|Title||Estimating the effect of LinkNYC on voter registration and participation|
|C1 Background and Explanation of Rationale||
Beginning in 2015, New York City began installing LinkNYC kiosks to replace the city’s network of payphones. Each kiosk features 55’’ high definition displays that show content including advertising and transit information, as well as tablets for accessing city services and maps, free USB charging stations, and phones for making domestic calls.
On September 21st, 2018, LinkNYC launched a voter registration drive to encourage voter registration on kiosk tablets, which connect users to an electronic voter registration page. While these devices are designed to bring internet access and city services to users’ fingertips, their ability to promote electoral participation is an open question; their usage for this purpose – and whether messages encouraging participation actually work – has not been studied. An important question is whether increasing access to, and the visibility of, voter registration resources improves electoral participation.
The present study is a randomized implementation of registration encouragement using LinkNYC large screens to display either (1) a localized neighborhood-identity based participation appeal; (2) an urgency appeal; or (3) no appeal (a control group). Randomization occurs at the Census tract level (n = 828). Outcomes measured include individual counts of voter registration site usage, with date-times, duration of use, and location of specific kiosks where each use occurred, as well as voter registration and turnout data.
From 9/21/18 through 10/12/18 (registration deadline), all kiosks in a given census tract will display their assigned message, or lack of a message, at regular intervals, alongside other usual (non-experimental) content. At the end of that period, the implementing organization (CityBridge LLC, which operates the kiosks) will provide timestamped counts of registrations on each kiosk during the study period. Although we will not have information about individual registrants, we use the NYC voter file to obtain characteristics of people who registered during the study period under each of the three conditions based on the census tract in which they live and the kiosks nearest their home address.
|C2 What are the hypotheses to be tested?||
Outcomes are measured from two main sources: (1) counts of individuals who use LinkNYC kiosks to register at the kiosk level, provided by CityBridge, which can be aggregated up to the tract level, and (2) voter information from the NYC voter file. The former includes date and time information, while the latter includes dates, allowing us to determine which registrations occurred during the message campaign and to study heterogeneous effects by proximity to the registration deadline. Moreover, we will obtain an updated NYC voter file after Election Day to determine whether those who registered after exposure to the interventions actually turned out to vote. Though we will be unable to identity which individuals registered on a LinkNYC kiosk, since treatment is randomized at the tract level we can treat all registrants in a tract as having been treated to some degree (i.e., it is plausible that individuals saw the ads and subsequently registered via another medium). The following hypotheses will be tested with data from both sources unless otherwise specified.
H1: Encouragement effect: Registration will be higher in tracts where the messages are shown compared to those in which they were not shown.
H2: Neighborhood-specific vs. countdown message: We will test the efficacy of two different messages. The first appeals to neighborhood identity (e.g., “Greenpoint added XXX new voters in 2017. Tap ‘register now’ on the tablet to join in” ) and the second shows a dynamic countdown to the registration deadline (“06 days 04 hours 22 seconds until the voter registration deadline for the November general elections. Tap ‘register now’ on the tablet to join in”).
H3: Heterogeneous effects: Subject to sample size constraints, we plan to explore the possibility of heterogeneous treatment effects by attributes of individuals’ context, as well as by individual characteristics. Are the effect of the messages moderated by local racial or socioeconomic context, or by individual characteristics? Since the literature is largely silent on these effects with respect to voter registration, this exercise is largely exploratory. See pre-analysis plan for details.
H4: Subsequent voting: How likely to vote are people who registered during the messaging campaign in treated tracts vs. those who registered in control tracts? Are people who were exposed to the neighborhood identity message vs. the countdown message more or less likely to vote?
|C3 How will these hypotheses be tested? *||
The main outcome variable is the tract level registration rate, calculated as the number of new registrants divided by the number of people eligible to register. The numerator comes from either the number of people who register at LinkNYC kiosks, or from the number of people who register via any method, during the study period. Subject to data availability we also calculate a registration rate out of the typical number of sessions on each kiosk per day.
H1: Encouragement effect: We conduct t-tests of the difference in means of the tract-level registration rate in treatment tracts (pooled) vs. control tracts. We also run OLS regressions controlling for pre-treatment registration rates, which we were unable to block on, as well as poverty status, educational attainment, age distribution, residential mobility, homeownership, and employment, in order to increase precision.
In order to account for upward trending registration rates as the election approaches, we will calculate the difference-in-differences from the NYC voter file: (registration rate in treatment tracts at t0 - registration rate in control tracts at t0) - (registration rate in treatment tracts at t1 - registration rate in control tracts at t1), where t0 includes the 22 days prior to 9/21 and t1 includes 9/21 – 10/12 (same number of days in t0 and t1).
H2: Neighborhood-specific vs. countdown message: We will regress tract level registration rates on indicators for which type of message was received as well as the controls described above.
H3: Heterogeneous effects: See pre-analysis plan for details.
H4: Subsequent voting: We regress an indicator for whether the individual voted on tract-level treatment indicators and controls for individual characteristics (race, age, gender) and contextual variables, clustering standard errors at the tract level.
|C4 Country||United States|
|C5 Scale (# of Units)||The experiment will involve 838 Census tracts|
|C6 Was a power analysis conducted prior to data collection?||No|
|C7 Has this research received Insitutional Review Board (IRB) or ethics committee approval?||not provided by authors|
|C8 IRB Number||Not human subjects research|
|C9 Date of IRB Approval||not provided by authors|
|C10 Will the intervention be implemented by the researcher or a third party?||CityBridge, llc (Intersection)|
|C11 Did any of the research team receive remuneration from the implementing agency for taking part in this research?||No|
|C12 If relevant, is there an advance agreement with the implementation group that all results can be published?||Yes|
|C13 JEL Classification(s)||not provided by authors|