Week One - Homework

Practice Exercises 1

1. The night shift was not necessarily safer. It might just be that there are fewer workers on the night shift. You would need to know the number of workers and the hours they are working to figure out the rate at which accidents happen. This is a problem of a missing denominator. 2. (a) There are many more Camrys on the road than Escalades (b) Because the theft rate is higher, the Escalade is more likely to get stolen.

3. They should look at the complaint rate per hundred prescriptions -- not the total number of complaints. The reason most of the complaints are for CVS pharmacy could easily be because most of the prescriptions are written from CVS pharmacies. It could turn out that CVS pharmacies are doing better than other pharmacies.

Practice Exercises 2

4. (a) observational (b) Prisoners who volunteer to take these courses are more likely to be interested in being rehabilitated than the ones who don’t volunteer. The confounding factor created here is the level of prisoner motivation to be rehabilitated: it influences whether or not they take the courses and also whether or not they commit crimes to return to prison. The education program may work well, but you can’t tell from this study because of this confounding factor. (c) A better way would be to choose a random sample of inmates and divide them into "treatment" and "control" groups. You then could offer the treatment group the option of attending counseling and compare the recidivism rate for this entire group -- not just the ones you actually go to counseling -- with the control group. This means you will compare everyone who was offered counseling to everyone who was not offered counseling. (Within the treatment group there will be some people who don't go to counseling -- but they should be all considered people who were given the "treatment.")

5. Cars with alarms are more likely to be nicer than cars without alarms. This explains why they happen to get stolen more often. The value of the car is a confounding factor here: it affects whether or the car has an alarm, and also how desirable the car is to thieves.

8. No, it wouldn't be an example of a confounding factor because both groups you are comparing would have the same type of prior health problems. If you randomly divide the groups in a randomized experimental study there should not be any confounding factors.