The methods used to estimate the Jewish population at the national, state, and sub-state levels continue to develop as new data are added to the data synthesis. The addition of data increases the power to estimate smaller geographic areas. As the power to estimate smaller geographic areas is improved, so too is the ability to explore factors that can affect estimation of these areas.
Previous work included the clustering of respondents within surveys, as well as within geographic areas, such as states or counties (See Publications). Also included in past population models are demographic characteristics related to the representativeness of survey samples and to the distribution of the Jewish population. These demographics include sex, age, educational attainment, and race/ethnicity. All estimates are for the household population, and do not include group quarters (e.g., college residence halls, residential treatment centers, skilled nursing facilities, group homes, military barracks, correctional facilities, and workers’ dormitories).
Bayesian multilevel models, using R and Stan, are used to fit the data. A set of simulations from the models are stored to generate estimates of the proportion of adults who are Jewish for each demographic group within each geographic region poststratified to US population totals.
Sufficient data at the ZIP Code level were available to allow for estimation of smaller geographic areas within counties. These data were used to improve estimation of counties by taking into account variation within counties.
The definitions of county-level areas change on the basis of available data. With the addition of ZIP Code level data, more counties were able to be estimated singly than in previous years. Where possible, counties (and ZIP Code areas within counties) were grouped to align with Federation areas.
Counties outside of these areas were grouped based on recent work with Jewish organizations providing services throughout the US, Birthright registration data, and on information about economic areas from the Cluster Mapping project.
Revised November 2020.
For 2020 models, metropolitan areas were limited to the top 40 metro areas defined by the Consolidated Statistical Based Area (CBSA) definitions of metropolitan areas.
Estimates of those not represented in the model, specifically, Jewish adults who do not identify by religion (JNR), and Jewish children are estimated from independent sources of data and added to the model-based estimates of Jewish adults to obtain estimates of the total Jewish population. See Chapter 4 of the Technical Report: ZIP Code-based Jewish Population Estimates 2020 for details of the analyses.
Methods of estimation are detailed in Technical Report: ZIP Code-based Jewish Population Estimates 2020.
The definitions of county-level areas change on the basis of available data. Currently, the lowest level of geography for national models are counties. Counties were estimated singly, or were grouped, depending on available sample size, with a target of a minimum of 1,000 observations per geographic area for reliable estimation. Counties were grouped to align with UJA Federation areas, or as close an approximation to those areas as possible.
Counties outside of these areas were grouped based on information about economic areas from the Cluster Mapping project. This project is a partnership between the U.S. Department of Commerce, the U.S. Economic Development Administration, and the Harvard Business School's Institute for Strategy and Competitiveness and synthesizes data on industry locations and regional business environments to map geographic clusters that represent regional economies.
Revised November 2019.
For 2019 models, metropolitan areas were limited to 70 areas. This includes the top 40 metropolitan areas along with other major metropolitan areas. Previous metropolitan area models used Consolidated Statistical Based Area (CBSA) definitions of metropolitan areas for clustering of respondents within geographic areas. CBSAs are larger geographic areas than counties. For example, the New York metropolitan area as defined using CBSAs consists of 23 counties in New York, New Jersey, and Pennsylvania. There are more surveys that share the metropolitan area of respondents than there are surveys that share the county of the respondent. Models based on this larger level of geographic clustering have been done in the past to make use of the additional data and because they are relatively easy to fit. In addition, they provide estimates that are "in the ball park", that is, estimates based on the metropolitan area model typically are within the 95% certainty intervals of models based to the counties. They do not, however, yield estimates identical to those based to counties. Estimates could be in either the high or low end of the 95% certainty interval depending on how respondents might have been distributed over the larger geographic area. To avoid confusion for those who wish to compare the two geographic areas, estimates for metropolitan areas have been revised to reflect the county-level distributions.
To limit higher incidence areas influencing estimates of lower incidence areas (and vice versa), counties were grouped into 26 clusters based on census division, population density, and past estimates of Jewish population incidence. Previous models clustered data into 80 groups based on population density and Jewish population incidence. Estimates using the simpler set of 26 clusters were compared to estimates using the more detailed clustering to determine that they did not change estimates of individual county groups.
Estimates of those not represented in the model, specifically, Jewish adults who do not identify by religion (JNR) and Jewish children, are estimated from independent sources of data and added to the model-based estimates of Jewish adults to obtain estimates of the total Jewish population. At the national level, estimates of these two groups were based on the Pew 2013 Survey of American Jews. For counties and metropolitan areas, estimates of these groups are based on recent local Jewish community surveys where available. For areas without such local sources of data, estimates are based on secondary analysis of Pew 2013. Past estimates used the lowest level of geography for which a reliable estimate could be obtained (national, census region, division, state, metropolitan area, or county). Because estimates of these two groups rely solely on single sources of data that cannot be validated against independent sources of data such as a census, new estimates use the lowest possible estimate for the group, regardless of geography so as to err on the side of conservative estimates given the lack of data for these groups. For example, if the national estimate is that 21% of the Jewish population are children aged 17 years and younger, and the county estimate from Pew was 10%, we used the 10% estimate even if this estimate had low reliability.