may be my mistake? Because I want to test long run returns, and I have yearly returns for several countries. I made a new variable ( long_term_return_70) which is the average of all returns over the 70 year period. while all the other variables are still yearly. I am really struggling with forming my model correctly.

[George Ford]

That ain't gonna work. You've got 1 outcome for each country.

[Sjaakie de Boer]

I'm sorry to ask. But I think it is not possible for you to suggest how I should adress this? I am feeling really lost.

[George Ford]

This problem sounds a lot like a financial event study. You've got a Y (returns) and an event (crisis). With that method, you can compute cumulative abnormal returns over extended periods. But 70 years is too long. A lot happens over time and the effect of a crisis will eventually decay.

A very simple setup would be to dummy the crisis year and each of the 10 years following. The coefficients should move to zero over time. That's not the model I'd use (or maybe I would), but it would give you a sense of what the process looks like. But in this setup, all crisis are the same. You could use different dummies for different types of crisis. I suspect a dynamic model may be useful, where you lag the outcome. This requires particular techniques (xtabond2, for example). https://www.stata-journal.com/articl...article=st0159

You might look for papers that quantify the impact of a crisis. Say, the effect on GDP of a recession, or a war. That literature is probably fairly rich and the modeling may have been mostly settled. Some citations here (https://www.cato.org/cato-journal/wi...uantifying-war).