Appendix A: Methodology

Download the Methodology as a PDF (180 KB, 17pages)

This study provides comprehensive demographic estimates of the size and distribution of eight major religious groups in the 232 countries and territories for which the United Nations Population Division provides general population estimates as of 2010.¹⁶ It includes estimates for Christians, Muslims, the religiously unaffiliated, Hindus, Buddhists, folk religionists, adherents of other religions and Jews. This study also provides estimates for the median ages of these religious groups at the regional and global level.

This appendix describes various stages of the work that led to this study. First, it describes the data and procedures used to derive estimates of the overall religious composition of each country. Second, it describes the procedures used to estimate religious composition in each country by age and sex (e.g., males between the ages 0-4, or females ages 95+). (This data was needed to help calculate the median ages of the religious groups.) Third, this appendix describes how additional data on fertility, migration and religious switching were collected. (This information was used to project the religious composition of countries to 2010 when the primary data on religious composition for a country was collected prior to 2008.) The appendix ends with a discussion about measuring religion in China, notes about estimates that vary from previously published Pew Forum reports and a list of the countries included within each of the six regions used in this report.

General Procedures for Composition Estimates

Data Collection and Documentation

Pew Forum researchers acquired and analyzed religious composition information from about 2,500 data sources, including censuses, demographic surveys, general population surveys and other studies – the largest project of its kind to date. Censuses were the primary source for Pew Forum religious composition estimates in 90 countries, which together cover 45% of all people in the world.¹⁷ Large-scale demographic surveys were the primary sources for an additional 43 countries, representing 12% of the global population. General population surveys were the primary source of data for an additional 42 countries, representing 37% of the global population.

Together, censuses or surveys provided estimates for 175 countries representing 95% of the world’s population. In the remaining 57 countries, representing 5% of the world’s population, the primary sources for the religious-composition estimates include population registers and institutional membership statistics reported in the World Religion Database and other sources. A list of the primary sources used to estimate the overall religious composition of each country is provided in Appendix B.

Censuses and nationally representative surveys can provide valid and reliable measures of religious landscapes when they are conducted following the best practices of social science research. Valid measurement in censuses and surveys also requires that respondents are free to provide information without fear of negative governmental or social consequences. However, variation in methods among censuses and surveys (including sampling, question wording, response categories and period of data collection) can lead to variation in results. Social, cultural or political factors also may affect how answers to census and survey questions are provided and recorded.

Standardization of Religious Categories and Measurement Strategies

At least three researchers worked together to produce each country’s religious composition estimates, observing the general procedures and considerations described below.

Pew Forum researchers standardized religion categories in all available censuses and surveys for each country. Censuses and surveys collect religious identity at different levels of specificity. For example, depending on the source, the most specific level of affiliation measured could be Christian, Protestant, Baptist or Southern Baptist. Researchers coded religious identities into standard categories that aggregate into the eight major religious categories used in this report.

The measure of religious identity in this study is sociological rather than theological. In order to have statistics that are comparable across countries, the study attempts to count individuals who self-identify with each religion. This includes people who hold beliefs that may be viewed as unorthodox or heretical by others who claim the same religion. It also includes people who do not regularly practice the behaviors prescribed by their religion, such as those who seldom pray or attend worship services.

Identifying Primary Source(s) for Religious Composition Estimates

After considering all evidence available, researchers identified one or more primary sources for each religious composition estimate. Researchers sought a recent, reliable source – ideally, a census or large-scale demographic survey. Researchers favored sources in which religion was measured with a single question that permitted respondents to identify specific affiliations or no affiliation at all.¹⁸

Nationally representative surveys were occasionally chosen as a primary source rather than a census or demographic survey due to limitations in the measurement of religion in the latter sources. In Vietnam, for example, the 2009 census and the 2005 Demographic and Health Survey did not adequately measure folk religion identities. Researchers instead relied on the 2005 Asian Barometer survey, which measured a wider range of religious identities, including identification with folk religions.

Making Adjustments for Groups Not Adequately Measured

As necessary, researchers made adjustments to the primary source(s) to account for omitted or underrepresented groups since small minority groups are sometimes not measured or not reported in surveys and censuses. Multiple survey sources, denomination counts and estimates produced by country experts for each nation were used to assess whether minority religious groups were omitted or undercounted in the selected primary source(s).

In cases where censuses and surveys lacked sufficient detail on minority groups, the estimates also drew on estimates provided by the World Religion Database, which takes into account other sources of information on religious affiliation, including statistical reports from religious groups themselves.

Adjusting for Limitations in a Survey Questionnaire

Usually, researchers assumed that members of underrepresented groups were included in the sample but were not adequately measured by the survey instrument. Adjustments frequently were made among people who responded “other religion” or failed to answer the religion question. In a few cases, the study made adjustments based on evidence that political, legal, or cultural dynamics in a country compromised the validity of self-reported religious identity.

In India, for instance, there is evidence of a Christian undercount in the 2001 census; some Christians who belong to Scheduled Castes (historically referred to as Untouchables or Dalits) choose to identify as Hindu when completing official forms such as the census.¹⁹ This is due to a mandate in the Indian constitution that specifies that only Hindus, Sikhs and Buddhists can receive caste-based government affirmative-action benefits (known as reservations in India).²⁰ After analyzing Indian survey data and convening a special consultation on this topic with leading India demographers at the 2010 Asian Population Association’s meeting in New Delhi, Pew Forum researchers adjusted the Christian population of India’s 2001 population from 2.3% to 2.6%, assuming a 10% undercount in the census. In this case, the adjustment comes from the Hindu category. (Hinduism is the most common religion in India.)

Adjusting for Sampling Limitations

In some situations, underrepresented groups are likely to be omitted from the sample itself. For example, recent migrants who may not be fluent in the language used in a survey are often missing in samples. Accounting for groups not included in the sample requires proportionately deflating survey data to account for underrepresented populations. For example, researchers made adjustments to survey-based estimates in Europe where they found evidence that some survey samples and population registers underrepresented Muslim migrants.

In this study, researchers sought to ensure that primary sources were representative of the entire country. When this was not the case, it was usually due to concerns about the safety of interviewers and census takers or disputes about political boundaries. In such cases, researchers attempted to make appropriate adjustments or find an alternative data source that was nationally representative.

For example, the 2001 Sri Lankan census was not conducted in a handful of northern and eastern districts because of perilous conditions due to armed conflict. After analyzing religion data from earlier censuses, Pew Forum researchers determined that the areas that were not covered by the 2001 census historically had a different makeup than the rest of the country. Pew Forum researchers adjusted the 2001 census data for Sri Lanka based on 1981 census data about regions omitted in the 2001 census.²¹

Religious identity is sometimes linked to ethnic identity, particularly for religious minorities. In a small number of countries where the census did not measure religious affiliation or where survey data on religious affiliation had sampling limitations, researchers used ethnicity data to estimate the religious affiliation of small groups. For example, ethnicity data from the 2002 Russian census was used together with 2005 Generations and Gender Survey data to estimate the proportion of Muslims in Russia. The survey did not adequately sample the country’s predominantly Muslim areas but it did provide information on the share of Muslims within ethnic groups associated with Islam.²² This information, combined with census ethnicity data, was used to adjust the Muslim composition estimate in regions the survey sampled inadequately.

Making Adjustments for the Religious Affiliation of Infants

Parents are sometimes hesitant to report a religious affiliation for their infant children even though they will claim a religion for the child when he or she is slightly older. Forum researchers observed evidence of this phenomenon in some Christian-majority countries where Christian parents were disproportionately describing their infants as religiously unaffiliated.

This is evident when comparing census numbers over multiple years. In Brazil, for example, the 2000 census reported that 11% of those ages 0-4 were unaffiliated. By the time of the 2010 census, only 8% of the same birth cohort (who were then 10 to 14 years old) was unaffiliated. While some of this change may be explained by mortality and migration, it is at least partly due to parents being more willing to describe their older children as Christians.

In order to compensate for this measurement bias in Brazil and a few other countries where there was considerable evidence of this phenomenon, Pew Forum researchers applied the religious composition of older children (those 5-9 years old) to infants and young children (those 0-4 years old). This adjustment was made only where there was a substantial difference between the religious composition of the youngest age group and children ages 5-9. (For countries in which this adjustment was made, there is a corresponding note in the list of sources by country in Appendix B.)

Making Adjustments for Missing Religion Data

Some degree of missing data is found in most surveys and censuses. Census agencies typically make adjustments for missing data before reporting results. For example, though some respondents fail to answer questions about sex and age, census agencies follow procedures to impute missing values so it is not necessary to report “sex not stated” and “age not stated” as sex and age categories. Some census agencies, such as Statistics Canada, have historically imputed religion values for respondents who have not answered the census religion question.

The likelihood that religion data will be missing increases when religion questions are labeled as optional, as is the case in censuses in countries such as Australia, the United Kingdom and the Czech Republic.²³ Census agencies in countries where religion is labeled as an optional question often report “religion not stated” results alongside standard categories of affiliation and non- affiliation. This strategy allows the census agencies to demonstrate that answering their religion question was indeed optional. However, for purposes of this study, the “religion not stated” category is not a meaningful religious identity. Therefore, after making any necessary adjustments for undercounted groups, religious shares were re-calculated based on the population of all people who gave valid responses to the census or survey. The effect of this approach was to proportionately raise the shares of all religious groups, including the religiously unaffiliated.

