Government Funding and Private Donations: Crowding-in Versus Crowding-out in the Context of a Big Data Field Experiment

How to cite

Jilke, Sebastian; Xu, Chengxin; Shinohara, Shugo; Lu, Jiahuan. 2017. "Government Funding and Private Donations: Crowding-in Versus Crowding-out in the Context of a Big Data Field Experiment." AEA RCT Registry. AEARCTR-0002345. https://doi.org/10.1257/rct.2345-3.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Sebastian Jilke* , Chengxin Xu , Shugo Shinohara , Jiahuan Lu

* Primary investigator

Registry ID
AEARCTR-0002345
DOI
https://doi.org/10.1257/rct.2345-3.0
First registered
Last updated
Status
Completed
Trial dates
2017-08-21 – 2017-12-31
Countries
United States of America (NYC)
Registry record
http://www.socialscienceregistry.org/trials/2345

Abstract

The study explores whether government funding of nonprofit organizations crowds in/out private donations using a large-scale, big data field experiment. We will buy advertisements on Facebook to solicit donations among Facebook users to real food banks in NYC that are either government funded, donation funded, or without funder information. Clusters of Facebook users will be randomly exposed one of the experimental conditions: 1) government-funded, 2) donation-funded, and 3) no funding information. As outcome measure, we will monitor people’s intention to donate to the food banks by their actual click-through-rates.

Intervention

The intervention includes advertisements on Facebook for food banks in NYC with randomly varied funding information (i.e., government-funded, donation-funded, or no such information). Yet, the advertisements will not solicit donations to specific food banks but instead enourage users to click on the "Rutgers Observatory for [government-funded/ donation-funded/ no-info] Food Banks", where they will be directed to a website off Facebook property.

Experimental design

We test crowding-in and crowding-out claims using a large-scale, big data field experiment. We will buy advertisements on Facebook to advertise donations to real food banks from New York City that are either government funded, or donation funded. Respondents from the NYC area will be targeted to keep the experiment regionally focused and at the same time ensure experimental realism. We will use Facebook’s online advertising facilities and randomly allocate groups of Facebook users into three experimental conditions, thereby manipulating the following funding information: 1) government funded, 2) donation funded, 3) no such information. Randomization will be done on a cluster leve of usersl, stratified by age, gender and zip code. A total of 600 clusters will be composed. Subjects within clusters will be repeatedly exposed to donation advertisements for 24 hours.

No individual-level date will be collected. For those who actually clicked the advertisements, they will be directed to a webpage on the School of Public Affairs and Administration (SPAA) at Rutgers University website, called “The SPAA Food Bank Observatory.” On this webpage, visitors will be informed of the purpose of the study and provided with a list (depdening on the experimental condition) of either government funded or primarily donation funded (or all) food banks in NYC, and encouraged to donate money to them (i.e., depending on the exprimental condition, three versions of this webpage exist). The food banks listed on the websites are all real food banks that operate in NYC, as collected through the National Center for Charitable Statistics (NCCS).

Design details

Randomization method
Randomization will be done by a computer (Stata 14.0); 600 clusters will be blocked into 200 groups of 3 clusters each. Within each group, 1 cluster will get randomly assigned to the government-funding condition, 1 cluster to the donation-funding condition, and 1 cluster to the no information condition.
Randomization unit
The unit of randomization is clusters of facebook users, stratified by age, gender and zip code. Randomly, 100 NYC zip codes were chosen. In addition, two genders (male and female), 3 age groups (18-27; 28-38; 39-60) were used for stratification --> 100 zipcodes * 2 genders * 6 age groups = 600 clusters.
Clusters
600 clusters of Facebook users.
Observations
Will be determined by Facebook auctions once ads are fielded. Potenital reach per cluster ranges between 1,000 to 110,000 users, and we will pay $10.00 per each cluster which will potenitally result in a reach of about 300-700 per cluster (i.e., a reach between 180,000 - 420,000 in total). To deal with potenitally unequal cluster sizes, cluster randomization will be blocked by size (3 clusters per block; 1 donation-funded, 1 government-funded, 1 control).
Arms
200 cluster per treatment arm (i.e., 200 clusters control, 200 clusters donation-funded, 200 clusters government-funded)
Primary outcomes
Our primary outcome of interest is subjects' intention-to-donate, which we operationalize as the unique-outbound-link-click-rate (i.e., reach divided through unique outbound clicks) per cluster.

Papers from this trial

  1. In this article we introduce and showcase how social media can be used to implement experiments in public administration research. To do so, we pre-registered a placebo-controlled field experiment and … https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3164576

Cited by

No citation data available for this registration yet.

Keywords