Search

Recent studies consistently showed that making cash visible with a windowed envelope during mail contact increases response rates in surveys. The visible cash aims to pique interest and encourage sampled households to open the envelope. This article extends prior research by examining the effect of additional interventions implemented during mail recruitment to a survey panel on recruitment rates and costs. Specifically, we implemented randomized experiments to examine size (small, large) and location (none, front, back) of the window displaying cash, combined with what part of the cash is shown through the window envelope (numeric amount, face/image), and various prepaid incentive amounts (two $1, one $2, one $5). We used the recruitment effort for NORC's AmeriSpeak Panel as the data source for this study. The probability-based AmeriSpeak Panel uses an address-based sample and multiple modes of respondent contact, including mail, phone, and in-person outreach during recruitment. Our results were consistent with prior research and showed significant improvement in recruitment rates when cash was displayed through a window during mail contact. We also found that placing the window on the front of the envelope, showing $5 through the envelope compared to $2 and $1, and showing the tender amount compared to the image on the cash through the window were more likely to improve the recruitment rates. Our cost analyses illustrated that the cost difference in printing window versus no window envelope is small. There is no difference in printing cost between front window and back window as they both require custom manufacturing. There is also no cost difference in printing envelopes with small windows versus large windows. Lastly, we found no evidence of mail theft based on our review of the United States Postal Service's "track and trace" reports, seed mailings sent to staff, and undeliverable mailing rates. [ABSTRACT FROM AUTHOR]

Big data has been fruitfully leveraged as a supplement for survey data—and sometimes as its replacement—and in the best of worlds, as a "force multiplier" to improve survey analytics and insight. We detail a use case, the big data classifier (BDC), as a replacement to the more traditional methods of targeting households in survey sampling for given specific household and personal attributes. Much like geographic targeting and the use of commercial vendor flags, we detail the ability of BDCs to predict the likelihood that any given household is, for example, one that contains a child or someone who is Hispanic. We specifically build 15 BDCs with the combined data from a large nationally representative probability-based panel and a range of big data from public and private sources, and then assess the effectiveness of these BDCs to successfully predict their range of predicted attributes across three large survey datasets. For each BDC and each data application, we compare the relative effectiveness of the BDCs against historical sample targeting techniques of geographic clustering and vendor flags. Overall, BDCs offer a modest improvement in their ability to target subpopulations. We find classes of predictions that are consistently more effective, and others where the BDCs are on par with vendor flagging, though always superior to geographic clustering. We present some of the relative strengths and weaknesses of BDCs as a new method to identify and subsequently sample low incidence and other populations. [ABSTRACT FROM AUTHOR]

Hypernetwork sampling aims to generate representative samples of populations for which a sample frame does not exist or is too costly to construct. This multi-level sampling method relies on nominations from one sample source (Stage 1 sample) to construct another sample (Stage 2 sample). However, nonresponse from the Stage 1 sample has the potential to produce bias in Stage 2 of the hypernetwork sample if Stage 1 respondents differ from nonrespondents. This paper examines nonresponse in a hypernetwork sample of religious congregations in the U.S. generated from a probabilitybased household panel that includes background information for all panelists including Stage 1 nonrespondents. This study also illustrates the benefits of constructing a hypernetwork sample by using a sample of already recruited panelists for whom information has already been collected. We find Stage 1 nonrespondents tend to be from rural areas and not from the Midwest, compared to Stage 1 respondents. Results also suggest that the impact of subsequent survey reminders on key Stage 1 estimates decreased after the third reminder during Stage 1 fielding. Additionally, we find that Stage 1 nonresponse impacts the Stage 2 estimates for congregational characteristics. Specifically, the congregations nominated by Stage 1 late respondents tend to have the following characteristics: located in the South, predominantly African American, more likely to be conservative/evangelical Protestant or black Protestant, younger, urban or suburban, helped people register to vote, less likely to have a school, and have fewer child participants. Post-survey weighting adjustment of the Stage 1 sample decreased the risk for nonresponse bias in the Stage 1 hypernetwork sample and in the Stage 2 sample of congregations. [ABSTRACT FROM AUTHOR]

This is a correction to: David Dutwin and others, Leveraging Predictive Modelling from Multiple Sources of Big Data to Improve Sample Efficiency and Reduce Survey Nonresponse Error, Journal of Survey Statistics and Methodology, 2023, smad016, https://doi.org/10.1093/jssam/smad016In the originally published version of this manuscript, the author Joshua Lerner was inadvertently omitted from the author byline.This error has been corrected.By David Dutwin; Patrick Coyle; Joshua Lerner; Ipek Bilgen and Ned EnglishReported by Author; Author; Author; Author; Author [Extracted from the article]

With proliferation of web surveys, the relative affordability of recruitment, and increasing nonresponse in other survey modes, nonprobability methods are increasingly being considered by researchers and government offices alike. However, research needs to more fully understand how the demographic characteristics of respondents may depend heavily on the source of sample, mode of recruitment, and context of the survey experience. As a first step in exploring the potential implications of recruitment source on response quality, we use data from a web survey fielded in 2013 to compare data quality indicators in survey data from the two recruitment platforms (Google and Facebook advertisements). In so doing, taking into account demographic differences that may arise from various steps in the recruitment process, we explore the effect of demographics, device and technology usage, incentives, and recruitment platform on data quality and response strategy. Our results show differences between platforms in comparability to national benchmarks, breakoffs, completion time, nonsubstantive answers, and numeric response strategies. Importantly, some variation in substantive responses was explained by demographic differences related to mobile device usage, which varied by recruitment platform. With the use of nonprobability samples on the rise, future work should build from these results to more directly assess the role of recruitment source in data quality. [ABSTRACT FROM AUTHOR]