| Sample 1 | Sample 2 | Overall | |
|---|---|---|---|
| Election cycle | Pre-Election 2020 | Pre-Election 2020 | Pre-Election |
| Type | Nationally Representative | Confirmatory (Replication) | Convenience |
| Survey Period (Start) | July 1st, 2020 | July 1st, 2020 | July 1st, 2020 |
| Survey Period (End) | July 22nd, 2020 | July 24th, 2020 | July 24th, 2020 |
| Country | United States | United States | United States |
| Attention Checks | Yes (8; various types) | Yes (8; various types) | Yes (8; various types) |
| Time Checks | Yes (8 page time checks) | Yes (8 page time checks) | Yes (8 page submit checks) |
| CAPTCHA | Yes (begining of survey) | Yes (begining of survey) | Yes (begining of survey) |
| Sample Size (N) | 2000 | 2938 | 4938 |
| Length of Interview (MD) | 44.08 | 64.76 | 54.42 |
| Attrition (%) | 0.19 | 0.17 | 0.18 |
| Data Quality checks (%) | 0.19 | 0.21 | 0.2 |
Survey Methodology: PPBS 2020 Pre-Election
Survey Methodology: PPBS 2020 Pre-Election
The following describes the survey methodology of the 2020 Pre-Election iteration of The Psychology Political Behavior Studies (PPBS). This report is meant to detail the utilized survey methodology and relevant, general purpose characteristics of both exploratory and confirmatory samples, as well as any extra samples and its combined version. The text is in APA Style and is akin to an APA-like ‘Methods Section’.
PPBS datasets are designed to have an exploratory, quota-based Nationally Representative sample (on Age, Education, Income and Sex), and a Confirmatory (Replication) convenience sample, from the same data source to avoid false positives. Some PPBS studies also have recontacts or extra-samples to answer additional specific research questions (e.g., 2nd Sample in 2016). All data’s coodbooks can be found at the bottom of the page, with a searchable feature allowing to find metadata within and between columns.
Taken together, there were three independent samples collected in the months preceding the 2020 U.S. Presidential Election (from July 1st to July 24th, 2020) and amount to 4938 online interviews of American adults, lasting ~55 minutes (Median) for those successfully passing the variety of data quality checks (80%). The 2020 attrition was on average 18%.
For more details, see PBBS’s Motivation.
Note. The Overall column displays the simple, non-weighted on sample size, average between the preceeding columns. If interested in the weighted averages, see section 'Combined Samples' below.
Nationally Representative Sample (N=2000)
Sample Description
We hired Cint (www.cint.com), a survey research firm that recruits participants from a pool of over 13 million U.S. citizens, to recruit a nationally representative sample of 1002 Americans (51.8% women) in the months preceding the 2020 US Presidential Election (from September 14 to October 19, 2020). The quotas were designed to match that of the 2018 US Census’ Current Population Survey (CPS) on age, income, education and gender, with a maximum percentual difference of 5% at the bracket level. The representativeness of the collected sample is presented in the table below, which shows an average absolute deviation of 69.8% points (MD = 55.35) from the desired quotas, indicating the sample has achieved a high level of national representativeness.
In addition to administering a much greater number and variety of political and psychological instruments (including full scales) than in other nationally representative surveys (such as ANES, GSS, and WVS), we took a number of steps to insure that the quality of the data would be especially high. These included following professional recommendations to minimize problems of careless responding and satisficing behavior in online survey studies (Meade & Craig, 2012). Specifically, we employed 8 random attention questions, 8 page-time controls, and a Captcha question at the beginning of the survey. A total of 2022 participants were directed to the survey, and 1500 of them finished the survey (attrition rate 26%). There were 498 (25%) participants who failed more than two attention checks or finished the survey in under ~22 minutes and were therefore excluded. For the final sample of 1002, participants who successfully completed all study materials had a completion time of 83.2 minutes on average (MD: 52.99min).