Following the procedures described above, researchers produced national-level estimates of the religious composition of each country for the year measured by the primary source. In order to report 2010 population figures, the religious composition percentages based on data collected in 2008 or later have been multiplied by the U.N.’s 2010 population estimate for each country and territory.²⁴ Estimates based on data collected prior to 2008 have been projected forward to 2010. These projections required additional information, including the religious composition of age and sex groups within each country.

Researchers from the International Institute for Applied Systems Analysis (IIASA) in Laxenburg, Austria, collaborated with Pew Forum researchers on data collection and religious composition estimates for Europe; Pew Forum researchers gathered data outside Europe. IIASA researchers also collaborated on development of the procedures described below, which were used to standardize information across countries for age breakdowns, childbearing patterns, migration flows and religious switching patterns. Researchers from the Vienna Institute of Demography provided initial estimates of country-to-country migrant flows by age and sex.

Age Structure Procedures

Religious affiliation varies by age.²⁵ In this section, the phrase “age structure” is used as shorthand to refer to the religious composition of age-sex groups. In order to calculate the median ages of religious groups and carry out population projections, Pew Forum researchers assembled age structures for each of the eight religious groups in every country. Data on age structures were collected in 20 age categories (measured in five-year increments with a top value of 95 and above) for males and females (e.g. males between the ages of 15 and 19), resulting in a total of 40 categories.

Age structures were compiled in three steps. First, census or survey data were used to capture the religious affiliation of each available age group. Second, survey data on religion by age were adjusted to account for small sample sizes. And third, results were adjusted to match the religious breakdown by age and sex to each country’s overall religious composition. These steps are described in detail below.

Estimating Religion by Age and Sex

Pew Forum researchers constructed initial age structures by analyzing survey datasets, census datasets and tables published by census agencies. While censuses usually enumerate religion for the entire population, including children, general population surveys do not usually include interviews with children. Since age structures require religious affiliation data for children, children were assigned religious affiliations when necessary based on the best methods available. For datasets that measured religious affiliation only for adult respondents, yet included the number and ages of children (and other adults) in the household, researchers were able to estimate the religious affiliation of remaining household members. In most cases, the religious affiliation of the respondent or head of the household was assigned to all additional members of the household who were not surveyed.²⁶ For surveys that did not offer household information, such as the AmericasBarometer or the European Values Study, children were assigned a religious affiliation based on the fertility patterns and religious affiliation of women of childbearing age, as well as information about the religious affiliation of the youngest respondents measured in the survey.

For many countries, reliable age data were not available for all eight religious groups. Sometimes a survey indicated the overall size of a small religious group yet lacked sufficient numbers to reliably estimate the group’s age breakdown. In such cases, the age breakdowns of minority religious groups were based on the country’s overall age distribution or the combined age distribution of respondents from all minority religions in a survey.

Adjustments to Minimize Errors Due to Sample Size

The reliability of survey estimates is partly dependent on the number of people surveyed (the sample size). Since respondents who identify with religious groups are divided into 40 age and sex categories, the number of Muslims, for example, in any one age-sex category may be small and produce less reliable estimates than a larger count would produce. This introduces significant variation in patterns of religious affiliation by age: affiliation levels may bounce between highs and lows for consecutive age groups. To eliminate unlikely variation, researchers smoothed data using statistical procedures intended to reveal the general underlying pattern.²⁷

Census data are not smoothed because census data represent a complete enumeration of all individuals living in a country. However, in some cases, the age categories reported by census agencies are in 10-year age groups or aggregated for all adults above a certain age, such as 60. Researchers used statistical modeling techniques to distribute the composition of these aggregated age groups across the more specific five-year age categories used in this study.

Matching Religion by Age and Sex to Overall Population by Age and Sex

The overall religious affiliation resulting from the age structure procedures sometimes varies from the religious composition estimated for the country using the procedures described in the first section of this appendix. This difference exists for two reasons. First, the data sources used for the age-structure procedure may be different from the data sources used for the religious composition. Second, overall religious compositions were adjusted manually to account for undercounts and sampling issues.

In order to match the overall religious composition figures to the data on religion by age and sex, the age structure was adjusted. The adjustment procedure used is often referred to as Iterative Proportional Fitting (IPF), or raking. Raking makes adjustments to the percentages of religious affiliation for each age group without significantly altering the underlying religious affiliation patterns by age group. Additionally, raking is used to match each country’s counts of males and females in particular age groups to the U.N.’s estimate of the country’s overall age distribution.

When survey or census data on the differential religious composition of age-sex groups was not available, each age-sex group was assigned the same religious composition. Lack of differential religious composition data by age-sex group was most common in countries with very small populations.

Sources for Age Structure Data

As mentioned above, the data source used for the country’s age structure is sometimes different from the source used for the overall religious composition (see Appendix B for a list of religious-composition sources by country).²⁸ This is the case, for example, when census data with overall religious composition results are available but a detailed breakdown by age and sex is not released by the census bureau; another source must therefore be used to generate the age structure. Sources are also different when multiple waves of a survey series have to be combined in order to have a sample size large enough to construct reliable age structures.

Age structures were further adjusted in countries where the age structure data source is much older than the source used for the religious composition of the country. In order to harmonize the data on overall religious affiliation with the age structure, the age structure data is aged in five-year projections while holding the religious composition constant.

In a small number of countries, age structures were estimated based in part on ethnicity or citizenship data. For example, all six Gulf Cooperation Council (GCC) countries release information on the age distribution of citizens and non-citizens, but only Bahrain further breaks this information down by Muslims and non-Muslims.²⁹ For this reason, age-distribution estimates for citizens and non-citizens in other GCC countries are modeled on Bahrain, where almost all citizens are Muslim but a substantial share of the non-citizen population is non-Muslim due to the influx of migrant workers.

Projecting Populations to 2010

For many countries, Pew Forum researchers drew on data collected in or around 2010. In some cases, however, religious affiliation data were collected in or around the years 2000 or 2005. In those cases, researchers used additional data on differential fertility, age structures and migration to project populations forward to 2010. This is the first time that formal demographic projections have been used to standardize multi-national religious population data collected in different years. This is also the first project to collect and incorporate data on the size, sex composition and religious composition of migrant flows in global population projections.

Since people can also move between religious groups within a given country, religious switching was also taken into account for some countries.

Below is a description of the projection approach, as well as the procedures used to estimate differential fertility rates, the flow of religious migrants between countries and patterns of religious switching.

The Projection Approach

The projections in this report were made by researchers in the Age and Cohort Change Project of the International Institute for Applied Systems Analysis using an advanced variation of the standard demographic method of making population projections. The standard approach is called the cohort-component method, and it takes the age and sex structure of a population into account when projecting the population forward in time. This has the advantage of recognizing that an initial baseline population can be relatively “young,” with a high proportion of people in younger age groups (such as Nigeria) or relatively “old,” with a high proportion of older people (such as Japan).

Cohorts are groups of people that had an experience in a particular time. A birth cohort, the type of cohort referenced in this context, comprises people born in a certain period. Birth cohorts can also be described as males or females (or males and females) who have reached a certain age in a particular year. For example, the cohorts of females ages 15-19 in the year 2000 and males ages 15-19 in the year 2000 shared the experience of being born between 1981 and 1985.

Components are the three ways in which populations grow or shrink: new entrants via births, exits via deaths and net changes from migration. Each cohort of the population is projected into the future by adding likely gains – births and people moving into the country (immigrants) – and subtracting likely losses – deaths and people moving out (emigrants) – year-byyear. The very youngest cohorts, those ages 0-4, are created by applying age-specific fertility rates to each female cohort in the childbearing years (ages 15-49).³⁰

The cohort-component method has been in existence for more than a century. First suggested by the English economist Edwin Cannan in 1895, then further improved by demographers in the 1930s and ’40s, it has been widely adopted since World War II. It is used by the United Nations Population Division, the U.S. Census Bureau, other national statistical offices and numerous academic and research institutions.

The advanced variant of this approach, multistate cohort component projection, was developed at IIASA by the American geographer Andrei Rogers in the 1970s. The multistate approach permits simultaneous projection of the eight religious groups in this study, taking into account variation by religion in age, sex, childbearing patterns, and propensity and direction of migration. This approach also enables modeling of religious switching as a transition between religious “states.”

The country-by-country source list in Appendix B acknowledges when the religious composition of a country has been projected. Projections have been carried out for all countries with populations greater than 100,000 for which the primary source data used for composition estimates were collected prior to 2008. For countries with populations below 100,000, and in larger countries with data collected from 2008 or afterward, the population shares estimated for the year of the primary source have been applied to the country’s population as of 2010.

