| Treatment group | Control group | |
|---|---|---|
| Turnout rate among those contacted by canvassers | 54.43 (395) | |
| Turnout rate among those not contacted by canvassers | 36.48 (1,050) | 37.54 (5,645) |
| Overall turnout rate | 41.38 (1,445) | 37.54 (5,645) |
Non-compliance
Study: Getting out the vote
During the weeks before the 1998 midterm elections, households of over 30,000 voters in New Haven, Connecticut were randomly assigned to a treatment group or a control group in a study run by Gerber and Green. Treated households were visited by canvassers who knocked on their doors to talk to them about the importance of voting. At the end of the election, the public registries were checked to see which of the original group of voters had cast a vote in the election.
What kind of experiment is this? What are the units?
CR[1]. Experimental units are households. Measurement units are voters.
What is \(ITT_Y\)?
\(41.38 - 37.54 = 3.84\)
What is \(ITT_D\)?
\(395 / 1445 = 27.3\)
| Treatment group | Control group | |
|---|---|---|
| Turnout rate among those contacted by canvassers | 54.43 (395) | |
| Turnout rate among those not contacted by canvassers | 36.48 (1,050) | 37.54 (5,645) |
| Overall turnout rate | 41.38 (1,445) | 37.54 (5,645) |
\[ CACE = \frac{ITT_Y}{ITT_D} \]
\[ CACE = 3.84 / .273 = 14.07 \]
- The naive estimate of, \(ITT_Y\), yields an underestimate.
- The corrected estimate (\(CACE\)) is only the ATE of compliers.
\(CACE\) vs \(ITT_Y\)
Which parameter are you more interested in?
You are a political scientist and want to understand the interactions that cause people to change their voting behavior as a general rule.
You are a campaign consultant and want to know how much extra turnout you can get from canvassing a city, relative to spending the effort on more web ads.