Door-to-door canvassing can be effective in the two weeks prior to Election Day, whereas earlier canvassing efforts are not found to be as effective. Multiple visits may not increase effectiveness of campaigns. Candidate-centered messages are more effective than partisan-centered messages.

Nickerson, David W. 2006. "Forget Me Not? The Importance of Timing in Voter Mobilization." Paper presented at the annual meeting of the American Political Science Association, Marriott, Loews Philadelphia, and the Pennsylvania Convention Center, Philadelphia, PA.

David Nickerson presents the results of a randomized field experiment designed to determine at what point in time face-to-face contact becomes effective in mobilizing voter turnout.  The experiment was conducted by the Young Democrats of America during the 2005 Virginia gubernatorial election.  Areas of Northern Virginia were randomly assigned to differing numbers of visits from canvasser.  The timing of visits to subjects was randomly assigned as well. The experiment uncovered strong evidence that door-to-door canvassing can be effective within the  two weeks prior to Election Day, but earlier canvassing efforts appear decidedly less effective.  The results also indicate that multiple visits may not be necessary and suggest that candidate-centered material may be more effective than party-centered literature as a means to mobilize voters.

Electoral Context:  The experiment was conducted throughout the 2005 Virginia gubernatorial election. The race was tightly contested and candidates ran highly professional, well-funded campaigns on par with candidates running for national office.

Subject Population: The Young Democrats of America targeted areas in Northern Virginia with high densities of Democratic and Independent registered voters under the age of 36.

Randomization Procedure:  The randomization occurred at two levels.  First, areas with high densities of Democratic and Independent voters were divided into 67 “turfs” that could be walked by paid canvassers.   Turfs were randomly assigned some number of attempted visits by canvassers and a schedule for the visits.  The schedules were set by dividing the canvassing calendar into seven two-week intervals ending with Election Day.  Second, subjects were randomized at the household level. Roughly 13% of the households (N =23,359) were randomly assigned to be an internal control group in order to construct treatment variance within turfs and maximize statistical power. Of the remaining 87% of households, half were randomly selected to receive a flyer comparing the Republican and Democratic parties and half of the households received a flyer comparing Tim Kaine to Jerry Kilgore. The goal was to determine whether young voters responded better to party labels or individual personalities.

Treatment:  The treatments were two-fold.  For the treatment that was designed to explore the most effective timing and frequency of visits, a schedule of visits was pre-set and divided the calendar into seven time periods consisting of approximately two-week intervals.  Subjects could not be contacted two consecutive periods until the final two time periods leading up to Election Day. 

The portion of the experiment that was randomized at the household level examined the difference between the effectiveness of information describing the differences between the Democratic and Republican parties versus information comparing the Democratic and Republican candidates.  Flyers were left at the door in the event that there was no answer. When someone answered the door, canvassers used the randomly selected flyer for talking points and left the flyer with the subject. 

Canvassers in Northern Virginia faced two obstacles.  First, canvassers found very few of the subjects to be home during canvassing hours.  Canvassers adjusted their schedules somewhat but found that young voters with children objected to canvassing efforts that ran too late into the evening.  Second, the majority of young voters in Northern Virginia live in apartment buildings with restricted access.  Overall, canvassers contacted approximately 13% of the targets.

Findings:  Using OLS regression, Nickerson estimates that voters assigned to be contacted in the last few days of the campaign were more likely to vote than those assigned to the control group during that time period.  The estimated effect of 3.5 percentage points (with a standard error of 2.0 percentage points) implies that the 8,700 attempts to contact subjects face-to-face in the final days of the campaign generated an additional 305 votes.  Taking the contact rate into account, Nickerson estimates that voters contacted in the final days were 27 percentage points more likely to vote than individuals in the control group (standard error: 15.4 percentage points); however the high standard error renders the estimate significant but uncertain.  Canvassing in the last two weeks of October appeared to increase the likelihood of turnout by 5.3 percentage points though the estimate does not reach statistical significance (standard error: 3.5 percentage points).  After controlling for past turnout, Nickerson finds that late October canvassing increased the likelihood of voting by 4.7 percentage points (standard error: 2.5 percentage points); 5,046 attempts to contact subjects via canvassing in late October resulted in an additional 237 votes.  Earlier canvassing attempts appear to be ineffective in increasing the likelihood of turnout with the exception of canvassing efforts in early August.  Subjects contacted in early August were 6.2 percentage points more likely to vote (standard error: 3.0) than control group subjects.  Pooling the results of the time periods prior to mid-October, Nickerson estimates a 0.7 percentage point increase in the likelihood of voting (standard error: 1.2 percentage points).  Taking the contact rate into account, the treatment on the treated effect is estimated to be 5.3 percentage points (standard error: 9.2 percentage points.) 

Analysis estimating the differential effects of multiple contacts reveals no systematic pattern, suggesting that contacting subjects multiple times does not enhance mobilization or persuasion effects. From his comparison of the different combinations of frequency and timing of visits, Nickerson concludes that face-to-face canvassing ceases to increase turnout prior to the final 18 days leading up to Election Day.  He further argues that earlier canvassing may educate or persuade voters, but it does not appear to boost turnout. The results from canvassing in early August, Nickerson argues, could most likely be attributed to sampling variability.

Analysis of partisan-centered versus candidate-centered literature reveals no robust difference in the differential effects of the literature when considering the entire sample (which is possible since flyers were left at doors regardless of contact.)  Among subjects who were actually contacted by canvassers, however, individuals who received the information contrasting the candidates were 4.8 percentage points more likely to vote (standard error: 1.4 percentage points) than subjects contacted with information comparing the parties.  The strength of the estimate diminishes slightly when additional controls are added.