Estimating Fertility

In many countries, there are substantial differences in the number of children born to women across religions. Furthermore, religious groups often vary in the share of women in their population who are of childbearing age. Differences in childbearing and age-structure patterns combine to produce differences in the rates at which babies are born to adult members across religions.³¹

Fertility data was gathered from censuses and surveys, and fertility rates were estimated via direct and indirect measures. Some censuses and surveys directly measure recent births or the count of children a woman has ever given birth to by the time of the survey. In other cases, fertility data were gathered indirectly, for example, by using data on the age of a mother’s children to estimate her past birth patterns. These various sources of fertility data were used to estimate age specific and total fertility rates for religious groups in each country.³²

In many countries, data on differential fertility is available for the largest religious groups but sufficient detail is not available for all minority religious groups. In Nigeria, for example, more than 98% of women of reproductive age are either Christian or Muslim, and there is sufficient data for estimating fertility for these two groups. For other groups in Nigeria, however, researchers had to base estimates on more limited data. Similarly, there are many countries in which one religious group makes up 95% or more of the women of reproductive age, resulting in a relatively small number of women of other religions providing fertility data.

In some countries, differential fertility data by religion were not available. In these cases, researchers applied prevailing national fertility rates to all religious groups equally.

Estimating the Size and Religious Composition of Migrant Flows

Accounting for migration in religious-group projections is important because migration can significantly change the religious landscape of countries. For example, the movement of Turkish and North African Muslims to Europe during the past decade has steadily increased the Muslim population in most Western European countries. Similarly, the movements of Hindus from India and Christians from the Philippines to Gulf Cooperation Council countries are making these Muslim-majority countries less Muslim. To accurately estimate the impact of these changes, projections require input data on migrant flows between countries by religion.

Data describing the size and religious composition of migrant flows were pieced together in two steps. The first step was to estimate the size of migrant flows for males and females to and from every country in the world since the year 2000. This was carried out using immigrant stock data from the World Bank and the Pew Forum’s Global Religion and Migration Database.³³ The plausibility of this data was checked against U.N. net migration estimates and available flow data from countries. Slight adjustments were made to flow estimates where underlying data were less reliable.

The second step was to identify the religious composition of migrants moving from one country to another. Sometimes researchers assumed migrants leaving a country had the religious composition of the country they were leaving. However, in many cases researchers had evidence to challenge this assumption. Specifically, religious minorities in a sending country may be disproportionately likely to migrate to a country in which their religion is in the majority. For example, in the Middle East, Christians are more likely than Muslims to move to the United States. The Pew Forum’s Global Religion and Migration Database captures such dynamics for many of the world’s large migrant pathways. It was therefore used to improve estimates of the religious breakdown of large migrant flows between a series of key countries.³⁴

Researchers have identified general age and sex patterns of cross-national migration. Using empirical data and research-based observations about the age and sex structures of migrant flows, researchers were able to model flows of migrants by five year age-sex cohorts and religion.

Estimating Religious Switching

In many advanced industrial countries, religious switching is producing substantial changes in the religious landscape. While there is evidence of many different types of religious switching, in many countries one of the profound shifts is a net movement toward being religiously unaffiliated. In a few of the countries that were projected to 2010 for this report (specifically, Austria, Germany and Italy), there was evidence that substantial religious switching was occurring and data were available to model switching in the population projections.

The input data for the analysis of religious switching in this report is a comparison of reports of religious origins and current religion for adults ages 30-54 in recent general population surveys. By comparing reports of respondents’ childhood and current religion, researchers estimated transition rates between all religious groups. These switching rates were calculated separately for males and females. The overall rate of religious change was then applied to cohorts ages 15-29 in population projections. For projection purposes, each person was allowed one switch, which is all that is directly measured in the surveys this method draws upon.

Religious switching results and elaboration of this approach will be described in greater detail in future Pew Forum reports.

a title="Download the Methodology as a PDF (180 KB, 17pages)" href="/uploadedFiles/Topics/Religious_Affiliation/globalReligion-appA.pdf">Download the Methodology as a PDF (180 KB, 17pages) This study provides comprehensive demographic estimates of the size and distribution of eight major religious groups in the 232 countries and territories for which the United Nations Population Division provides general population estimates as of 2010.¹⁶ It includes estimates for Christians, Muslims, the religiously unaffiliated, Hindus, Buddhists, folk religionists, adherents of other religions and Jews. This study also provides estimates for the median ages of these religious groups at the regional and global level. This appendix describes various stages of the work that led to this study. First, it describes the data and procedures used to derive estimates of the overall religious composition of each country. Second, it describes the procedures used to estimate religious composition in each country by age and sex (e.g., males between the ages 0-4, or females ages 95+). (This data was needed to help calculate the median ages of the religious groups.) Third, this appendix describes how additional data on fertility, migration and religious switching were collected. (This information was used to project the religious composition of countries to 2010 when the primary data on religious composition for a country was collected prior to 2008.) The appendix ends with a discussion about measuring religion in China, notes about estimates that vary from previously published Pew Forum reports and a list of the countries included within each of the six regions used in this report.

General Procedures for Composition Estimatesa title="Download the Methodology as a PDF (180 KB, 17pages)" href="/uploadedFiles/Topics/Religious_Affiliation/globalReligion-appA.pdf">Download the Methodology as a PDF (180 KB, 17pages) This study provides comprehensive demographic estimates of the size and distribution of eight major religious groups in the 232 countries and territories for which the United Nations Population Division provides general population estimates as of 2010.¹⁶ It includes estimates for Christians, Muslims, the religiously unaffiliated, Hindus, Buddhists, folk religionists, adherents of other religions and Jews. This study also provides estimates for the median ages of these religious groups at the regional and global level. This appendix describes various stages of the work that led to this study. First, it describes the data and procedures used to derive estimates of the overall religious composition of each country. Second, it describes the procedures used to estimate religious composition in each country by age and sex (e.g., males between the ages 0-4, or females ages 95+). (This data was needed to help calculate the median ages of the religious groups.) Third, this appendix describes how additional data on fertility, migration and religious switching were collected. (This information was used to project the religious composition of countries to 2010 when the primary data on religious composition for a country was collected prior to 2008.) The appendix ends with a discussion about measuring religion in China, notes about estimates that vary from previously published Pew Forum reports and a list of the countries included within each of the six regions used in this report.

General Procedures for Composition Estimates

Data Collection and Documentation

Pew Forum researchers acquired and analyzed religious composition information from about 2,500 data sources, including censuses, demographic surveys, general population surveys and other studies – the largest project of its kind to date. Censuses were the primary source for Pew Forum religious composition estimates in 90 countries, which together cover 45% of all people in the world.¹⁷ Large-scale demographic surveys were the primary sources for an additional 43 countries, representing 12% of the global population. General population surveys were the primary source of data for an additional 42 countries, representing 37% of the global population. Together, censuses or surveys provided estimates for 175 countries representing 95% of the world’s population. In the remaining 57 countries, representing 5% of the world’s population, the primary sources for the religious-composition estimates include population registers and institutional membership statistics reported in the World Religion Database and other sources. A list of the primary sources used to estimate the overall religious composition of each country is provided in Appendix B. Censuses and nationally representative surveys can provide valid and reliable measures of religious landscapes when they are conducted following the best practices of social science research. Valid measurement in censuses and surveys also requires that respondents are free to provide information without fear of negative governmental or social consequences. However, variation in methods among censuses and surveys (including sampling, question wording, response categories and period of data collection) can lead to variation in results. Social, cultural or political factors also may affect how answers to census and survey questions are provided and recorded.

Standardization of Religious Categories and Measurement Strategies

At least three researchers worked together to produce each country’s religious composition estimates, observing the general procedures and considerations described below. Pew Forum researchers standardized religion categories in all available censuses and surveys for each country. Censuses and surveys collect religious identity at different levels of specificity. For example, depending on the source, the most specific level of affiliation measured could be Christian, Protestant, Baptist or Southern Baptist. Researchers coded religious identities into standard categories that aggregate into the eight major religious categories used in this report. The measure of religious identity in this study is sociological rather than theological. In order to have statistics that are comparable across countries, the study attempts to count individuals who self-identify with each religion. This includes people who hold beliefs that may be viewed as unorthodox or heretical by others who claim the same religion. It also includes people who do not regularly practice the behaviors prescribed by their religion, such as those who seldom pray or attend worship services.

Identifying Primary Source(s) for Religious Composition Estimates

After considering all evidence available, researchers identified one or more primary sources for each religious composition estimate. Researchers sought a recent, reliable source – ideally, a census or large-scale demographic survey. Researchers favored sources in which religion was measured with a single question that permitted respondents to identify specific affiliations or no affiliation at all.¹⁸ Nationally representative surveys were occasionally chosen as a primary source rather than a census or demographic survey due to limitations in the measurement of religion in the latter sources. In Vietnam, for example, the 2009 census and the 2005 Demographic and Health Survey did not adequately measure folk religion identities. Researchers instead relied on the 2005 Asian Barometer survey, which measured a wider range of religious identities, including identification with folk religions.

Making Adjustments for Groups Not Adequately Measured

As necessary, researchers made adjustments to the primary source(s) to account for omitted or underrepresented groups since small minority groups are sometimes not measured or not reported in surveys and censuses. Multiple survey sources, denomination counts and estimates produced by country experts for each nation were used to assess whether minority religious groups were omitted or undercounted in the selected primary source(s). In cases where censuses and surveys lacked sufficient detail on minority groups, the estimates also drew on estimates provided by the World Religion Database, which takes into account other sources of information on religious affiliation, including statistical reports from religious groups themselves.

