# Regarding a categorical dependent variable

Say someone is analyzing a basketball tournament that includes 64 teams and the dependent variable is the number of wins in the tournament by each team. We'll have a bunch of independent variables.

My question: Should the dependent variable (wins) be a categorical variable with values 0 through 6. Or continuous? What kind of regression would you use?

Thanks!

• If each variable represents "the number of" wins for a team then it is clear that it has to be discrete. Also, why the range of the values should be 0 through 6? – Sobi Dec 16 '15 at 22:14

## 1 Answer

You can model "wins in the tournament" as a continuous variable. You will have to apply a post-processing step to your output where you round the final number to a meaningful integer number of wins.

This method is restricted to regressions.

Another possibility is use a multinomial logistic regression. This will ignore the ordering of the # of wins and treat each possible outcome of # of wins as a distinct class. In this case, you would have 7 classes. Again, the multinomial logistic regression assumes the outcome variable is nominal, not ordinal i.e 0 wins, 1 win, 2 wins, aren't seen as numerically ordered but as distinct classes just as red hair" and "blonde hair" are seen as distinct classes with no notion of "greater than" or "less than" between the classes. Modeling an ordinal variable as a nominal variable may reduce the accuracy of the model

The paper linked in the hyperlinks above can give you more guidance on more advanced methods. In addition, you can look at this paper here