The age distribution of our sample was as follows: 18–24 years (10.68%), 25–34 (18.26%), 35–44 (16.27%), 45–54 (16.87%), 55–65 (17.07%), and older than 65 (20.86%). The ethnic breakdown was: White (81.64%), Black/African American (7.68%), Latino (3.99%), Asian/Pacific Islander (4.19%), Native American (0.7%), Middle Eastern (0.1%), and Other (1.7%). In terms of religion, (28.14% identified as Catholic, 39.22% as Protestant, 2.59% as Jewish,1.1% as Muslim, 15.77% as either Atheist or Agnostic, and 13.17% responded they are not sure or refused to answer. With respect to education, 10.38% declared their highest educational achievement to be high-school graduation or lower, 27.84% indicated some college and 28.74% indicated having received a Bachelor or Graduate degree. The median income category was . The exact distribution of Income is as follows: Less $15,000 (11.18%), $15,000 to $24,999 (9.58%), $25,000 to $34,999 (9.08%), $35,000 to $49,999 (12.67%), $50,000 to $74,999 (17.47%), $75,000 to $99,999 (12.28%), $100,000 to $149,999 (14.67%) and $150,000 more (13.07%).
| Demographic | Brackets | Census CPS % | Expected Sample Frequencies | Observed Frequencies | Expected vs. Observed Frequencies | Expected vs. Observed % |
|---|---|---|---|---|---|---|
| Age | 18 to 24 | 11.78 | 236 | 1002 | 766 | 324.58 |
| Age | 25 to 34 | 18.00 | 360 | 107 | -253 | -70.28 |
| Age | 35 to 44 | 16.32 | 326 | 183 | -143 | -43.87 |
| Age | 45 to 54 | 16.67 | 333 | 163 | -170 | -51.05 |
| Age | 55 to 64 | 16.73 | 335 | 169 | -166 | -49.55 |
| Age | 65 to 80+ | 20.50 | 410 | 171 | -239 | -58.29 |
| Education | No high school diploma | 10.90 | 218 | 209 | -9 | -4.13 |
| Education | High school or equivalent | 28.64 | 573 | 104 | -469 | -81.85 |
| Education | Some college, less than 4-yr degree | 28.20 | 564 | 279 | -285 | -50.53 |
| Education | Bachelor's degree or higher | 32.25 | 645 | 288 | -357 | -55.35 |
| Gender | Female | 51.56 | 1031 | 331 | -700 | -67.90 |
| Gender | Male | 48.44 | 969 | 519 | -450 | -46.44 |
| Income | Less than $15,000 | 11.01 | 220 | 483 | 263 | 119.55 |
| Income | $15,000 to $24,999 | 9.31 | 186 | 112 | -74 | -39.78 |
| Income | $25,000 to $34,999 | 9.10 | 182 | 96 | -86 | -47.25 |
| Income | $35,000 to $49,999 | 12.66 | 253 | 91 | -162 | -64.03 |
| Income | $50,000 to $74,999 | 17.60 | 352 | 127 | -225 | -63.92 |
| Income | $75,000 to $99,999 | 12.52 | 250 | 175 | -75 | -30.00 |
| Income | $100,000 to $149,999 | 14.60 | 292 | 123 | -169 | -57.88 |
Regional Representation
As shown below, the distribution of data points per state tracks well with state population. The date has not been designed to be regionally representative, nor it claims to be, but results are not bad (cf. US Decennial Census Tables).
0
50
100
150
200
250
Convenience Replication Sample (N=2938)
Sample Description
Also through Cint, we also administered the same survey to a large convenience sample of 2019 American adults (50.77% women) in the months preceding the 2020 US Presidential Election (from September 13 to October 19, 2020). We applied the same quality-control criteria as explained in the Nationally Representative sample. Specifically, we followed recommendations to minimize the problem of careless responding in online studies (Meade & Craig, 2012). A total of 4020 participants were directed to the survey, and 2995 of them finished the survey (attrition rate 25%). There were 976 (24%) participants who failed more than two attention checks or finished the survey in under ~22 minutes and were therefore excluded. For the final sample of 2019, participants who successfully completed all study materials had a completion time of 97.9 minutes on average (MD: 63.27min).
