Experiments

STAT 20: Introduction to Probability and Statistics

Concept Questions

We run an experiment to see if informative phone calls encourage people to vote. We obtain phone numbers for 1000 registered voters and randomly assign half of them to receive calls from volunteers. The volunteers tell them the location of their polling place and the date of the election. Our outcome is whether or not the subjects vote in the next election.

Which of the following are possible covariates for our study?

1. The party affiliation of the voter (Republican or Democrat).
2. Whether the next election is a presidential election or not.
3. Whether the voter voted in the last election.
4. Whether the voter votes primarily for Republicans or Democrats in the next election.
5. Whether the treated voters hung up on the volunteer.

The point of this question is to get the students thinking about whether variables are measured before or after treatment assignment.A and C are correct answers, D and E are not (although E is measured before the primary outcome, it is still after treatment). B is technically a covariate but probably not a useful one; unless our experiments spans multiple election cycles, it will have the same value for all subjects.

01:00

Which of the following claims about covariate balance are true?

1. Randomized treatment assignment tends to produce covariate balance.
2. A standardized mean difference that is not equal to zero likely means that randomization was not conducted correctly.
3. The Love plot shows the difference in means or proportions for each covariate across treatment groups.
4. In a hypothesis test for balance, the null hypothesis is that the treatment assignment and the covariate are independent.

01:00