regression on categorical variables.

[Beunhez]

hello,
i want to do some regressions on categorical variables, both the dependend and independend variables are categorical.
can i just use the regress command. and can the outcomes be interpreted the same (so if the dependend variable rises with 1 the independend variable rises with the coëfficiënt)
can i make logaritmic functions of these variables, and can i interpreted them the same (so if the dependend variable rises with 1 percent, the independend variable rises with the coëfficiënt as a percentage)

the variables are:
ch11e004 (about hapiness) with 5 categories (very sad, sad, normal, happy, very happy)
and cw12e138 (about working hours) with 4 categories ( works shifts, works sometimes outside office hours, works regularly outside office hours, never works outside office hours)
or cw12e140 (about working hours) with 4 categories ( works never in weekend, works sometimes in weekend, works every few weeks in weekends, works almost alltime in weekend) this variable is a little bit more useful i guess because it is ordinal?

i used egen newvar = group(cw12e140/138/ch11e004) to make them numeric, but they still are categorical

i know how to do normal regressions with continious variables, but i don't what to do with these categories. and can i make numbers for cw12e138 because it can't be said if working shifts is "higher" than not working outside office hours or work a lot outside office hours?

thanks in advance for your tips and advice
kind regards Erwin

[Richard Williams]

It sounds like you have a 5 category ordinal dependent variable. If you used OLS regression with it, you wouldn't be the first. But I would suggest reading up on logistic regression and ordinal regression first. Just reading through the manual entries for logit and ologit and mlogit would be a good start. Or get a copy of Long and Freese's book:

http://www.stata.com/bookstore/regre...ent-variables/

A new edition is in the works.

[Beunhez]

thank you for answering.
so then i should use: ologit ch11e004 i.cw12e138? can i take the logaritmic functions en do ologit lnch11e004 i.lncw12e138 and say the increase of the coëfficiënt in percentages?

[Maarten Buis]

No, you cannot use the logarithm trick in ordinal regression as you could (but probably shouldn't, see http://blog.stata.com/2011/08/22/) in linear regression. For the proper way of interpreting these models you really need to read up on these techniques. Rich already gave you an excelent suggestion on where to start.