Regression With and Without One Observation

[Jantje Beton]

Hello everyone,

Currently I am working on my thesis about Germany. Berlin is of course the capital. I want to do some regressions with and without Berlin, thus temporarily drop the observation of Berlin.
However, I need it in a later stage again. Is there a possibility of running a regression excluding this observation and directly afterwards include it again for another regression?

Or is it better to make a dummy out of this. Then gen Berlin = observation326 --> then 0=without and 1=with.

I prefer the option without creating a new variable.

Thanks in advance.

Jantje

[Clyde Schechter]

So, if we assume that the Berlin observation is identified by a string variable called city, then just add -if city != "Berlin"- to your analysis commands. If there are a large number of commands, making that inconvenient you can do this:

// ANALYSES WITH BERLIN INCLUDED HERE

preserve
drop if city == "Berlin"
// SAME ANALYSIS COMMANDS AS BEFORE GO HERE
restore
// ALL THE DATA ARE NOW BACK

[Jantje Beton]

Thank you very much. Didn't know about the probably basic preserve restore actions.
I was looking from something with "ignore in 326".
Just dropped the observation and restored it and it worked of course.

[Carlo Lazzaro]

Jantje:
Clyde's code is obviously smart and effective, but, as a matter of personal taste, I would strongly favour creating a new variable (I made some mishaps in the past and deleted observations permanently):

gen flag=1 if city=="Berlin"
replace flag=0 if flag==.
reg depvar indepvars if flag!=1

Whenever you want to include Berlin into your regression, just type:

reg depvar indepvars

A simpler approach would be:

reg depvar indepvars if city!="Berlin"

[Eric de Souza]

The advantage of creating a dummy for Berlin is that is allows you to test whether Berlin is really an outlier.