How to use psacalc command with the option mcontrol so that to take fixed-effects as unrelated controls ?

[Simplice Adjisse]

Hello Dear All,

I am trying to use psacalc command in Stata to build an argument for omitted variables bias.

My regression takes the form: "reg Depvar X1 X2 ... Xn i.CODE i.ETHNY i.OCC i.REL i.COL, cluster(IDENTIFIER)" (1).
My will is that psacalc command to consider all the fixed effects (i.CODE i.ETHNY i.OCC i.REL i.COL) as unrelated controls and include them in all the regressions, uncontrolled and controlled ones.

So I run the following command "psacalc delta X1, mcontrol(i.CODE i.ETHNY i.OCC i.REL i.COL)".
But Stata returns the following message "Unrelated control not in regression". I've replaced "mcontrol" by "mcontrols" but it did not work neither.

Question 1: Does someone know what is going please ? How can I implement that correctly please?

In addition, after the regression (1) above, I run the command "psacalc beta X1" and Stata gave me the results attached to this post. My concern is that the beta retuned by the command is 4641.34207 which is too big to me. I was thinking beta should be my estimate on X1, which should not be that big! Or maybe I am wrong ?

Question 2: Does someone know how to clearly interpret the results attached to my post please? For example, what does the "Yes" on the results really mean please ?

Any attempt to explain how psacalc command works properly in general would be very appreciated. Thanks in advance for your kind replies. Sorry if some of my questions seem to be stupid.

[Phil Bromiley]

You didn't get a quick answer. You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex. Posting pictures is not encouraged. Being able to replicate your problem is often extremely helpful in solving it.

There are tests for endogeneity available that work with regress. Since you are using a user-written procedure and have not received an answer, it is likely that none of the active folks use that procedure. You may need to contact the authors, although if you search for endogeneity tests and Stata you'll find other ways to handle such issues.

[Sven-Kristjan Bormann]

I have used this command before, so I can offer some help. But as already stated by Phil Bromiley , I need more information about your data and the exact commands that you have run to get your result.

However, I can answer some of your questions. The first error message tells you that you did not include the control variables in your original estimation. To see why this happens, I need to see the exact commands. But my guess is that the command does not understand your factor variable notation (i.CODE i.ETHNY i.OCC i.REL i.COL). Therefore, you have to create new dummy variables for each individual level of your categorical variables and include these new variables in your estimation and in your call to psacalc. You could run something like

foreach var of varlist CODE ETHNY OCC REL COL{
tab `var',gen(`var'_)
}
reg depvar ... CODE_* ETHNY_* ... 
psacalc delta X1, mcontrol(CODE_1 CODE_2 ... )

This code should give you an idea how to proceed.
If you feel courageous, you can modify the code of the psacalc command yourself. I could do it myself, but I would not be willing to do it, at least not now. You could contact the author of the command to suggest some changes to the code to allow factor variable notation. You can find the author in the help file of the command.

To interpret the output of the command, you should consult the paper on which the command is based. I've used this command something like six months ago and already forgot how interpret the results properly. But overall, the bias of the unobservables seem to change the direction of your estimated effect (turning negative to positive or vice versa).

Hope this helps

[Simplice Adjisse]

Dear Phil Bromiley and Sven-Kristjan Bormann, thanks a lot for your respective reply. Your answers are very helpfully indeed!. In fact, I have contacted the Author who suggested the same thing as you, Sven-Kristjan Bormann. And, it actually worked. I am very sorry for the way I posed the problem though. My bad! I promise to improve next. time.