Adjusting for Limitations in a Survey Questionnaire

Usually, researchers assumed that members of underrepresented groups were included in the sample but were not adequately measured by the survey instrument. Adjustments frequently were made among people who responded “other religion” or failed to answer the religion question. In a few cases, the study made adjustments based on evidence that political, legal, or cultural dynamics in a country compromised the validity of self-reported religious identity. In India, for instance, there is evidence of a Christian undercount in the 2001 census; some Christians who belong to Scheduled Castes (historically referred to as Untouchables or Dalits) choose to identify as Hindu when completing official forms such as the census.¹⁹ This is due to a mandate in the Indian constitution that specifies that only Hindus, Sikhs and Buddhists can receive caste-based government affirmative-action benefits (known as reservations in India).²⁰ After analyzing Indian survey data and convening a special consultation on this topic with leading India demographers at the 2010 Asian Population Association’s meeting in New Delhi, Pew Forum researchers adjusted the Christian population of India’s 2001 population from 2.3% to 2.6%, assuming a 10% undercount in the census. In this case, the adjustment comes from the Hindu category. (Hinduism is the most common religion in India.)

Adjusting for Sampling Limitations

In some situations, underrepresented groups are likely to be omitted from the sample itself. For example, recent migrants who may not be fluent in the language used in a survey are often missing in samples. Accounting for groups not included in the sample requires proportionately deflating survey data to account for underrepresented populations. For example, researchers made adjustments to survey-based estimates in Europe where they found evidence that some survey samples and population registers underrepresented Muslim migrants. In this study, researchers sought to ensure that primary sources were representative of the entire country. When this was not the case, it was usually due to concerns about the safety of interviewers and census takers or disputes about political boundaries. In such cases, researchers attempted to make appropriate adjustments or find an alternative data source that was nationally representative. For example, the 2001 Sri Lankan census was not conducted in a handful of northern and eastern districts because of perilous conditions due to armed conflict. After analyzing religion data from earlier censuses, Pew Forum researchers determined that the areas that were not covered by the 2001 census historically had a different makeup than the rest of the country. Pew Forum researchers adjusted the 2001 census data for Sri Lanka based on 1981 census data about regions omitted in the 2001 census.²¹ Religious identity is sometimes linked to ethnic identity, particularly for religious minorities. In a small number of countries where the census did not measure religious affiliation or where survey data on religious affiliation had sampling limitations, researchers used ethnicity data to estimate the religious affiliation of small groups. For example, ethnicity data from the 2002 Russian census was used together with 2005 Generations and Gender Survey data to estimate the proportion of Muslims in Russia. The survey did not adequately sample the country’s predominantly Muslim areas but it did provide information on the share of Muslims within ethnic groups associated with Islam.²² This information, combined with census ethnicity data, was used to adjust the Muslim composition estimate in regions the survey sampled inadequately.

Making Adjustments for the Religious Affiliation of Infants

Parents are sometimes hesitant to report a religious affiliation for their infant children even though they will claim a religion for the child when he or she is slightly older. Forum researchers observed evidence of this phenomenon in some Christian-majority countries where Christian parents were disproportionately describing their infants as religiously unaffiliated. This is evident when comparing census numbers over multiple years. In Brazil, for example, the 2000 census reported that 11% of those ages 0-4 were unaffiliated. By the time of the 2010 census, only 8% of the same birth cohort (who were then 10 to 14 years old) was unaffiliated. While some of this change may be explained by mortality and migration, it is at least partly due to parents being more willing to describe their older children as Christians. In order to compensate for this measurement bias in Brazil and a few other countries where there was considerable evidence of this phenomenon, Pew Forum researchers applied the religious composition of older children (those 5-9 years old) to infants and young children (those 0-4 years old). This adjustment was made only where there was a substantial difference between the religious composition of the youngest age group and children ages 5-9. (For countries in which this adjustment was made, there is a corresponding note in the list of sources by country in Appendix B.)

Making Adjustments for Missing Religion Data

Some degree of missing data is found in most surveys and censuses. Census agencies typically make adjustments for missing data before reporting results. For example, though some respondents fail to answer questions about sex and age, census agencies follow procedures to impute missing values so it is not necessary to report “sex not stated” and “age not stated” as sex and age categories. Some census agencies, such as Statistics Canada, have historically imputed religion values for respondents who have not answered the census religion question. The likelihood that religion data will be missing increases when religion questions are labeled as optional, as is the case in censuses in countries such as Australia, the United Kingdom and the Czech Republic.²³ Census agencies in countries where religion is labeled as an optional question often report “religion not stated” results alongside standard categories of affiliation and non- affiliation. This strategy allows the census agencies to demonstrate that answering their religion question was indeed optional. However, for purposes of this study, the “religion not stated” category is not a meaningful religious identity. Therefore, after making any necessary adjustments for undercounted groups, religious shares were re-calculated based on the population of all people who gave valid responses to the census or survey. The effect of this approach was to proportionately raise the shares of all religious groups, including the religiously unaffiliated. Following the procedures described above, researchers produced national-level estimates of the religious composition of each country for the year measured by the primary source. In order to report 2010 population figures, the religious composition percentages based on data collected in 2008 or later have been multiplied by the U.N.’s 2010 population estimate for each country and territory.²⁴ Estimates based on data collected prior to 2008 have been projected forward to 2010. These projections required additional information, including the religious composition of age and sex groups within each country. Researchers from the International Institute for Applied Systems Analysis (IIASA) in Laxenburg, Austria, collaborated with Pew Forum researchers on data collection and religious composition estimates for Europe; Pew Forum researchers gathered data outside Europe. IIASA researchers also collaborated on development of the procedures described below, which were used to standardize information across countries for age breakdowns, childbearing patterns, migration flows and religious switching patterns. Researchers from the Vienna Institute of Demography provided initial estimates of country-to-country migrant flows by age and sex.

Age Structure Proceduresa title="Download the Methodology as a PDF (180 KB, 17pages)" href="/uploadedFiles/Topics/Religious_Affiliation/globalReligion-appA.pdf">Download the Methodology as a PDF (180 KB, 17pages) This study provides comprehensive demographic estimates of the size and distribution of eight major religious groups in the 232 countries and territories for which the United Nations Population Division provides general population estimates as of 2010.¹⁶ It includes estimates for Christians, Muslims, the religiously unaffiliated, Hindus, Buddhists, folk religionists, adherents of other religions and Jews. This study also provides estimates for the median ages of these religious groups at the regional and global level. This appendix describes various stages of the work that led to this study. First, it describes the data and procedures used to derive estimates of the overall religious composition of each country. Second, it describes the procedures used to estimate religious composition in each country by age and sex (e.g., males between the ages 0-4, or females ages 95+). (This data was needed to help calculate the median ages of the religious groups.) Third, this appendix describes how additional data on fertility, migration and religious switching were collected. (This information was used to project the religious composition of countries to 2010 when the primary data on religious composition for a country was collected prior to 2008.) The appendix ends with a discussion about measuring religion in China, notes about estimates that vary from previously published Pew Forum reports and a list of the countries included within each of the six regions used in this report.

General Procedures for Composition Estimates

Data Collection and Documentation

Pew Forum researchers acquired and analyzed religious composition information from about 2,500 data sources, including censuses, demographic surveys, general population surveys and other studies – the largest project of its kind to date. Censuses were the primary source for Pew Forum religious composition estimates in 90 countries, which together cover 45% of all people in the world.¹⁷ Large-scale demographic surveys were the primary sources for an additional 43 countries, representing 12% of the global population. General population surveys were the primary source of data for an additional 42 countries, representing 37% of the global population. Together, censuses or surveys provided estimates for 175 countries representing 95% of the world’s population. In the remaining 57 countries, representing 5% of the world’s population, the primary sources for the religious-composition estimates include population registers and institutional membership statistics reported in the World Religion Database and other sources. A list of the primary sources used to estimate the overall religious composition of each country is provided in Appendix B. Censuses and nationally representative surveys can provide valid and reliable measures of religious landscapes when they are conducted following the best practices of social science research. Valid measurement in censuses and surveys also requires that respondents are free to provide information without fear of negative governmental or social consequences. However, variation in methods among censuses and surveys (including sampling, question wording, response categories and period of data collection) can lead to variation in results. Social, cultural or political factors also may affect how answers to census and survey questions are provided and recorded.

Standardization of Religious Categories and Measurement Strategies

At least three researchers worked together to produce each country’s religious composition estimates, observing the general procedures and considerations described below. Pew Forum researchers standardized religion categories in all available censuses and surveys for each country. Censuses and surveys collect religious identity at different levels of specificity. For example, depending on the source, the most specific level of affiliation measured could be Christian, Protestant, Baptist or Southern Baptist. Researchers coded religious identities into standard categories that aggregate into the eight major religious categories used in this report. The measure of religious identity in this study is sociological rather than theological. In order to have statistics that are comparable across countries, the study attempts to count individuals who self-identify with each religion. This includes people who hold beliefs that may be viewed as unorthodox or heretical by others who claim the same religion. It also includes people who do not regularly practice the behaviors prescribed by their religion, such as those who seldom pray or attend worship services.

