Regression help I am doing a medical study for my thesis concerning the number of organs a donor (numbers ranging from from 0-7) will produce given certain circumstances. I am monitoring their blood sugar, ejection fraction, etc (which are whole numbers) but I also have categorical figures such as Race, Gender, etc. On top of that, I have some yes or no questions as well. I can convert these to binary numbers, but would that mess up a multiple regression or give me bad results. Should I look into another model?
 A: What you are looking for might be an ordinal logistic regression. The outcome variable takes a small number of discrete values that have a defined order. The predictor variables can be continuous, binary, categorical, etc., as for standard linear regression. This page gives examples; searches on the name will provide more. This is a type of generalized linear model. In general, a single model including all predictor variables will be better than several parallel models for different groups, as you will have better pooled estimates of the residual error around the fit and more degrees of freedom for statistical tests.
As a student working on a thesis, you presumably have access to a local statistics department. From the nature of parts of your question (for example, wondering if binary predictor variables will "mess up" a regression, which they will not) it seems that you do not have a lot of experience with regression analysis and statistics. Look for and get some local help to ensure that the statistically based parts of your thesis meet professional standards.
With your data, such statistical help will be particularly important for dealing with the large number of missing data points on ejection fraction. If you wish to include that as a predictor, you will have to do imputation of the missing data or else regression software will remove cases without values for that variable. Handling imputations properly with the preferred multiple imputation methods can take some expertise and experience.
A: You could treat it as a binomial variable with the number of organs out of 7. In Stata, it would look like:
glm y x1 x2 ... , fam(bin 10) link(logit) robust
