Methodology
This survey, funded by New America, was conducted by NORC at the University of Chicago. Staff from NORC and New America collaborated on all aspects of the study. Data were collected using both probability and non-probability sample sources. Interviews for the survey were conducted between April 6 and April 20, 2026, with adults age 18 and over representing the 50 states and the District of Columbia.
The probability sample source is AmeriSpeak®, NORC’s probability-based panel designed to be representative of the U.S. household population. During the initial recruitment phase of the panel, randomly selected U.S. households were sampled with a known, nonzero probability of selection from the NORC National Sample Frame and then contacted by U.S. mail, email, telephone, and field interviewers (face to face). The National Sample Frame provides coverage of approximately 97 percent of the U.S. household population. The U.S. Postal Service Delivery Sequence File (USPS DSF) and National Consumer Address File have been used in a subset of years as a supplemental sample during AmeriSpeak panel recruitment for low-coverage states. Those excluded from the sample include people with P.O. box–only addresses, some addresses not listed in the USPS DSF, and some newly constructed dwellings.
Panel members were randomly drawn from AmeriSpeak, and 2,673 completed the survey—2,627 via the web and 46 via telephone. Panel members were invited by email or by phone by a NORC telephone interviewer. Interviews were conducted in both English and Spanish, depending on respondent preference. Respondents were offered a small monetary incentive ($3) for completing the survey. The weighted household panel recruitment rate is 24.9 percent and the survey completion rate is 40.6 percent, for a weighted cumulative response rate of 10.1 percent.
Dynata provided 672 nonprobability interviews with men age 18–29, men of color, and fathers of children age 0–8. The nonprobability sample was derived based on quotas related to age, race and ethnicity, gender, and education. Interviews were conducted in English and Spanish and via the web only. For panel recruitment, Dynata uses invitations of all types—including email invitations, phone alerts, banners, and messaging on panel community sites—to include people with a diversity of motivations to take part in research. Because nonprobability panels do not start with a frame where there is a known probability of selection, standard measures of sampling error and response rates cannot be calculated.
Quality assurance checks were conducted to ensure data quality. In total, 208 interviews were removed for nonresponse to at least 50 percent of the questions asked of them or for completing the survey in less than one-third the median interview time for the full sample. These interviews were excluded from the data file prior to weighting.
In order to incorporate the nonprobability sample, NORC used TrueNorth calibration, an innovative hybrid calibration approach developed at NORC to explicitly account for potential bias associated with the nonprobability sample. TrueNorth uses a tree-based non-parametric supervised learning algorithm to classify respondents based on their survey responses and apply a ratio adjustment to adjust the weights for the nonprobability sample to bring weighted distributions of the nonprobability sample in line with the population distribution for characteristics correlated with the survey variables. Such calibration adjustments help to reduce potential bias, yielding more accurate population estimates. A raking process is used to adjust for any survey nonresponse as well as any noncoverage or under- and oversampling in both probability and nonprobability samples resulting from the study-specific sample design. Raking variables for both the probability and the nonprobability samples included age, sex, census region, race and ethnicity, education, parental status, and 2024 vote choice. Population control totals for the ranking variables were obtained from the 2020–2024 5-year American Community Survey. For the general population, the weighted data reflect the U.S. population of people age 18 and over. For fathers, the weighted data reflect the U.S. population of fathers age 18 and over.
The overall margin of sampling error for the combined sample of 3,345 completed interviews is +/- 2.8 percentage points at the 95 percent confidence level, including the design effect. Men overall and fathers specifically were sampled at a higher rate than their proportion of the population for reasons of analysis. The overall margin of sampling error for the combined sample of 2,229 completed interviews with men is +/- 3.2 percentage points at the 95 percent confidence level including the design effect. The overall margin of sampling error for the combined sample of 1,298 completed interviews with fathers is +/- 4.0 percentage points at the 95 percent confidence level including the design effect. Sampling error is only one of many potential sources of error and there may be other unmeasured error in this or any other survey.
For the open-ended questions BELIEF1, CGAT8B, and F12, responses were initially classified using Microsoft Copilot. These codes were reviewed, validated, and adjusted as needed by a human coder for reporting.
Additional information on the TrueNorth approach can be found here. For more information, email [email protected].
Better Life Lab staff members supplemented survey data with interviews of select survey respondents who opted in to being contacted by a reporter after their survey completion. Unless otherwise noted, each respondent granted us permission to use their real names. Like most large-scale surveys of this kind, this study measured gender as binary (man/woman). It also found about 1 percent of respondents identify as transgender. We recognize this framing does not reflect the full diversity of gender identity, and we present these findings as a description of the data collected and general trends around caregiving and gender.