Identifying Primary Source(s) for Religious Composition Estimates

After considering all evidence available, researchers identified one or more primary sources for each religious composition estimate. Researchers sought a recent, reliable source – ideally, a census or large-scale demographic survey. Researchers favored sources in which religion was measured with a single question that permitted respondents to identify specific affiliations or no affiliation at all.¹⁸ Nationally representative surveys were occasionally chosen as a primary source rather than a census or demographic survey due to limitations in the measurement of religion in the latter sources. In Vietnam, for example, the 2009 census and the 2005 Demographic and Health Survey did not adequately measure folk religion identities. Researchers instead relied on the 2005 Asian Barometer survey, which measured a wider range of religious identities, including identification with folk religions.

Making Adjustments for Groups Not Adequately Measured

As necessary, researchers made adjustments to the primary source(s) to account for omitted or underrepresented groups since small minority groups are sometimes not measured or not reported in surveys and censuses. Multiple survey sources, denomination counts and estimates produced by country experts for each nation were used to assess whether minority religious groups were omitted or undercounted in the selected primary source(s). In cases where censuses and surveys lacked sufficient detail on minority groups, the estimates also drew on estimates provided by the World Religion Database, which takes into account other sources of information on religious affiliation, including statistical reports from religious groups themselves.

Adjusting for Limitations in a Survey Questionnaire

Usually, researchers assumed that members of underrepresented groups were included in the sample but were not adequately measured by the survey instrument. Adjustments frequently were made among people who responded “other religion” or failed to answer the religion question. In a few cases, the study made adjustments based on evidence that political, legal, or cultural dynamics in a country compromised the validity of self-reported religious identity. In India, for instance, there is evidence of a Christian undercount in the 2001 census; some Christians who belong to Scheduled Castes (historically referred to as Untouchables or Dalits) choose to identify as Hindu when completing official forms such as the census.¹⁹ This is due to a mandate in the Indian constitution that specifies that only Hindus, Sikhs and Buddhists can receive caste-based government affirmative-action benefits (known as reservations in India).²⁰ After analyzing Indian survey data and convening a special consultation on this topic with leading India demographers at the 2010 Asian Population Association’s meeting in New Delhi, Pew Forum researchers adjusted the Christian population of India’s 2001 population from 2.3% to 2.6%, assuming a 10% undercount in the census. In this case, the adjustment comes from the Hindu category. (Hinduism is the most common religion in India.)

Adjusting for Sampling Limitations

In some situations, underrepresented groups are likely to be omitted from the sample itself. For example, recent migrants who may not be fluent in the language used in a survey are often missing in samples. Accounting for groups not included in the sample requires proportionately deflating survey data to account for underrepresented populations. For example, researchers made adjustments to survey-based estimates in Europe where they found evidence that some survey samples and population registers underrepresented Muslim migrants. In this study, researchers sought to ensure that primary sources were representative of the entire country. When this was not the case, it was usually due to concerns about the safety of interviewers and census takers or disputes about political boundaries. In such cases, researchers attempted to make appropriate adjustments or find an alternative data source that was nationally representative. For example, the 2001 Sri Lankan census was not conducted in a handful of northern and eastern districts because of perilous conditions due to armed conflict. After analyzing religion data from earlier censuses, Pew Forum researchers determined that the areas that were not covered by the 2001 census historically had a different makeup than the rest of the country. Pew Forum researchers adjusted the 2001 census data for Sri Lanka based on 1981 census data about regions omitted in the 2001 census.²¹ Religious identity is sometimes linked to ethnic identity, particularly for religious minorities. In a small number of countries where the census did not measure religious affiliation or where survey data on religious affiliation had sampling limitations, researchers used ethnicity data to estimate the religious affiliation of small groups. For example, ethnicity data from the 2002 Russian census was used together with 2005 Generations and Gender Survey data to estimate the proportion of Muslims in Russia. The survey did not adequately sample the country’s predominantly Muslim areas but it did provide information on the share of Muslims within ethnic groups associated with Islam.²² This information, combined with census ethnicity data, was used to adjust the Muslim composition estimate in regions the survey sampled inadequately.

Making Adjustments for the Religious Affiliation of Infants

Parents are sometimes hesitant to report a religious affiliation for their infant children even though they will claim a religion for the child when he or she is slightly older. Forum researchers observed evidence of this phenomenon in some Christian-majority countries where Christian parents were disproportionately describing their infants as religiously unaffiliated. This is evident when comparing census numbers over multiple years. In Brazil, for example, the 2000 census reported that 11% of those ages 0-4 were unaffiliated. By the time of the 2010 census, only 8% of the same birth cohort (who were then 10 to 14 years old) was unaffiliated. While some of this change may be explained by mortality and migration, it is at least partly due to parents being more willing to describe their older children as Christians. In order to compensate for this measurement bias in Brazil and a few other countries where there was considerable evidence of this phenomenon, Pew Forum researchers applied the religious composition of older children (those 5-9 years old) to infants and young children (those 0-4 years old). This adjustment was made only where there was a substantial difference between the religious composition of the youngest age group and children ages 5-9. (For countries in which this adjustment was made, there is a corresponding note in the list of sources by country in Appendix B.)

Making Adjustments for Missing Religion Data

Some degree of missing data is found in most surveys and censuses. Census agencies typically make adjustments for missing data before reporting results. For example, though some respondents fail to answer questions about sex and age, census agencies follow procedures to impute missing values so it is not necessary to report “sex not stated” and “age not stated” as sex and age categories. Some census agencies, such as Statistics Canada, have historically imputed religion values for respondents who have not answered the census religion question. The likelihood that religion data will be missing increases when religion questions are labeled as optional, as is the case in censuses in countries such as Australia, the United Kingdom and the Czech Republic.²³ Census agencies in countries where religion is labeled as an optional question often report “religion not stated” results alongside standard categories of affiliation and non- affiliation. This strategy allows the census agencies to demonstrate that answering their religion question was indeed optional. However, for purposes of this study, the “religion not stated” category is not a meaningful religious identity. Therefore, after making any necessary adjustments for undercounted groups, religious shares were re-calculated based on the population of all people who gave valid responses to the census or survey. The effect of this approach was to proportionately raise the shares of all religious groups, including the religiously unaffiliated. Following the procedures described above, researchers produced national-level estimates of the religious composition of each country for the year measured by the primary source. In order to report 2010 population figures, the religious composition percentages based on data collected in 2008 or later have been multiplied by the U.N.’s 2010 population estimate for each country and territory.²⁴ Estimates based on data collected prior to 2008 have been projected forward to 2010. These projections required additional information, including the religious composition of age and sex groups within each country. Researchers from the International Institute for Applied Systems Analysis (IIASA) in Laxenburg, Austria, collaborated with Pew Forum researchers on data collection and religious composition estimates for Europe; Pew Forum researchers gathered data outside Europe. IIASA researchers also collaborated on development of the procedures described below, which were used to standardize information across countries for age breakdowns, childbearing patterns, migration flows and religious switching patterns. Researchers from the Vienna Institute of Demography provided initial estimates of country-to-country migrant flows by age and sex.

Age Structure Procedures

Religious affiliation varies by age.²⁵ In this section, the phrase “age structure” is used as shorthand to refer to the religious composition of age-sex groups. In order to calculate the median ages of religious groups and carry out population projections, Pew Forum researchers assembled age structures for each of the eight religious groups in every country. Data on age structures were collected in 20 age categories (measured in five-year increments with a top value of 95 and above) for males and females (e.g. males between the ages of 15 and 19), resulting in a total of 40 categories. Age structures were compiled in three steps. First, census or survey data were used to capture the religious affiliation of each available age group. Second, survey data on religion by age were adjusted to account for small sample sizes. And third, results were adjusted to match the religious breakdown by age and sex to each country’s overall religious composition. These steps are described in detail below.

Estimating Religion by Age and Sex

Pew Forum researchers constructed initial age structures by analyzing survey datasets, census datasets and tables published by census agencies. While censuses usually enumerate religion for the entire population, including children, general population surveys do not usually include interviews with children. Since age structures require religious affiliation data for children, children were assigned religious affiliations when necessary based on the best methods available. For datasets that measured religious affiliation only for adult respondents, yet included the number and ages of children (and other adults) in the household, researchers were able to estimate the religious affiliation of remaining household members. In most cases, the religious affiliation of the respondent or head of the household was assigned to all additional members of the household who were not surveyed.²⁶ For surveys that did not offer household information, such as the AmericasBarometer or the European Values Study, children were assigned a religious affiliation based on the fertility patterns and religious affiliation of women of childbearing age, as well as information about the religious affiliation of the youngest respondents measured in the survey. For many countries, reliable age data were not available for all eight religious groups. Sometimes a survey indicated the overall size of a small religious group yet lacked sufficient numbers to reliably estimate the group’s age breakdown. In such cases, the age breakdowns of minority religious groups were based on the country’s overall age distribution or the combined age distribution of respondents from all minority religions in a survey.

