Comparing museum visitors numbers using non-parametric method

Hi all,

I'm supervising a student who is going to do her research on visitor numbers to a certain exhibition.

The museum will be actively promoting this exhibition in the month to come. The museum wants to know whether this marketing campaign had an effect on visitor numbers.

So, we want to count the number of visitors to the exhibition before the campaign and after the campaign. Of course, many factors can affect attendance; the weather, for instance. Therefore, we plan on using the fraction exhibition/total instead of just the raw numbers. This will cancel out a lot of noise, hopefully. If we find that there's a significant increase in visitor numbers (as a proportion of the total), we will conclude that it's likely due to the marketing campaign (we will look at alternative explanations, but they don't seem very likely - except perhaps word-of-mouth).

Now, the problem is this: due to time constraints, she won't be able to count a lot of days. We're now planning on counting on 5 days for both before and after (same days). Since this is such a small sample size, I was thinking of using Mann-Whitney U. But even then, it seems like it will be tricky to get enough power (I have no clue yet of the variance).

Then again, a day seems like such an arbitrary choice. Why couldn't one choose hours, for instance? They won't be independent, but neither are days of the same week. Presumably the variance will increase with smaller time units, but is there some kind of optimal trade-off? I'm a bit confused here; I don't really know how to approach this problem.

An alternative that we'll probably also use is to ask people whether they've seen any of the communication material (yes/no) and do a chi-square with attendance to the exhibition (yes/no). That one will probably be easier to pull off, but we want to do the first one too.