Written and maintained by CASRAI Editorial Board
Published · Last updated
On this page
The Pew Religious Landscape Study (RLS) is Pew Research Center’s large, periodic survey of religion among adults in the United States, fielded in 2007, 2014 and 2023-24. The latest edition completed 36,908 interviews, large enough to report results for every state and for 34 metropolitan areas, and Pew publishes the dataset, questionnaire and methodology so that other researchers can analyse it. This guide explains how the 2023-24 study was built, what its margins of error mean for the groups you may want to study, and how far it can be compared with the earlier editions.
What is the Pew Religious Landscape Study?
Pew describes the RLS as a survey of Americans’ religious affiliation, beliefs, practices and social, political and demographic characteristics. Three editions exist: 2007, 2014 and 2023-24. The 2023-24 edition sampled adults in all 50 states and the District of Columbia, and Pew’s own summary of its headline findings includes that about 35% of adults no longer identify with the religion they were raised in, and that the decline of Christianity may have leveled off. Those findings are reported on Pew’s study pages; this guide focuses on the design behind the numbers, because that is what determines how the data can be used in research.
The study is a Pew Research Center product, and it was conducted for Pew by NORC at the University of Chicago. For how Pew work is treated in scholarly practice, including the question of whether it is peer reviewed, see Is Pew Research Center peer reviewed?. For the field that studies this subject matter, see what religious studies is.
The three editions: 2007, 2014 and 2023-24
The 2007 and 2014 editions were conducted entirely by telephone, using random-digit dialing and live interviewers, according to Pew’s appendix on comparing the studies. The 2023-24 edition changed the design: it used address-based sampling with mailed invitations, and respondents could answer online, on paper or by phone. The appendix notes that telephone response rates now hover near 3%, compared with 20% for the new survey.
That mode change is the most important fact for anyone planning a trend analysis, and it is covered in the comparison section below.
How the 2023-24 study was designed
Sample and fieldwork
Pew’s methodology page reports 36,908 completed interviews with U.S. adults, fielded from July 17, 2023 to March 4, 2024 in English and Spanish. NORC drew an address-based sample from the U.S. Postal Service Computerized Delivery Sequence file, maintained by Vericast and updated monthly, and mailed invitations to 205,100 sampled addresses. The protocol was sequential: online first, then a paper questionnaire to nonrespondents, with a toll-free phone option in every contact. Respondents completed 25,250 interviews online, 10,733 on paper and 925 by phone.
Stratification
Addresses were assigned to 93 strata defined by state, the District of Columbia and, where sample sizes allowed, target metropolitan areas and non-metropolitan areas. The allocation targeted at least 300 completes in every state and at least 250 in large metropolitan areas. A minimum per state means small states are sampled at higher rates than large ones, so selection probabilities are unequal, and the weights correct for that (see survey weighting: design weights, post-stratification and raking).
Response rate
Pew reports an AAPOR Response Rate 3 of 20%. The American Association for Public Opinion Research publishes the Standard Definitions that govern how disposition codes and outcome rates such as RR3 are calculated, so the figure can be compared with other surveys that follow the same definitions.
Weighting
The methodology page lists the steps in order: a base weight equal to the inverse of each address’s selection probability; an eligibility and nonresponse adjustment within each stratum; a person-level weight that multiplies the household weight by the number of adults in the household, truncated at 3; an adaptive mode adjustment that doubled the relative weight of respondents who answered offline (by paper or phone); raking to benchmarks for noninstitutionalized U.S. adults, with weights trimmed at the 1st and 99th percentiles within strata; and, for the 34 metropolitan areas with at least 250 interviews, a further calibration within each area. A mode adjustment of this kind is a design choice that your own analysis inherits, so it belongs in your methods section if you use the weights.
Margins of error: what they are and what the table shows
The national margin of error is plus or minus 0.8 percentage points at the 95% confidence level. Pew states that it accounts for the complex design and weighting and that it is calculated with a formula based on the variability of respondent weights rather than the bootstrap replicates, a formula that can be quite conservative. Pew also publishes a table of unweighted sample sizes and margins of error for the country, regions, every state, 34 metropolitan areas, and dozens of religious, demographic and attitudinal groups. The data table below this guide reproduces 21 of those rows and adds two computed columns, so you can see how fast the margin grows as the group shrinks.
