De (on)mogelijkheid om het leeftijds-, periode- en cohorteffect te bepalen. Een evaluatie aan de hand van het kerkbezoek in Nederland tussen 1970 en 1995

Samenvatting

Since the variables age, period, & cohort are collinear, regression coefficients cannot be estimated properly, & straightforward application of their effect analysis is impossible. Three proposed solutions to this problem are tested through evaluation of variance inflation factors & condition numbers, & through simulations with small perturbations in original data on church attendance, 1975-1990, in the Netherlands among 26,591 people born 1900-1977. Solution 1, in which one of the three effects is set to zero, yields effects that are not harmed by collinearity; however, both strength & sign of the effects depend on which effect has been set to zero. Solution 2, in which the categories are merged into broader categories, is not harmed by collinearity either; however, significance of the effects depends on how the categories are combined. Solution 3, in which only a few categories are merged or a variable is measured more precisely, yields highly unstable results due to collinearity. An alternative, more content-based solution, ie, using more substantive variables that reflect underlying hypotheses is proposed. 2 Tables, 6 Figures, 27 References. Adapted from the source document.