How to predict categorical reponse? I am trying to predict categorical response by using several categorical variables and quantitative variables? I tried linear regression model in R, but I don't think it works well as the response is categorical.
Is there any way to predict categorical response? Any recommended book or online pdf?
Thanks.
 A: I'm surprised no one has mentioned non-regression methods.  While logistic and Polytomous/Multinomial regression are certainly viable options here, they aren't the only ones.  What you are trying to do generally falls under the umbrella of what statisticians generally call "classification methods."  If you do a web search on that term, you'll find all sorts of other non-regression methods for your analysis.  For example, You could try to predict the categories using discriminant analysis, non-parametric nearest neighbor methods (one of my personal favorites), decision tress etc.  
Since you asked for an online book, you could investigate the freely downloadable Elements of Statistical learning, although it's not my my favorite book (since the notation seems to change from page to page), but it does seem to be widely used and addresses many different approaches to your question.
A: It seems like you need to perform logistic regression, and you are interested in doing so in R.
If your dependent categorical variable is ordinal, this may be of use to you: 
http://www.ats.ucla.edu/stat/r/dae/ologit.htm
Otherwise, if your dependent categorical variable is nominal, try this:
http://www.ats.ucla.edu/stat/r/dae/mlogit.htm
A: Yes, the others are right that logistic regression is the area you want to look into. You're basically going to have two options. If you have more than two outcome categories, Ordinal or 'regular old' Multinomial response will be what you want. If your data are in some meaningful order, Ordinal could yield some benefits, but the assumptions needed for that model are rather strict. Be sure to test both approaches. 
Another thing you should consider is what I call 'disaggregated' multinomial. If you have N categories on your response variable, then the multinomial model is N-1 logistic regression equations nested within a single model, where all the right had side variables are the same across equations. If you're up for it, you can customize these equations, so that the equation for response type 1 is different from the equation for response type 2,3...N. The wikipedia article: http://en.wikipedia.org/wiki/Multinomial_logistic_regression#As_a_set_of_independent_binary_regressions on multinomial logistic regression gives a good explanation on how to set this up. This disaggregated approach can be especially useful if accurate prediction is more important than estimating the effect of a specific variable. 
A: As the response variable is categorical, you can consider following modelling techniques:
1) Nominal Logistic 
2) Bootstrap Forest
3) Decision Tree
4) Boosted Tree
5) Neural Net
