This section describes the sources and variables used for the analyses in this report. It explains known limitations of the data. Next, it outlines the strengths and weaknesses of the map used, as well as possible alternatives and reasons why data is shown as missing in some areas. Finally, the multilevel mixed-effects modeling used in the “Causes of change: Fertility” section in Chapter 2 is described in more detail.
Quantitative analyses in this report rely on two main sources: India’s decennial census and the National Family Health Survey (NFHS). Information about population sizes from 1951 to 2011 come from the census. The census has collected detailed data on individuals and households, including religion, since 1881, when India was still under British rule.
Data on fertility and how it is related to the state a woman lives in, her age, years of education, household wealth and urban versus rural status comes from the National Family Health Survey. The NFHS is a large, nationally representative household survey with more extensive information on childbearing than the census. The first round was conducted in 1992-93. Total fertility rates from each of its four available rounds are included. The analysis of how various factors are related to the fertility of women in their 40s relies only on wave four of the survey, conducted in 2015 and 2016. As of the publication of this study, the fifth wave of the NFHS is in progress. Microdata will be made available later, permitting analysis of trends by religion. Preliminary total fertility rate (TFR) results have been released for states in which data collection is already complete: https://pib.gov.in/Pressreleaseshare.aspx?PRID=1680702.
Brief references to historical patterns of fertility for Indian women overall come from the United Nations World Population Prospects. Data on migration of all people born in India also comes from the UN and underlie a Pew Research Center analysis of their religious composition. Comparisons of the development levels of Indian states are based on Human Development Index scores – an aggregate measure that includes dimensions like life expectancy, average years of schooling and per capita income.
Survey responses pertaining to religious switching and interfaith marriage come from a Pew Research Center survey of 29,999 Indian adults conducted in late 2019 and early 2020. Projections of future population sizes were also made by the Center.
There are known issues with census data on religion in India. Christians in particular appear to be undercounted; some Christians who belong to Scheduled Castes may choose to identify as Hindu when completing official forms such as the census. This is due to a mandate in the Indian Constitution specifying that only Hindus, Sikhs and Buddhists can receive some types of caste-based government affirmative action benefits (known in India as “reservations”). Analyses in Pew Research Center’s Global Religious Landscape and Future of World Religions reports accounted for this by estimating that 10% of Christians in India state their affiliations as Hindu in the census and by adjusting the population figures accordingly. This report relies on unadjusted numbers due to the uncertainty surrounding how a bias in responses might have changed over six decades.
Buddhists, Jains and Sikhs are grouped with Hindus in some laws that enable access to social services or employment and educational preferences to other religious minority groups. There may be some confusion among census takers and enumerators as to which religious affiliations apply to which circumstances and government forms.
Anecdotes presented during a consultation on religion in the Indian census with leading India demographers at the 2010 Asian Population Association’s meeting in New Delhi described how, rather than asking respondents for their religious affiliation, some census enumerators made their own inferences based on individuals’ names or items and symbols in the home. Others may have asked a religious or community leader about the characteristics of nearby individuals. These practices are inconsistent with census protocols and may have biased responses toward the majority religion, Hinduism, but there is no known quantitative data on this topic.
While, constitutionally, India is a secular country with protections for religious minorities, Indians still generally experience “high” levels of government restrictions on religion, according to an annual Pew Research Center study. Nine states have made it illegal for Christians and Muslims to proselytize. Government restrictions on religion may influence census takers to state a religious affiliation other than the one they practice or believe in, though a Pew Research Center survey also found that religious switching is unusual.
Map borders and missing data
The map template used throughout this report is based on the Census of India map. It is one of many maps of India, and it has both strengths and weaknesses. The map used here depicts the boundaries of states and territories according to the Indian government, including areas within them for which data was unavailable in 2011. The reasons for missing data vary. For example, a portion of the Western state of Gujarat is usually under standing water during the monsoon season. North Sentinel Island, officially part of the Andaman and Nicobar Island administrative region, is home to the Sentinelese, an uncontacted tribe. The tribe has a long history of vigorous defense against outsiders, including attacks on approaching boats that have killed prospective visitors. The island is protected and monitored remotely by the Indian government but is effectively sovereign.
A large swath of Jammu and Kashmir, as it is depicted in the map, covers the region of Kashmir occupied by Pakistan, where demographic data on Indians is not collected. The northeastern section of the state has been occupied by the Chinese military since the 1950s but is mostly uninhabited. While the Indian government regards the full area labeled Jammu and Kashmir as unequivocally part of India and views long-standing occupations as illegal, the governments of Pakistan and China do not. International organizations like the United Nations recognize these regions as disputed. Pew Research Center takes no position on the matter of which countries’ claims to this region are legitimate. Rather, this map was chosen because it is consistent with the main data source and is commonly accepted as accurate by people in India.
Details on modeling
The section of this report on the relationship between religion, fertility and its correlates presents the predicted number of children an average Indian woman in her 40s would have, controlling for other factors, in order to estimate the degree to which fertility gaps can be explained by other differences between religious groups. These were calculated using a weighted multilevel mixed-effects generalized linear model assuming a Gaussian distribution. The sample consists of 149,294 ever-married women ages 40 to 49 (1,49,294 in the Indian number system) in wave four of the NFHS (the most recent for which microdata is available). The model accounted for religious affiliation, urban versus rural setting and the interaction between the two as factors. It also accounted for linear and nonlinear effects of the common household wealth score, years of schooling and age. Women were modeled as nested within districts (the primary sampling unit in the NFHS), with districts nested within states. The model includes a random slope to allow the effect of religion to differ from state to state.
These groups also differ in their average age at first marriage and first birth. Both Hindu and Muslim women in their 40s married for the first time at age 18, on average, and the average age for Christians was 21. On average, Hindu women had their first child at age 21, Muslims at 20 and Christians at 22. Since these factors could also be related to fertility, they were included in robustness checks of the model. The results generally hold with or without these additional variables. The predicted number of children for Christians is 3.8 if ages at first marriage and birth are accounted for, compared to 3.5 if they are not. Estimates for Hindus and Muslims change by less than 0.1 when these variables are included. Because age at first marriage and birth are closely related to education, and excluding them does not meaningfully affect results, they were dropped from the final model to avoid overfitting.