- #1

- 1,089

- 10

I have the following problem:

My department head wants to accurately estimate class attendance, as a measure

of the effectiveness of my teaching. For that, I used to take attendance weekly, i.e., once

each week, selecting the day at random, from which I would construct a confidence interval

at the end of the year. The department head wants me to switch from counting weekly to

doing so bi-weekly, since she must enter the data in the computer , submit, etc. I think

this is not a good idea, and I told her so.

I wonder what criterion I could use to argue that this swich is likely to cause a

non-trvial distortion in the estimate of the true mean weekly attendance.

What I have considered, so far:

1) I take the mean of all weekly attendance values, and I construct

a confidence interval. Problem (for me) is that the confidence interval will become

wider as N:= sample-size decreases. So, fewer measurements means smaller accuracy.

2) Trying a difference-of-means test, between the weekly measurements and the

biweekly measurements, and showing that the initial hypothesis is not accepted

at, say, a 95% confidence level.

3)Just a general argument that the Law of Large numbers suggests that more

measurements I get , the more accurate the estimate will be.

Do my arguments work? Should I consider anything else?

Thanks for Your Suggestions.