多项式：如何获得观察数

Question

I am using multinom from nnet to create a logistic model from a massive clinical database. The syntax I'm using is:

library(nnet)
fit=multinom(group ~ sex + age + var3 + var4,
            data=d, na.action = na.omit)

Now, each row (a patient) has a different number of NAs, as not all clinical data were recorded for all patients, and it is not clear to me if the model only uses those rows for which all of the variables do not contain any NAs. More in general, it would be useful to obtain the Ns of observations that the model is based on, which I suspect is smaller that the number of rows of my dataframe (N of patients). I have looked everywhere but I don't seem to be able to find how to do this.

Answer 1

I believe length(residuals(fit)) should work.

If you want to explore the data set that was used to fit the model, you can do:

mf <- model.frame(group ~ sex + age + var3 + var4, data=d,
                  na.action=na.omit)

(this is how multinom processes your data) and then count the number of rows, tabulate the number of observations in different categories for different variables (eg table(mf$var3) ), etc.. lapply(mf,table) should tabulate the number of observations by category for every variable.

You might find the describe function from the Hmisc package useful:

Hmisc::describe(group ~ sex + age + var3 + var4, data=d)