See the latest developments and updates here.
The AJPP method has been validated using data from Canada and the UK where results could be compared to Canadian and UK Census data which include assessments of religious identification (Magidin de Kramer, Tighe, Saxe & Parmer, 2018; Claassen & Traunmüller, 2018). In the study of Canada, a comparison of three different methods for data synthesis showed that Bayesian Multilevel Regression with Poststratification (BMRP) yielded estimates that are on par with the Canadian Census and out-performed alternative methods. In the study with UK data, the method was demonstrated to yield accurate estimates of the Jewish, Muslim, and Hindu populations over a 20 year period (Claasen & Traunmüller, 2018).
The present release provides the most recent estimates of the proportion of adults who identify their religion as Jewish. The estimates are based on a set of independent samples of the US adult population collected across the years 2012 to 2018 and are adjusted to Census Population Estimates for the year 2018—the most recent year for which census total population data at the county level are available. Additional sources of data were used to provide total population estimates including Jewish children and those who might identify culturally, ethnically, or secularly as Jewish but do not identify as Jewish when asked about religion. These sources include the most recent Jewish population survey conducted by Pew Research Center as well as local Jewish community studies conducted from 2015 to 2018. Total population estimates are available for all geographic areas (states, counties, and metropolitan areas). County groupings and metropolitan areas differ from the last release due (See Methodology). Comparisons with previous estimates as a measure of absolute change, especially for small geographic areas, should be avoided as any differences between estimates published in past work likely represent changes and improvements in methodology more than they do changes in the distribution of the Jewish population within these areas.
Across county and metropolitan areas, estimates derived from the data synthesis vary in terms of how similar they are to population counts derived from other sources, such as those reported in the American Jewish Yearbook (AJYB). There are many areas, small and large, where the data synthesis yields similar estimates to the yearbook. For as many areas where our model-based estimates are similar to those reported in the AJYB, there are many where the estimates are disparate. There are also cases where the AJPP estimates appear different from a local Jewish community study. This can be due to differences in geographic areas covered by AJPP and the local study as well as differences in definitions of who is Jewish. See our detailed research note, Why Do These Estimates Differ from Other Published Estimates, for further discussion of these issues.
A majority of the surveys limit their sampling frames to the contiguous United States. Estimates for Alaska and Hawaii are, therefore, not included in our major national models. Future work can explore differences between surveys that include Hawaii and Alaska and those that do not to develop separate estimates for these states.
For the surveys in our data synthesis, the lowest level of geography for which there is reliable data across the largest number of surveys is county. There are more than 3,000 individual counties in the United States. While larger counties have sample sizes that enable reliable estimation, many smaller counties have too few cases to be estimated on their own. We aim for a minimum sample size of approximately 1,000 cases in each county group. For a few states—such as Maine, New Hampshire, North Dakota, and Vermont—this results in all counties within each state having to be combined. As we continue to develop models and add new sources of data, it may be possible to do custom analyses within these states to obtain estimates for specific counties of interest.
For the surveys in our data synthesis, the lowest level of geography for which there is reliable data across the largest number of surveys is county. We collect data at lower levels of geography where it is available; however, surveys vary in the type of information that is available. If there is a particular area you are interested in, contact us! We're happy to talk about our data and to see what is possible.
We collect data for all religiously defined groups represented in the surveys that are part of the data synthesis. This includes Protestant (denominations and subgroups within, including evangelical and mainline), Catholic, Mormon, and Muslim, as well as no religion and other religions. With the exception of Catholicism, religious representation varies considerably across surveys. For example, some surveys might not distinguish a general category of "Mormon" from the more specific LDS Church, or might only include general categories such as "Methodist", "Baptist", or "Lutheran" for Protestant denominations and ignore differences between specific denominations within those broad categories. We have not yet explored this variability or developed population models for these other groups. We have, however, begun to explore and model the large group who identify with no religion, which also varies across surveys in how it is assessed (atheist, agnostic, nothing in particular, etc.).
