Change 1: Making it easier to choose “no religion”
As the share of religious “nones” grows, censuses that measure religion are increasingly making “no religion” the first response category in religious identity questions rather than the last. Some respondents may choose this option simply because it appears first, a primacy effect.
In 2011, “no religion” was the last religion response category in Australia. In 2016, Australia made “no religion” the first option and the religiously unaffiliated share rose 9 percentage points from 24% to 33%. Reporting on Australia’s census results, CNN noted that the “nones” outnumbered Catholics for the first time. However, the report didn’t mention that there was a change in response categories that may have contributed to this shift (Berlinger 2017). During this period, the rise of religious “nones” was lower—about three points—on International Social Survey Programme surveys that kept their measure constant.
Other questionnaire design decisions also have made it easiest to choose “no religion.” In 1991, Canada’s census introduced a write-in box for respondents to spell out their religious identity. However, identifying with “no religion” only required filling in a bubble. New Zealand and Poland introduced a similar change in recent censuses. The relative ease of choosing “no religion” may contribute, at least in part, to the rise of the “nones” in these countries.
Change 2: Shifting between one-step and two-step measures of religious identity
One-step measures of religious identity ask a question like, “What is your religion, if any?” and typically offer “no religion” as a response category. By contrast, a two-step question first asks a filter question like “Do you have a religion?” and if respondents say they do, they are invited to specify it. Outside Muslim-majority countries, these two types of questions tend to produce different results (Brenner et al. 2024; Hackett 2014; Voas and Bruce 2004; Voas 2015).
The share of religious “nones” is higher in surveys that use a two-step measure. People with low levels of religious commitment who might volunteer a religious identity in a one-step question tend to fall into the “no religion” bucket in a two-step question.
Slovakia changed from a one-step to a two-step question between 2011 and 2021. In 2021, the unaffiliated share of the population was 25%, up from 15% in the 2011 census. It is possible, however, that much of the apparent change between census waves in Slovakia may have been a measurement artifact. Interestingly, the European Social Survey in 2012 used a two-step question and found an identical unaffiliated share, 25%, as the 2021 Slovakia census, which used a two-step question.
The opposite change could create an illusion of decline in the share of religious nones. For example, as Lithuania switched from a two-step question in 2001 to a one-step question in 2011, the unaffiliated share of the population dropped from 10% to 7%.
These types of changes are not limited to censuses. We observed that the unaffiliated share in Sweden dropped about 10 points between the 2010 ISSP, which used a two-step measure, and the 2011 wave, which used a one-step measure.
Change 3: Changes in survey mode
Changes in survey mode may affect social desirability and coverage biases.
When monthly surveys carried out in Spain by the Center for Sociological Research changed from in-person to phone interviews in April 2020, there was an immediate jump of 5 percentage points in the share of respondents who said they were atheist, agnostic or indifferent toward religion (González and Cabrera 2023).
Many censuses and surveys have transitioned from in-person interviews to self-administered questionnaires. For example, in Hungary, the option to complete the census online or by mail in 2011 may have contributed significantly to the 9-point rise in the unaffiliated share of the population from the 2001 face-to-face census. By contrast, with the same options for completing the census in place in 2021, the unaffiliated share rose only 2 additional points.
A change from primarily face-to-face interviews to a mail-to-web format saw the “no religion” share of U.S. General Social Survey respondents rise 5 points from 2018 to 2021. Some of this increase may have been the result of more religious Americans, including older adults, being less willing to take the survey online (Schnabel et al. 2024).
The need to test the impact of measurement changes
When media coverage overlooks methodological issues, it may mislead the public (Hackett 2013; 2023; Hackett and Tong 2025; Hackett 2026). Reporters may emphasize what appears to be a large change while overlooking, omitting or being unaware of measurement artifacts that exaggerate the change (Berlinger 2017).
Survey and census organizations should research how changes in the way religion is measured affect results. Without such study, it is difficult to distinguish real social change from methodological artifacts. For all who seek to understand religious change, including journalists, religious leaders, policymakers, researchers and the public, it’s vital to disentangle the two (Hackett 2020). Organizations should publish the results of experiments to measure the impact of measurement change. Approaches may include:
Experiments with split-samples: When changing question format, respondents can be randomly assigned to the new and old conditions. Differences between the groups can be used to quantify the measurement effect.
Experiments with two modes: When survey modes are changing, a mode experiment may use the old and new mode of data collection (Pew Research Center 2021).
External data comparisons: While it would be ideal for organizations to conduct their own experiments, they may also evaluate the extent to which religious change has occurred on other high-quality surveys that have maintained consistent methodology during the period of interest.
Organizations should thoroughly describe their methodological changes and draw attention to how changes may affect trend data (Sullivan et al. 2012). Researchers, journalists, and the public need help understanding whether apparent change is primarily the result of measurement change.
NEW: We highlight 3 changes in measurement of religious “nones” & call for organizations that make these changes to study their effects.
From Matthew Conrad & me. www.surveypractice.org/article/1595...