Adjustments to Minimize Errors Due to Sample Size

The reliability of survey estimates is partly dependent on the number of people surveyed (the sample size). Since respondents who identify with religious groups are divided into 40 age and sex categories, the number of Muslims, for example, in any one age-sex category may be small and produce less reliable estimates than a larger count would produce. This introduces significant variation in patterns of religious affiliation by age: affiliation levels may bounce between highs and lows for consecutive age groups. To eliminate unlikely variation, researchers smoothed data using statistical procedures intended to reveal the general underlying pattern.²⁷ Census data are not smoothed because census data represent a complete enumeration of all individuals living in a country. However, in some cases, the age categories reported by census agencies are in 10-year age groups or aggregated for all adults above a certain age, such as 60. Researchers used statistical modeling techniques to distribute the composition of these aggregated age groups across the more specific five-year age categories used in this study.

Matching Religion by Age and Sex to Overall Population by Age and Sex

The overall religious affiliation resulting from the age structure procedures sometimes varies from the religious composition estimated for the country using the procedures described in the first section of this appendix. This difference exists for two reasons. First, the data sources used for the age-structure procedure may be different from the data sources used for the religious composition. Second, overall religious compositions were adjusted manually to account for undercounts and sampling issues. In order to match the overall religious composition figures to the data on religion by age and sex, the age structure was adjusted. The adjustment procedure used is often referred to as Iterative Proportional Fitting (IPF), or raking. Raking makes adjustments to the percentages of religious affiliation for each age group without significantly altering the underlying religious affiliation patterns by age group. Additionally, raking is used to match each country’s counts of males and females in particular age groups to the U.N.’s estimate of the country’s overall age distribution. When survey or census data on the differential religious composition of age-sex groups was not available, each age-sex group was assigned the same religious composition. Lack of differential religious composition data by age-sex group was most common in countries with very small populations.

Sources for Age Structure Data

As mentioned above, the data source used for the country’s age structure is sometimes different from the source used for the overall religious composition (see Appendix B for a list of religious-composition sources by country).²⁸ This is the case, for example, when census data with overall religious composition results are available but a detailed breakdown by age and sex is not released by the census bureau; another source must therefore be used to generate the age structure. Sources are also different when multiple waves of a survey series have to be combined in order to have a sample size large enough to construct reliable age structures. Age structures were further adjusted in countries where the age structure data source is much older than the source used for the religious composition of the country. In order to harmonize the data on overall religious affiliation with the age structure, the age structure data is aged in five-year projections while holding the religious composition constant. In a small number of countries, age structures were estimated based in part on ethnicity or citizenship data. For example, all six Gulf Cooperation Council (GCC) countries release information on the age distribution of citizens and non-citizens, but only Bahrain further breaks this information down by Muslims and non-Muslims.²⁹ For this reason, age-distribution estimates for citizens and non-citizens in other GCC countries are modeled on Bahrain, where almost all citizens are Muslim but a substantial share of the non-citizen population is non-Muslim due to the influx of migrant workers.

Projecting Populations to 2010a title="Download the Methodology as a PDF (180 KB, 17pages)" href="/uploadedFiles/Topics/Religious_Affiliation/globalReligion-appA.pdf">Download the Methodology as a PDF (180 KB, 17pages) This study provides comprehensive demographic estimates of the size and distribution of eight major religious groups in the 232 countries and territories for which the United Nations Population Division provides general population estimates as of 2010.¹⁶ It includes estimates for Christians, Muslims, the religiously unaffiliated, Hindus, Buddhists, folk religionists, adherents of other religions and Jews. This study also provides estimates for the median ages of these religious groups at the regional and global level. This appendix describes various stages of the work that led to this study. First, it describes the data and procedures used to derive estimates of the overall religious composition of each country. Second, it describes the procedures used to estimate religious composition in each country by age and sex (e.g., males between the ages 0-4, or females ages 95+). (This data was needed to help calculate the median ages of the religious groups.) Third, this appendix describes how additional data on fertility, migration and religious switching were collected. (This information was used to project the religious composition of countries to 2010 when the primary data on religious composition for a country was collected prior to 2008.) The appendix ends with a discussion about measuring religion in China, notes about estimates that vary from previously published Pew Forum reports and a list of the countries included within each of the six regions used in this report.

General Procedures for Composition Estimates

Data Collection and Documentation

Pew Forum researchers acquired and analyzed religious composition information from about 2,500 data sources, including censuses, demographic surveys, general population surveys and other studies – the largest project of its kind to date. Censuses were the primary source for Pew Forum religious composition estimates in 90 countries, which together cover 45% of all people in the world.¹⁷ Large-scale demographic surveys were the primary sources for an additional 43 countries, representing 12% of the global population. General population surveys were the primary source of data for an additional 42 countries, representing 37% of the global population. Together, censuses or surveys provided estimates for 175 countries representing 95% of the world’s population. In the remaining 57 countries, representing 5% of the world’s population, the primary sources for the religious-composition estimates include population registers and institutional membership statistics reported in the World Religion Database and other sources. A list of the primary sources used to estimate the overall religious composition of each country is provided in Appendix B. Censuses and nationally representative surveys can provide valid and reliable measures of religious landscapes when they are conducted following the best practices of social science research. Valid measurement in censuses and surveys also requires that respondents are free to provide information without fear of negative governmental or social consequences. However, variation in methods among censuses and surveys (including sampling, question wording, response categories and period of data collection) can lead to variation in results. Social, cultural or political factors also may affect how answers to census and survey questions are provided and recorded.

Standardization of Religious Categories and Measurement Strategies

At least three researchers worked together to produce each country’s religious composition estimates, observing the general procedures and considerations described below. Pew Forum researchers standardized religion categories in all available censuses and surveys for each country. Censuses and surveys collect religious identity at different levels of specificity. For example, depending on the source, the most specific level of affiliation measured could be Christian, Protestant, Baptist or Southern Baptist. Researchers coded religious identities into standard categories that aggregate into the eight major religious categories used in this report. The measure of religious identity in this study is sociological rather than theological. In order to have statistics that are comparable across countries, the study attempts to count individuals who self-identify with each religion. This includes people who hold beliefs that may be viewed as unorthodox or heretical by others who claim the same religion. It also includes people who do not regularly practice the behaviors prescribed by their religion, such as those who seldom pray or attend worship services.

Identifying Primary Source(s) for Religious Composition Estimates

After considering all evidence available, researchers identified one or more primary sources for each religious composition estimate. Researchers sought a recent, reliable source – ideally, a census or large-scale demographic survey. Researchers favored sources in which religion was measured with a single question that permitted respondents to identify specific affiliations or no affiliation at all.¹⁸ Nationally representative surveys were occasionally chosen as a primary source rather than a census or demographic survey due to limitations in the measurement of religion in the latter sources. In Vietnam, for example, the 2009 census and the 2005 Demographic and Health Survey did not adequately measure folk religion identities. Researchers instead relied on the 2005 Asian Barometer survey, which measured a wider range of religious identities, including identification with folk religions.

Making Adjustments for Groups Not Adequately Measured

As necessary, researchers made adjustments to the primary source(s) to account for omitted or underrepresented groups since small minority groups are sometimes not measured or not reported in surveys and censuses. Multiple survey sources, denomination counts and estimates produced by country experts for each nation were used to assess whether minority religious groups were omitted or undercounted in the selected primary source(s). In cases where censuses and surveys lacked sufficient detail on minority groups, the estimates also drew on estimates provided by the World Religion Database, which takes into account other sources of information on religious affiliation, including statistical reports from religious groups themselves.

Adjusting for Limitations in a Survey Questionnaire

Usually, researchers assumed that members of underrepresented groups were included in the sample but were not adequately measured by the survey instrument. Adjustments frequently were made among people who responded “other religion” or failed to answer the religion question. In a few cases, the study made adjustments based on evidence that political, legal, or cultural dynamics in a country compromised the validity of self-reported religious identity. In India, for instance, there is evidence of a Christian undercount in the 2001 census; some Christians who belong to Scheduled Castes (historically referred to as Untouchables or Dalits) choose to identify as Hindu when completing official forms such as the census.¹⁹ This is due to a mandate in the Indian constitution that specifies that only Hindus, Sikhs and Buddhists can receive caste-based government affirmative-action benefits (known as reservations in India).²⁰ After analyzing Indian survey data and convening a special consultation on this topic with leading India demographers at the 2010 Asian Population Association’s meeting in New Delhi, Pew Forum researchers adjusted the Christian population of India’s 2001 population from 2.3% to 2.6%, assuming a 10% undercount in the census. In this case, the adjustment comes from the Hindu category. (Hinduism is the most common religion in India.)

