PDA

View Full Version : Proper statistical test needed

hominid
06-01-2010, 02:16 PM
A relative recently told me that her horoscope said that she might be in a car accident in the month of May, and indeed she was in a car accident. She asked me what I thought about this. I want to reply that the astrologer's claims are testable. Were people under the sign of Aries in more car accidents last month than in previous months of May?

Car accident reports are public records, and they include people's birthdays. My relative is an Aries, born between March 20 and April 20. She could collect car accident reports and filter them to include only people born between those days. Then she could look at car accident rates (as a proportion of all Aries) in the month of May for the past several years and compare them to the car accident rate last month (when the astrologer made the claim).

So my question is: what would be the appropriate statistical test for a difference in the proportions?

She'd be comparing two proportions, one with a standard deviation (for multiple past months of May) to one without (just last month). Both proportions would also have confidence intervals, since she couldn't collect every car accident report, thus she would be estimating from samples of the population.

I'm thinking a simple t-test would suffice, but I wanted some advice.

Of course, this is all hypothetical, since she's not really going to cull through car accident reports to get an answer. I just wanted to make sure my answer was correct before I gave it. It's also a good refresher on the statistics that I took long ago. That class didn't focus on the equations. It focused entirely on word problems and choosing the right statistical test for a situation, which is the hardest part. :-)