# Thread: Cohort as level 1 regressor in hierarchical age-period-cohort models (HAPC models)

1. ## Cohort as level 1 regressor in hierarchical age-period-cohort models (HAPC models)

Dear Talkstats,

What a great site! I cannot believe I only just discovered the site. I look forward to many interesting debates with you guys in the future.

Currently, I'm having some trouble understanding the way Raphael J. Nawrotzki and Fred C. Pampel make use of 'Cohort' as a level 1 regressor in their study' Cohort change and the diffusion of environmental concern: a cross-national analysis'. Their data is two waves in a repeated cross sectional survey (The ISSP).
They construct the cohort variable in the following way: "by subtracting the age of an individual from the survey year. The age of respondents ranged from 16 to 96 years with a mean of 45 years, and data were collected for two waves, 1993 and 2000; thus, we include cohorts born between 1900 and 1984".
They conclude that the "Cross-classified multilevel modeling allows us to identify a nonlinear interaction between cohort and education, our core measure of SES, in predicting environmental concern, while controlling for age and period."
I have attached a picture of their final model.

I am very interested in replicating this way of having the cohort as a level-1 regressor, and age and period on level 2, with repeated cross sectional data.

However, I cannot find any literature on whether you are allowed to treat 'Cohort' as a level 1 regressor in HAPC models? In the book 'Age-Period-Cohort Analysis: New Models, Methods, and Empirical Applications' by Yang Yang and Kenneth C. Land, they always use 'Age' as the level 1 regressor, and period and cohort as the level 2 regressor. They write: "time periods in a repeated crosssectional survey design are not fully nested within cohorts and cohorts are not fully nested within time periods. Rather, one obtains a cross-classified structure with individuals nested within cells defined by birth cohorts and time periods."
However, Raphael J. Nawrotzki and Fred C. Pampel write that "The individual data are nested within a crossclassification of period and age. By measuring age in five-year categories, this approach helps to deal with the dependence of cohort on age and period.".

Can you guys help me make sense of this? Can you freely choose between whether to have the age or cohort at the level-1? To me it makes most intuitive sense to see people as nested in cohorts (and periods) and not age (and periods).

Thank you

 Tweet