Adjusting for Sampling Limitations

In some situations, underrepresented groups are likely to be omitted from the sample itself. For example, recent migrants who may not be fluent in the language used in a survey are often missing in samples. Accounting for groups not included in the sample requires proportionately deflating survey data to account for underrepresented populations. For example, researchers made adjustments to survey-based estimates in Europe where they found evidence that some survey samples and population registers underrepresented Muslim migrants. In this study, researchers sought to ensure that primary sources were representative of the entire country. When this was not the case, it was usually due to concerns about the safety of interviewers and census takers or disputes about political boundaries. In such cases, researchers attempted to make appropriate adjustments or find an alternative data source that was nationally representative. For example, the 2001 Sri Lankan census was not conducted in a handful of northern and eastern districts because of perilous conditions due to armed conflict. After analyzing religion data from earlier censuses, Pew Forum researchers determined that the areas that were not covered by the 2001 census historically had a different makeup than the rest of the country. Pew Forum researchers adjusted the 2001 census data for Sri Lanka based on 1981 census data about regions omitted in the 2001 census.²¹ Religious identity is sometimes linked to ethnic identity, particularly for religious minorities. In a small number of countries where the census did not measure religious affiliation or where survey data on religious affiliation had sampling limitations, researchers used ethnicity data to estimate the religious affiliation of small groups. For example, ethnicity data from the 2002 Russian census was used together with 2005 Generations and Gender Survey data to estimate the proportion of Muslims in Russia. The survey did not adequately sample the country’s predominantly Muslim areas but it did provide information on the share of Muslims within ethnic groups associated with Islam.²² This information, combined with census ethnicity data, was used to adjust the Muslim composition estimate in regions the survey sampled inadequately.

Making Adjustments for the Religious Affiliation of Infants

Parents are sometimes hesitant to report a religious affiliation for their infant children even though they will claim a religion for the child when he or she is slightly older. Forum researchers observed evidence of this phenomenon in some Christian-majority countries where Christian parents were disproportionately describing their infants as religiously unaffiliated. This is evident when comparing census numbers over multiple years. In Brazil, for example, the 2000 census reported that 11% of those ages 0-4 were unaffiliated. By the time of the 2010 census, only 8% of the same birth cohort (who were then 10 to 14 years old) was unaffiliated. While some of this change may be explained by mortality and migration, it is at least partly due to parents being more willing to describe their older children as Christians. In order to compensate for this measurement bias in Brazil and a few other countries where there was considerable evidence of this phenomenon, Pew Forum researchers applied the religious composition of older children (those 5-9 years old) to infants and young children (those 0-4 years old). This adjustment was made only where there was a substantial difference between the religious composition of the youngest age group and children ages 5-9. (For countries in which this adjustment was made, there is a corresponding note in the list of sources by country in Appendix B.)

Making Adjustments for Missing Religion Data

Some degree of missing data is found in most surveys and censuses. Census agencies typically make adjustments for missing data before reporting results. For example, though some respondents fail to answer questions about sex and age, census agencies follow procedures to impute missing values so it is not necessary to report “sex not stated” and “age not stated” as sex and age categories. Some census agencies, such as Statistics Canada, have historically imputed religion values for respondents who have not answered the census religion question. The likelihood that religion data will be missing increases when religion questions are labeled as optional, as is the case in censuses in countries such as Australia, the United Kingdom and the Czech Republic.²³ Census agencies in countries where religion is labeled as an optional question often report “religion not stated” results alongside standard categories of affiliation and non- affiliation. This strategy allows the census agencies to demonstrate that answering their religion question was indeed optional. However, for purposes of this study, the “religion not stated” category is not a meaningful religious identity. Therefore, after making any necessary adjustments for undercounted groups, religious shares were re-calculated based on the population of all people who gave valid responses to the census or survey. The effect of this approach was to proportionately raise the shares of all religious groups, including the religiously unaffiliated. Following the procedures described above, researchers produced national-level estimates of the religious composition of each country for the year measured by the primary source. In order to report 2010 population figures, the religious composition percentages based on data collected in 2008 or later have been multiplied by the U.N.’s 2010 population estimate for each country and territory.²⁴ Estimates based on data collected prior to 2008 have been projected forward to 2010. These projections required additional information, including the religious composition of age and sex groups within each country. Researchers from the International Institute for Applied Systems Analysis (IIASA) in Laxenburg, Austria, collaborated with Pew Forum researchers on data collection and religious composition estimates for Europe; Pew Forum researchers gathered data outside Europe. IIASA researchers also collaborated on development of the procedures described below, which were used to standardize information across countries for age breakdowns, childbearing patterns, migration flows and religious switching patterns. Researchers from the Vienna Institute of Demography provided initial estimates of country-to-country migrant flows by age and sex.

Age Structure Procedures

Religious affiliation varies by age.²⁵ In this section, the phrase “age structure” is used as shorthand to refer to the religious composition of age-sex groups. In order to calculate the median ages of religious groups and carry out population projections, Pew Forum researchers assembled age structures for each of the eight religious groups in every country. Data on age structures were collected in 20 age categories (measured in five-year increments with a top value of 95 and above) for males and females (e.g. males between the ages of 15 and 19), resulting in a total of 40 categories. Age structures were compiled in three steps. First, census or survey data were used to capture the religious affiliation of each available age group. Second, survey data on religion by age were adjusted to account for small sample sizes. And third, results were adjusted to match the religious breakdown by age and sex to each country’s overall religious composition. These steps are described in detail below.

Estimating Religion by Age and Sex

Pew Forum researchers constructed initial age structures by analyzing survey datasets, census datasets and tables published by census agencies. While censuses usually enumerate religion for the entire population, including children, general population surveys do not usually include interviews with children. Since age structures require religious affiliation data for children, children were assigned religious affiliations when necessary based on the best methods available. For datasets that measured religious affiliation only for adult respondents, yet included the number and ages of children (and other adults) in the household, researchers were able to estimate the religious affiliation of remaining household members. In most cases, the religious affiliation of the respondent or head of the household was assigned to all additional members of the household who were not surveyed.²⁶ For surveys that did not offer household information, such as the AmericasBarometer or the European Values Study, children were assigned a religious affiliation based on the fertility patterns and religious affiliation of women of childbearing age, as well as information about the religious affiliation of the youngest respondents measured in the survey. For many countries, reliable age data were not available for all eight religious groups. Sometimes a survey indicated the overall size of a small religious group yet lacked sufficient numbers to reliably estimate the group’s age breakdown. In such cases, the age breakdowns of minority religious groups were based on the country’s overall age distribution or the combined age distribution of respondents from all minority religions in a survey.

Adjustments to Minimize Errors Due to Sample Size

The reliability of survey estimates is partly dependent on the number of people surveyed (the sample size). Since respondents who identify with religious groups are divided into 40 age and sex categories, the number of Muslims, for example, in any one age-sex category may be small and produce less reliable estimates than a larger count would produce. This introduces significant variation in patterns of religious affiliation by age: affiliation levels may bounce between highs and lows for consecutive age groups. To eliminate unlikely variation, researchers smoothed data using statistical procedures intended to reveal the general underlying pattern.²⁷ Census data are not smoothed because census data represent a complete enumeration of all individuals living in a country. However, in some cases, the age categories reported by census agencies are in 10-year age groups or aggregated for all adults above a certain age, such as 60. Researchers used statistical modeling techniques to distribute the composition of these aggregated age groups across the more specific five-year age categories used in this study.

Matching Religion by Age and Sex to Overall Population by Age and Sex

The overall religious affiliation resulting from the age structure procedures sometimes varies from the religious composition estimated for the country using the procedures described in the first section of this appendix. This difference exists for two reasons. First, the data sources used for the age-structure procedure may be different from the data sources used for the religious composition. Second, overall religious compositions were adjusted manually to account for undercounts and sampling issues. In order to match the overall religious composition figures to the data on religion by age and sex, the age structure was adjusted. The adjustment procedure used is often referred to as Iterative Proportional Fitting (IPF), or raking. Raking makes adjustments to the percentages of religious affiliation for each age group without significantly altering the underlying religious affiliation patterns by age group. Additionally, raking is used to match each country’s counts of males and females in particular age groups to the U.N.’s estimate of the country’s overall age distribution. When survey or census data on the differential religious composition of age-sex groups was not available, each age-sex group was assigned the same religious composition. Lack of differential religious composition data by age-sex group was most common in countries with very small populations.

Sources for Age Structure Data

As mentioned above, the data source used for the country’s age structure is sometimes different from the source used for the overall religious composition (see Appendix B for a list of religious-composition sources by country).²⁸ This is the case, for example, when census data with overall religious composition results are available but a detailed breakdown by age and sex is not released by the census bureau; another source must therefore be used to generate the age structure. Sources are also different when multiple waves of a survey series have to be combined in order to have a sample size large enough to construct reliable age structures. Age structures were further adjusted in countries where the age structure data source is much older than the source used for the religious composition of the country. In order to harmonize the data on overall religious affiliation with the age structure, the age structure data is aged in five-year projections while holding the religious composition constant. In a small number of countries, age structures were estimated based in part on ethnicity or citizenship data. For example, all six Gulf Cooperation Council (GCC) countries release information on the age distribution of citizens and non-citizens, but only Bahrain further breaks this information down by Muslims and non-Muslims.²⁹ For this reason, age-distribution estimates for citizens and non-citizens in other GCC countries are modeled on Bahrain, where almost all citizens are Muslim but a substantial share of the non-citizen population is non-Muslim due to the influx of migrant workers.

Projecting Populations to 2010

For many countries, Pew Forum researchers drew on data collected in or around 2010. In some cases, however, religious affiliation data were collected in or around the years 2000 or 2005. In those cases, researchers used additional data on differential fertility, age structures and migration to project populations forward to 2010. This is the first time that formal demographic projections have been used to standardize multi-national religious population data collected in different years. This is also the first project to collect and incorporate data on the size, sex composition and religious composition of migrant flows in global population projections. Since people can also move between religious groups within a given country, religious switching was also taken into account for some countries. Below is a description of the projection approach, as well as the procedures used to estimate differential fertility rates, the flow of religious migrants between countries and patterns of religious switching.