Three patterns are visible in it. First, the national figure is very precise, but a group is only as precise as its own sample: Catholics (6,958 interviews) have a margin of 1.7 points, Muslims (273) 8.3, Hindus (247) 8.7 and Orthodox Christians (225) 9.9. Second, state estimates range from 2.4 points in California (3,746 interviews) to 7.8 in Iowa (310). Third, the published margin is larger than the margin a simple random sample of the same size would give, which is expected when the design involves unequal weights.
Worked example: what the design does to the margin
For a share of 50%, a simple random sample of n respondents has a 95% margin of 1.96 × √(0.25 / n). The arithmetic below is CASRAI’s own calculation from Pew’s published sample sizes and margins.
- U.S. adults: n = 36,908, so √(0.25 / 36,908) = 0.002603 and the margin is 1.96 × 0.002603 = 0.51 points. Pew publishes 0.8. The ratio (0.8 / 0.51)² is about 2.5, an approximate design effect.
- Muslims: n = 273, so √(0.25 / 273) = 0.03026 and the margin is 1.96 × 0.03026 = 5.93 points. Pew publishes 8.3, giving (8.3 / 5.93)² of about 2.0.
The published margin is therefore about 1.4 to 1.6 times what the sample size alone would suggest. The practical lesson is to use the published margins, not a margin you compute from n, when you report the study; the computed values are only a way of seeing the design effect.
Comparing two groups
The margin of error on a difference between two independent groups is larger than either margin. A common approximation combines them as the square root of the sum of squares. For Catholics (1.7) and evangelical Protestants (1.5) that is √(1.7² + 1.5²) = √(2.89 + 2.25) = 2.27 points. For Hindus (8.7) and Buddhists (8.4) it is √(75.69 + 70.56) = 12.1 points, so a gap of ten points between those two groups on a single question would not be distinguishable from sampling error. This approximation assumes the two groups do not overlap and the estimates are independent. When groups overlap (a state and the nation, or a religious group and an age group), use the replicate weights instead, as described below. For the general logic, see confidence intervals for a proportion and the dictionary entry on the confidence interval.
| Reading guidance | What the published margin tells you |
|---|---|
| Margin under about 2 points | Large groups: national, regions, Christians, Catholics, the unaffiliated. Differences of a few points between them are usually distinguishable. |
| Margin of 5 to 8 points | Mid-sized states and groups such as Jewish adults (5.0). Report ranges rather than point estimates. |
| Margin of about 8 to 11 points | Small groups with a few hundred interviews or fewer (Muslims, Hindus, Buddhists, Orthodox Christians). Estimates are indicative only. |
The bands in this table are the author’s reading of the published margins, not a rule from Pew.
Can you compare 2023-24 with 2007 and 2014?
Partly. Pew ran a separate phone bridge study of 1,519 adults (interviewed August 7 to November 6, 2023, margin of error 3.3 points) to measure how the new mode changes answers, and it sorted questions into three categories in its comparison appendix.
- Trendable: no significant mode effect. Includes religious identity, childhood religion, daily prayer, born-again identity, race, education, age and nativity.
- Compare cautiously: modest mode effects. Includes belief in heaven and hell, views of abortion and homosexuality, the importance of religion and church membership.
- Not comparable: relatively large mode or context effects. Includes religious service attendance, party identification, ideology, scripture reading, spiritual experiences, household income and sexual orientation.
Pew says phone surveys inflate reported attendance, so it does not compare the 2023-24 attendance figure with earlier years. If your research question involves a trend, check each variable against that list before analysing it, and say in your report which category it falls in. Related sources of error are discussed in survey bias.
Using the dataset in your own analysis
Pew’s methodology page says the 2023-24 dataset, topline and questionnaires are available on its site. The design has four consequences for analysis.
- Use the weights. The unweighted sample over-represents small states and was adjusted for mode, so unweighted shares are not national estimates.
- Use the replicate weights for variances. Pew created 500 Rao-Wu rescaling bootstrap replicates, so each respondent carries 501 weights: the full-sample weight plus 500 replicates. Survey software that accepts replicate weights uses them to estimate standard errors that reflect the complex design. No replicate weights were created for the metropolitan-area weights, so metropolitan estimates need the published margins instead.
- Check the cell sizes. A cross-tabulation of a small religious group by state can fall far below the group’s own sample size, and the margins in the table rise accordingly.
- Treat Pew’s margin as an upper-end guide. Pew notes its formula can be conservative.
