Elevating Voices 2026: Methodology Summary

This summary describes how data for Feeding America’s 2026 Elevating Voices Insights Report were collected, reviewed, and weighted. It is intended to help readers understand the study design, the people represented in the survey, and considerations for interpreting the findings. 

Survey Design and Fielding 

Feeding America partnered with the nonpartisan research firm PerryUndem to design a national survey of adults age 18 or older who reported that their household received charitable food assistance and/or experienced food insecurity within the past two years. NORC at the University of Chicago administered the survey, using its probability-based AmeriSpeak® Panel as the primary sample source and Dynata to provide a supplemental nonprobability sample. 

The survey was self-administered online in English and Spanish from June 16 through July 6, 2026. The median completion time was 20 minutes. The final analytic sample included 1,404 respondents: 1,257 from AmeriSpeak and 147 from Dynata. 

Respondent Eligibility 

Eligible respondents were adults age 18 or older who answered “Yes” to at least one of the prompts below: 

At any time in the last two years, have you or anyone in your household:

  • Worried whether your food would run out before you got money to buy more?
  • Had the experience where the food you bought didn’t last and you didn’t have money to buy more?
  • Gotten free groceries from a food pantry, food bank, church, or other place that helps with free food?
  • Received a free meal from a church, shelter, home-delivered meal service like Meals on Wheels, or other place that helps with free meals? 

Sampling Approach 

The primary sample consisted of 1,257 completed responses from members of NORC’s AmeriSpeak Panel. AmeriSpeak is a probability-based panel designed to represent the U.S. household population. An additional 147 completed interviews came from Dynata, a nonprobability sample provider. Only AmeriSpeak respondents were included in the study’s statistical weighting and population-level estimates. 

AmeriSpeak participants were selected using sampling strata based on age, race and ethnicity, education, and sex. The study oversampled non-Hispanic Black, Hispanic, and non-Hispanic Asian American and Pacific Islander adults to support more reliable subgroup analyses. NORC also supplemented the AmeriSpeak sample with Dynata interviews to reach requested interview targets for Puerto Rican adults and American Indian and Alaska Native adults. The non-Hispanic Asian American and Pacific Islander target was not fully reached before fielding ended. Complete demographic details for the sample can be found on pages 10 and 11 of the Elevating Voices Insights Report. 

AmeriSpeak panelists received the cash equivalent of $2 for completing the survey. To encourage participation among non-Hispanic Asian American and Pacific Islander panelists, NORC sent targeted reminders and increased the incentive to $5 on the final day of data collection. NORC did not determine the incentive offered to Dynata respondents. 

Survey Precision and Response 

For the weighted AmeriSpeak sample, the estimated margin of sampling error was ±3.78 percentage points at the 95% confidence level, incorporating a design effect of 1.87. Margins of error are larger for subgroup estimates. The AmeriSpeak weighted cumulative response rate was 8.8%. A conventional probability-based margin of sampling error does not apply to the Dynata sample. 

Data Quality 

NORC reviewed all completed surveys for data quality. Responses were excluded if respondents completed the survey in less than four minutes, skipped or refused more than half of the questions they were eligible to answer, or selected the same response across all applicable grid questions. Because some cases met more than one criterion, the counts overlap; 92 cases were removed in total. 

AmeriSpeak uses controlled invitations, secure panelist accounts, and unique identifiers to help ensure that only sampled panel members participate and that each person completes the survey once. 

Because nonprobability samples face additional risks from duplicate, automated, or otherwise invalid responses, Dynata respondents were also screened using attention checks, age and birth-year consistency checks, reCAPTCHA, and Research Defender. These safeguards blocked 12 suspicious cases and 18 cases that failed at least one attention check; no cases failed reCAPTCHA. These exclusions occurred before NORC’s review of completed interviews. 

Survey Weighting 

Survey weights were developed for the AmeriSpeak respondents to reduce potential bias associated with unequal selection probabilities, nonresponse, and coverage limitations. The final study weight combines the existing AmeriSpeak panel weight, each respondent’s probability of selection for this study, and study-specific adjustments for nonresponse. 

Because no external benchmark is available for the full target population of adults who recently experienced food insecurity, NORC used a two-step process. First, respondents who completed the eligibility questions were adjusted to February 2026 Current Population Survey benchmarks for U.S. adults. NORC then used the weighted profile of eligible respondents to create study-specific benchmarks and calibrated completed AmeriSpeak interviews to those benchmarks. Adjustments included age, race and ethnicity, education, sex, region, and American Indian and Alaska Native status. Extreme weights could be trimmed and recalibrated to limit their influence on estimates. 

NORC also created a separate analysis weight for the non-Hispanic Black, Hispanic, and non-Hispanic Asian American and Pacific Islander oversamples. This weight is intended for analyses focused on those groups or comparisons involving them; it should not be used to estimate results for the full sample. Dynata respondents were not included in either set of weights.

Interpreting the Findings 

Like all surveys, this study is subject to sampling error and other sources of uncertainty, including nonresponse, coverage limitations, recall error, and differences in how respondents interpret questions. Weighting reduces—but does not eliminate—potential bias. Estimates based on the Dynata interviews should be treated as descriptive because those respondents were selected through a nonprobability process and were not included in the population weights. Results for small subgroups should be interpreted with particular caution. 

This summary includes edits suggested by Microsoft 365 Copilot. All content was reviewed and verified by the authors.