This is an older question, but I'll attempt an answer.
This situation is not a good candidate for McNemar's test.
For McNemar's test, the categories have to be mutually exclusive. That is, in this example, if one school choses Question 1, they could not also choose Question 2, or 3, and so on.
To set up a McNemar's test, you should be able to construct a contingency table where the labels for the columns are the same as the labels for rows. Note also that it's necessary to be able to identify the "before" response for each subject (person, experimental unit) and identify the "after" response for the same subject.
In the following table, note that the column categories and row categories are the same. Also note that the table sums to 34 counts, which is the number of subjects. But note that in a sense there are 68 observations, since each of the 34 subjects had to have a response recorded twice.
After
Before Yes No
Yes 12 7
No 5 10
Since the response in the example in the question is a dichotomous yes/no, the example could probably be analyzed with logistic regression. The right hand side of the model would probably include Question and Time, and probably School as a random effect to take into account the repeated measures nature of the data.
A post-hoc analysis (for example with E.M. means) could be used to determine the differences among questions.
A partial set of sample data follows, with a potential model, written in R notation. Another model may be more appropriate, depending on the nature of the problem and data.
Question School Time Response
1 a Before Yes
1 a After Yes
1 b Before No
1 b After Yes
.
.
2
2
.
.
glmer(Response ~ Question + Time + (1 | School), data = Data, family = binomial)