Calculating the margins in Python
This snippet reproduces the arithmetic in the worked example. It computes the simple-random-sample margin, the implied design effect, and the combined margin for a difference between two independent groups.
import math
def srs_moe(n, p=0.5, z=1.96):
return 100 * z * math.sqrt(p * (1 - p) / n)
def diff_moe(m1, m2):
return math.sqrt(m1**2 + m2**2)
n, published = 273, 8.3
print(round(srs_moe(n), 2)) # 5.93
print(round((published / srs_moe(n))**2, 2)) # 1.96
print(round(diff_moe(1.7, 1.5), 2)) # 2.27
How to cite and report the study
Pew’s methodology page gives a recommended citation for the study report: Smith, Gregory, Alan Cooperman, Becka Alper, Besheer Mohamed, Chip Rotolo, Patricia Tevington, Justin Nortey, Asta Kallo, Jeff Diamant and Dalia Fahmy. 2025. “Decline of Christianity in the U.S. Has Slowed, May Have Leveled Off.” Pew Research Center. doi: 10.58094/4kqq-3112. When you use the data rather than the report, also cite the dataset page and state the edition (2023-24), the sample size, the fieldwork dates, the mode design and the margin of error that applies to the group you analyse. For a probability-sample survey with a design effect, name the weight variable you used and whether your variance estimates use the replicate weights.
When you describe the response rate, use the AAPOR definition number, as Pew does (RR3, 20%), so readers know what the figure includes. Margin of sampling error is only one component of total survey error; nonresponse, coverage and mode effects are not captured by it, which is the reason the comparison categories above matter. For frame and coverage issues see sampling frame and coverage error.
Sources
- Pew Research Center: Religious Landscape Study, checked 2026-11-17
- Pew Research Center: Religious Landscape Study methodology, checked 2026-11-17
- Pew Research Center: 2023-24 RLS sample sizes and margins of error, checked 2026-11-17
- Pew Research Center: Comparing results across Religious Landscape Studies, checked 2026-11-17
- American Association for Public Opinion Research: Standard Definitions, checked 2026-11-17
Data: Pew 2023-24 Religious Landscape Study: subgroup sample sizes, published margins of error and implied design effects
Unweighted sample sizes and published 95% margins of error for 21 groups in the 2023-24 Religious Landscape Study, taken from Pew Research Center's sample-size table. Two columns are computed by CASRAI: the margin of error a simple random sample of the same size would give at a 50% share (1.96 x sqrt(0.25/n)), and the ratio of the published margin squared to that value, an approximate design effect.
Period covered: 2023-07-17/2024-03-04. Coverage: United States.
| Group | Unweighted sample size (respondents) | Published margin of error (percentage points) | Simple-random-sample margin (computed) (percentage points) | Implied design effect (computed) (ratio) |
|---|---|---|---|---|
| U.S. adults | 36908 | 0.8 | 0.51 | 2.5 |
| Adults in the South | 13423 | 1.2 | 0.85 | 2 |
| Adults in the Northeast | 6717 | 1.7 | 1.2 | 2 |
| Adults in California | 3746 | 2.4 | 1.6 | 2.2 |
| Adults in Texas | 2988 | 2.5 | 1.79 | 2 |
| Adults in Alabama | 498 | 6.2 | 4.39 | 2 |
| Adults in Iowa | 310 | 7.8 | 5.57 | 2 |
| Christians | 23186 | 0.9 | 0.64 | 2 |
| Catholics | 6958 | 1.7 | 1.17 | 2.1 |
| Evangelical Protestants | 8298 | 1.5 | 1.08 | 1.9 |
| Mainline Protestants | 5333 | 2 | 1.34 | 2.2 |
| Religiously unaffiliated adults | 10729 | 1.4 | 0.95 | 2.2 |
| Jewish adults | 850 | 5 | 3.36 | 2.2 |
| Muslims | 273 | 8.3 | 5.93 | 2 |
| Buddhists | 348 | 8.4 | 5.25 | 2.6 |
| Hindus | 247 | 8.7 | 6.24 | 1.9 |
| Orthodox Christians | 225 | 9.9 | 6.53 | 2.3 |
| Methodists in the evangelical tradition | 148 | 9.9 | 8.06 | 1.5 |
| Members of the Global Methodist Church | 118 | 10.9 | 9.02 | 1.5 |
| 18- to 29-year-olds | 3524 | 2.2 | 1.65 | 1.8 |
| Adults ages 65 and older | 13289 | 1.2 | 0.85 | 2 |
Sources
- Pew Research Center: 2023-24 RLS sample sizes and margins of error (accessed 2026-11-17)
Compiled by CASRAI from the sources above. Free to reuse under CC BY 4.0 with attribution.







