# Qualitative Stats question (central limit theorem)

1. Dec 5, 2006

### habman_6

3) Which of the following are consequences of the Central Limit Theorem?
I) A SRS of resale house prices for 100 randomly selected transactions from all sale transactions in 2001 (in Toronto) will be obtained. Since the sample is large, we should expect the histogram for the sample to be nearly normal.
II) We will draw a SRS (simple random sample) of 100 students from all University of Toronto students, and measure each person’s cholesterol level. The average cholesterol level for the sample should be approximately normally distributed.
III) We want to estimate the proportion of Ontario voters who intend to vote for the Liberal party in the next election, and decide to draw a SRS of 400 voters. The percentage of the people in the sample who will say that they intend to vote Liberal is approximately normally distributed.
IV) We will draw a SRS of 100 adults from the Canadian military, and count the number who have the AIDS virus. The number of individuals in the sample who will be found to have the AIDS virus should be approximately normally distributed.
V) We are interested in the average income for all Canadian families for 2001. The mean income for all Canadian families should be approximately normal, due to the large number of families in the population.

The answer is II and III. I understand why I is wrong, and I understand why V is wrong. However, to me, IV seems exactly the same as III. Apparently its because the proportion is too low, but that does not make sense to me in terms of CLT. What should it matter what the proportion is, as long as the sample means have that same proportion?

2. Dec 6, 2006

### AlephZero

If the sample size or the proportion is too small the approximation is too poor to be useful. As an example, consider the Binomial distribution(N,p). The CLT applies to this case. The rule of thumb is that the normal approximation to the Binomial is poor unless Np > 10.

http://www.stat.yale.edu/Courses/1997-98/101/binom.htm

To convince yourself, look at the pdf of some binomial distributions that don't satisfy Np > 10, e.g. N = 100, p = 0.01.

3. Dec 6, 2006

### habman_6

Thanks Aleph, I understand now.