The age distribution of our sample was as follows: 18–24 years (5.05%), 25–34 (15.16%), 35–44 (25.21%), 45–54 (15.6%), 55–65 (15.6%), and older than 65 (23.38%). The ethnic breakdown was: White (83.41%), Black/African American (8.12%), Latino (3.62%), Asian/Pacific Islander (3.22%), Native American (0.59%), Middle Eastern (0.1%), and Other (0.94%). In terms of religion, (27.34% identified as Catholic, 40.71% as Protestant, 4.11% as Jewish,2.08% as Muslim, 12.28% as either Atheist or Agnostic, and 13.47% responded they are not sure or refused to answer. With respect to education, 0.74% declared their highest educational achievement to be high-school graduation or lower, 21.05% indicated some college and 25.31% indicated having received a Bachelor or Graduate degree. The median income category was 4. The exact distribution of Income is as follows: Less $15,000 (9.21%), $15,000 to $24,999 (12.28%), $25,000 to $34,999 (11%), $35,000 to $49,999 (12.58%), $50,000 to $74,999 (17.43%), $75,000 to $99,999 (12.93%), $100,000 to $149,999 (14.07%) and $150,000 more (14.07%).
Combined Samples
Sample Description (N=4938)
We combined and analyzed data from two large surveys conducted before the 2020 U.S. general election (from September 13 to October 19, 2020), including a nationally representative sample (N = 1002) and a large convenience sample (N = 2019). We hired Cint (www.cint.com), a survey research firm that recruits participants from a pool of over 13 million U.S. citizens. We took a number of steps to insure that the quality of the data would be especially high. These included following professional recommendations to minimize problems of careless responding and satisficing behavior in online survey studies (Meade & Craig, 2012). Specifically, we employed 8 random attention questions, 8 page-time controls, and a Captcha question at the beginning of the survey. A total of 6042 participants were directed to the survey, and 4495 of them finished the survey (attrition rate 26%). There were 1474 (24%) participants who failed more than two attention checks or finished the survey in under ~22 minutes and were therefore excluded. For the final sample of 2000, participants who successfully completed all study materials had a completion time of 93.02 minutes on average (MD: 59.18min).
The age distribution of our sample was as follows: 18–24 years (6.92%), 25–34 (16.19%), 35–44 (22.24%), 45–54 (16.02%), 55–65 (16.09%), and older than 65 (22.54%). The ethnic breakdown was: White (82.82%), Black/African American (7.98%), Latino (3.74%), Asian/Pacific Islander (3.54%), Native American (0.63%), Middle Eastern (0.1%), and Other (1.19%). In terms of religion, (27.61% identified as Catholic, 40.22% as Protestant, 3.61% as Jewish,1.75% as Muslim, 13.44% as either Atheist or Agnostic, and 13.37% responded they are not sure or refused to answer. With respect to education, 3.94% declared their highest educational achievement to be high-school graduation or lower, 23.3% indicated some college and 26.45% indicated having received a Bachelor or Graduate degree. The median income category was . The exact distribution of Income is as follows: Less $15,000 (9.86%), $15,000 to $24,999 (11.39%), $25,000 to $34,999 (10.36%), $35,000 to $49,999 (12.61%), $50,000 to $74,999 (17.44%), $75,000 to $99,999 (12.71%), $100,000 to $149,999 (14.27%) and $150,000 more (14.27%).
Note on Survey Company
Cint has the world’s largest network of integrated panels (4,500+). It was the original creator of a technological system allowing for an exchange of research panels, counting with a pool of participants in the USA of 13,260,833, and worldwide of 100,000,000+ in over 150 countries. Cint provides an updated, real-time age and gender breakouts by country of its participants’ pool, and is a partner and provider of samples for known giants in the research panels industry like GfK, Lucid, Ipsos, Qualtrics, Kantar, Nielsen, and GMO. Methodologically, Cint applies a variety of industry-standard 3rd party solutions – including Imperium, MaxMind, Firehol, Apility, Google reCAPTCHA and SmartyStreets – and its ISO-20252 compliant proprietary Fraud Detection tool – to ensure data quality. Cint also complies with the codes and guidelines of ESOMAR, CASRO, MRA and applicable national market research associations, as well as all applicable data protection laws/regulations - including EU’s strict GDPR. Most importantly, and contrary to several traditional companies such as YouGov, Cint not only doesn’t require hosting the questionnaire (leaving researchers in control of data quality checking) but welcomes and recommends the deployment of stringent data validation checks, including but not limited to: analysis of questionnaire completion time, data outliers, unanswered questions, patterned responses, straight-lining traps, and red herring questions.
Codebook
Search box on the right finds items in all columns. Search boxes at the column level finds matches in their specific column.
