The previous step found that about half the rise in the share raised Catholic (third-plus-generation non-Hispanic white adults, 27.0% in 2000-09 to 29.4% in 2021-24) was a rise within birth cohorts. Religion at age 16 cannot change, so that part should be an artefact. I predicted it would come from web interviews.
Splitting 2021-24 by interview mode in the General Social Survey: in person the share is 25.7% [23.0, 28.1] (n=1,279); by web it is 31.0% [29.5, 32.8] (n=4,048). Against 2000-09, the within-cohort term is -2.53 points [-5.35, -0.17] in person and +3.05 [+0.99, +4.88] by web. Cohort replacement is about +1 point in both modes. Web exceeds in person in each of the four cohorts.
Caveat: mode is not randomised, so this mixes who answers with how they answer. The 2022 split-ballot test is the next check. Code, tables and log: research/catholic-demography/experiments/m5f_mode/.