One observation per factor level combination in regression

Hi... I was wondering if you can perform multiple linear regression with a mix of categorical and quantitative variables, if you only have one observation per combination of the levels of your factors in the regression model? And how do you interpret predictions? Thanks in advance!


Less is more. Stay pure. Stay poor.
I would imagine tests around those groups would have limited power. Logistic reg will give you a warning "quasi seperation " however linear will run but may be give you large standard errors. Also unadjusted r^2 may be higher just because of the numbers of predictors.

I am guessing it will let you do it, you just have to proceed with caution.
Thank you! I was wondering about degrees of freedom here. What if there are many factor level combinations therefore many groups, and standard errors are small (within desirable limits)? Would that help mitigate? I suppose I would just proceed with caution if at all... I was thinking of creating a predictive model where I could make predictions for a group. The group is defined by the combination of factor levels. So my thinking was I want to be able to plug in what the level of the factor actially is into the model, and have a coefficient for each factor level. Thanks again... Will have to think about it some more! Just because I can plug it into a program doesn't mean I should! ;)