Deviation (Effect) coding in regression in Stata

[Katherine Picho]

Hello:

I'm using stata 12. I have 2 questions:

1. How do I code variables in Stata using effect/deviation coding? My searches on this issue turned up using xi3: which is not fully supported in recent Stata versions.
I have 2 variables that I'd like to 'effect' code

gender: males = -1 and females = 1
coed: The variable coed is derived from 2 variables: sex and type of school such that, 0 = males, coed -1 = females coed and 1 = females same sex

Admittedly I've never used effect coding before (dummy coding is what I'm used to). So I did the following in Stata to code my variables (probably a bad idea...I will let you be the judge). My question is: did I do this correctly?? If not, what is the correct way to do it?

gen gender = -2 if (sex==0)   /* where for variable  sex : 0= males, 1 = females*/
replace gender = 1 if (sex==1)

gen bcoed = 0 if (sex==0) & (schtype ==0)

gen gcoed = -1 if (sex==1) & (schtype==0)

gen gss = 1 if (sex==1) & (schtype==1)

egen coed = rowmax (bcoed gcoed gss)

2. I attempted to run a regression with these variables predicting a dependent variable (tscore) [ this is really just to check for redundancy & eliminate one of the group variables in a model I'm attempting to build-- so don't worry too much about the content for now). However, when I do this, stata gives me an error message that factor variables cant contain negative values. If this is the case, then how does stata handle effects coded variables in regression analyses?

 regress tscore i.gender##i.coed
gender:  factor variables may not contain negative values
r(452);

I did recode gender to be just 0/1 and it returned the same message for coed

 regress tscore i.gender##i.coed
coed:  factor variables may not contain negative values
r(452);

Any help on this would be much appreciated. Thank you!

Katherine Picho

[Andrew Lover]

Check out the older/deprecated -xi3- command:

http://www.stata.com/statalist/archi.../msg00597.html
http://fmwww.bc.edu/repec/bocode/x/xi3.ado

However you can accomplish all the same things (and more) using -margins- and -contrast- :

http://www.stata.com/statalist/archi.../msg00603.html

[Phil Clayton]

Here's a basic example:

sysuse auto, clear
regress price i.rep78
contrast g.rep78