The Projection Approach

The projections in this report were made by researchers in the Age and Cohort Change Project of the International Institute for Applied Systems Analysis using an advanced variation of the standard demographic method of making population projections. The standard approach is called the cohort-component method, and it takes the age and sex structure of a population into account when projecting the population forward in time. This has the advantage of recognizing that an initial baseline population can be relatively “young,” with a high proportion of people in younger age groups (such as Nigeria) or relatively “old,” with a high proportion of older people (such as Japan). Cohorts are groups of people that had an experience in a particular time. A birth cohort, the type of cohort referenced in this context, comprises people born in a certain period. Birth cohorts can also be described as males or females (or males and females) who have reached a certain age in a particular year. For example, the cohorts of females ages 15-19 in the year 2000 and males ages 15-19 in the year 2000 shared the experience of being born between 1981 and 1985. Components are the three ways in which populations grow or shrink: new entrants via births, exits via deaths and net changes from migration. Each cohort of the population is projected into the future by adding likely gains – births and people moving into the country (immigrants) – and subtracting likely losses – deaths and people moving out (emigrants) – year-byyear. The very youngest cohorts, those ages 0-4, are created by applying age-specific fertility rates to each female cohort in the childbearing years (ages 15-49).³⁰ The cohort-component method has been in existence for more than a century. First suggested by the English economist Edwin Cannan in 1895, then further improved by demographers in the 1930s and ’40s, it has been widely adopted since World War II. It is used by the United Nations Population Division, the U.S. Census Bureau, other national statistical offices and numerous academic and research institutions. The advanced variant of this approach, multistate cohort component projection, was developed at IIASA by the American geographer Andrei Rogers in the 1970s. The multistate approach permits simultaneous projection of the eight religious groups in this study, taking into account variation by religion in age, sex, childbearing patterns, and propensity and direction of migration. This approach also enables modeling of religious switching as a transition between religious “states.” The country-by-country source list in Appendix B acknowledges when the religious composition of a country has been projected. Projections have been carried out for all countries with populations greater than 100,000 for which the primary source data used for composition estimates were collected prior to 2008. For countries with populations below 100,000, and in larger countries with data collected from 2008 or afterward, the population shares estimated for the year of the primary source have been applied to the country’s population as of 2010.

Estimating Fertility

In many countries, there are substantial differences in the number of children born to women across religions. Furthermore, religious groups often vary in the share of women in their population who are of childbearing age. Differences in childbearing and age-structure patterns combine to produce differences in the rates at which babies are born to adult members across religions.³¹ Fertility data was gathered from censuses and surveys, and fertility rates were estimated via direct and indirect measures. Some censuses and surveys directly measure recent births or the count of children a woman has ever given birth to by the time of the survey. In other cases, fertility data were gathered indirectly, for example, by using data on the age of a mother’s children to estimate her past birth patterns. These various sources of fertility data were used to estimate age specific and total fertility rates for religious groups in each country.³² In many countries, data on differential fertility is available for the largest religious groups but sufficient detail is not available for all minority religious groups. In Nigeria, for example, more than 98% of women of reproductive age are either Christian or Muslim, and there is sufficient data for estimating fertility for these two groups. For other groups in Nigeria, however, researchers had to base estimates on more limited data. Similarly, there are many countries in which one religious group makes up 95% or more of the women of reproductive age, resulting in a relatively small number of women of other religions providing fertility data. In some countries, differential fertility data by religion were not available. In these cases, researchers applied prevailing national fertility rates to all religious groups equally.

Estimating the Size and Religious Composition of Migrant Flows

Accounting for migration in religious-group projections is important because migration can significantly change the religious landscape of countries. For example, the movement of Turkish and North African Muslims to Europe during the past decade has steadily increased the Muslim population in most Western European countries. Similarly, the movements of Hindus from India and Christians from the Philippines to Gulf Cooperation Council countries are making these Muslim-majority countries less Muslim. To accurately estimate the impact of these changes, projections require input data on migrant flows between countries by religion. Data describing the size and religious composition of migrant flows were pieced together in two steps. The first step was to estimate the size of migrant flows for males and females to and from every country in the world since the year 2000. This was carried out using immigrant stock data from the World Bank and the Pew Forum’s Global Religion and Migration Database.³³ The plausibility of this data was checked against U.N. net migration estimates and available flow data from countries. Slight adjustments were made to flow estimates where underlying data were less reliable. The second step was to identify the religious composition of migrants moving from one country to another. Sometimes researchers assumed migrants leaving a country had the religious composition of the country they were leaving. However, in many cases researchers had evidence to challenge this assumption. Specifically, religious minorities in a sending country may be disproportionately likely to migrate to a country in which their religion is in the majority. For example, in the Middle East, Christians are more likely than Muslims to move to the United States. The Pew Forum’s Global Religion and Migration Database captures such dynamics for many of the world’s large migrant pathways. It was therefore used to improve estimates of the religious breakdown of large migrant flows between a series of key countries.³⁴ Researchers have identified general age and sex patterns of cross-national migration. Using empirical data and research-based observations about the age and sex structures of migrant flows, researchers were able to model flows of migrants by five year age-sex cohorts and religion.

Estimating Religious Switching

In many advanced industrial countries, religious switching is producing substantial changes in the religious landscape. While there is evidence of many different types of religious switching, in many countries one of the profound shifts is a net movement toward being religiously unaffiliated. In a few of the countries that were projected to 2010 for this report (specifically, Austria, Germany and Italy), there was evidence that substantial religious switching was occurring and data were available to model switching in the population projections. The input data for the analysis of religious switching in this report is a comparison of reports of religious origins and current religion for adults ages 30-54 in recent general population surveys. By comparing reports of respondents’ childhood and current religion, researchers estimated transition rates between all religious groups. These switching rates were calculated separately for males and females. The overall rate of religious change was then applied to cohorts ages 15-29 in population projections. For projection purposes, each person was allowed one switch, which is all that is directly measured in the surveys this method draws upon. Religious switching results and elaboration of this approach will be described in greater detail in future Pew Forum reports.

Topics

Regions & Countries

Formats

Topics

Regions & Countries

Formats

General Procedures for Composition Estimates

Data Collection and Documentation

Standardization of Religious Categories and Measurement Strategies

Identifying Primary Source(s) for Religious Composition Estimates

Making Adjustments for Groups Not Adequately Measured

Adjusting for Limitations in a Survey Questionnaire

Adjusting for Sampling Limitations

Making Adjustments for the Religious Affiliation of Infants

Making Adjustments for Missing Religion Data

Age Structure Procedures

Estimating Religion by Age and Sex

Adjustments to Minimize Errors Due to Sample Size

Matching Religion by Age and Sex to Overall Population by Age and Sex

Sources for Age Structure Data

Projecting Populations to 2010

The Projection Approach

Estimating Fertility

Estimating the Size and Religious Composition of Migrant Flows

Estimating Religious Switching

Other Topics

China

Differences Between Current Estimates and Previous Pew Forum Estimates

A Note on Regions

A Note on Country and Territory Designation

A Note on Rounding

Sign up for our Religion newsletter

Report Materials

Table of Contents

General Procedures for Composition Estimates

Data Collection and Documentation

Standardization of Religious Categories and Measurement Strategies

Identifying Primary Source(s) for Religious Composition Estimates

Making Adjustments for Groups Not Adequately Measured

Adjusting for Limitations in a Survey Questionnaire

Adjusting for Sampling Limitations

Making Adjustments for the Religious Affiliation of Infants

Making Adjustments for Missing Religion Data

General Procedures for Composition Estimates

Data Collection and Documentation

Standardization of Religious Categories and Measurement Strategies

Identifying Primary Source(s) for Religious Composition Estimates

Making Adjustments for Groups Not Adequately Measured

Adjusting for Limitations in a Survey Questionnaire

Adjusting for Sampling Limitations

Making Adjustments for the Religious Affiliation of Infants

Making Adjustments for Missing Religion Data

Age Structure Procedures

Estimating Religion by Age and Sex

Adjustments to Minimize Errors Due to Sample Size

Matching Religion by Age and Sex to Overall Population by Age and Sex

Sources for Age Structure Data

General Procedures for Composition Estimates

Data Collection and Documentation

Standardization of Religious Categories and Measurement Strategies

Identifying Primary Source(s) for Religious Composition Estimates

Making Adjustments for Groups Not Adequately Measured

Adjusting for Limitations in a Survey Questionnaire

Adjusting for Sampling Limitations

Making Adjustments for the Religious Affiliation of Infants

Making Adjustments for Missing Religion Data

Age Structure Procedures

Estimating Religion by Age and Sex

Adjustments to Minimize Errors Due to Sample Size

Matching Religion by Age and Sex to Overall Population by Age and Sex

Sources for Age Structure Data

Projecting Populations to 2010

The Projection Approach

Estimating Fertility

Estimating the Size and Religious Composition of Migrant Flows

Estimating Religious Switching

Other Topics

Related

The Changing Global Religious Landscape

Q&A: The Muslim-Christian education gap in sub-Saharan Africa