A number of other variables are collected as part of the data synthesis project. The Jewish population models currently include only those variables related to model-based estimation of the population as a whole. These are male/female, age, educational attainment, and race/ethnicity, as well as geographic variables of state, counties, and metropolitan areas. Other variables collected as part of the project include: marital status, income, whether the respondent was born in the US, political orientation, political affiliation, Jewish denomination (Orthodox, Conservative, Reform, etc), religious service attendance, religious orientation (fundamentalist/evangelical), the importance of religion, household size, number of adults and children in the household, and community type (urban/suburban/rural; city, suburb, etc.). Surveys vary in which demographic data are included so sample sizes and numbers of samples vary for each analysis. A number of survey "meta-data" are recorded as well, such as date and day of the week the survey was conducted, incentives, and number of calls required to complete the survey. The dataset also includes methodological characteristics of all surveys included in the synthesis, such as response rates, length of time in the field, survey shop, sampling methods, survey purpose, and questions used to assess religious affiliation.
The "Low" and "High" values represent the 95% credible intervals associated with each estimate. Credible intervals based on Bayesian analysis are similar in concept to confidence intervals reported in single surveys, which are often described in terms of the "margin of error." Both the credible intervals and the margin of error represent the degree of certainty associated with the estimate. They differ in how they are calculated and how they are interpreted. For example, if one estimates the proportion of US adults who are Jewish to be 1.9% with a margin of error of 0.03, one would add or subtract this amount times some critical value to describe the variability in the estimate. For a 95% Confidence Interval, one would multiply the margin of error by 1.96, yielding a distribution that ranges from 1.84% at the lower end to 1.96% at the upper end. The confidence interval is based on the assumption that any sample drawn using the same sampling methods as those used in the particular survey of interest will yield a somewhat different estimate and a somewhat different range on that estimate. A 95% confidence interval means that, were one to repeat the sampling, one could expect 95% of the interval estimates to fall within the population parameter. In the Bayesian analysis, we are synthesizing data across repeated samples directly and are, therefore, reporting on results of the repeated samples themselves, rather than reporting on the assumption of repeated samples. For our national estimates, across all of the repeated independent samples, the estimated proportion of US adults who identify as Jewish is 1.8%, with a 95% credible interval of 1.86% to 1.96%. This means that there is a 95% probability, based on the data and prior information, that the true population proportion of US adults who identify their religion as Jewish is between this interval. For the demographic data, we have only included the point estimate on the map. For the full range of credible values associated with all of the detailed population estimates, see the detailed tables (requires registration).
Not at all! The data synthesis is not intended to replace the need for local or targeted studies of the Jewish community. Rather, it fills a specific gap that has persisted in survey research of American Jews for more than 40 years. All survey data requires adjustments or weighting to the known population from which it is sampled. Typically this is done using the US Census. However, this has not been possible because census data does not include a question about religion or Jewish identity. Weighting that is done in local surveys typically adjusts for standard demographics of the area, with the assumption that the Jews who ended up in the sample are representative of all Jews in the target area. There is no way to evaluate or adjust for whether the Jewish sample is representative of all Jews in that area. For many local study surveys, especially those which rely heavily on advertising to increase response rates among the Jewish population, this can lead to an inflation in the estimation of the size of the population. The data synthesis provides Census-like data for the Jewish population in order to provide a population profile that can serve as a baseline for those conducting targeted surveys and studies of the population.
This release of the map includes updates to the user interface and your computer may have stored a cookie from the previous version to speed up its loading. You can clear your browser's cookies or press Control+F5 (Cmd+Shift+R on a Mac) to completely reload the page and update the cookies.
We would love to speak with you to see how we can help. Get in touch with us using one of the links on the website, or send an email to email@example.com.
Don't see detailed tables you need on our data page? Contact us! We're happy to work with you to provide comparisons you need. Send an email to firstname.lastname@example.org.