I am working on panel attrition (loss of respondents at each wave of follow-up of a longitudinal survey). A multivariate model (discrete-time event history analysis) to analyse panel attrition is discussed in:
Watson, D. (2003), ‘Sample Attrition Between Waves 1 and 5 in the European Community Household Panel’, European Sociological Review , vol. 19, no. 4 (September), pp. 361-378.
My data contains only three waves and the age of the respondents varies at wave one (baseline) from 2 to 10 years. I would like to describe how panel attrition varies by age. At wave three the age of the oldest respondents is 15 years.
My question is: Is it reasonable to create one life table for all respondents containing as time-intervalls the age groups ([2-3), [3-4), [4-5), [5-6),...,[14-15))?
I finally found the answer!*
I finally found the answer in a German (sorry!!!) textbook (Handbuch Familiensoziologie, pp 269-318, Quantitative Auswertungsverfahren in der Familiensoziologie - Ereignisanalysen und dyadische Analysen Oliver Arránz Becker, Daniel Loi, 2014).
In this textbook two problems are described in using different levels of age to model the hazard rate:
a) The age-groups might have a different hazard rate at the same age (level-effect on the y axis).
b) The timing of the event might be different for the age-groups, which results in an effect on the x-axis.
