Substantive Analyses of Intensive Longitudinal Data
Barber (University of Michigan )
Recent, rapid improvements in technology have greatly facilitated survey researchers' ability to collect frequent, closely-spaced surveys. This is sometimes called intensive longitudinal data, or frequent assessment panel data. Truly illuminating analyses of the resulting data , which take full advantage of the frequent assessments, can be challenging.
Some traditional analytic methods, such as hazard models and linear growth curve models, are well-suited to intensive longitudinal data because they involve time-varying variables, and the intensive longitudinal data are easily translated into such time-varying variables. However, standard latent trajectory methods have difficulty with 100+ assessments per unit of analysis. In addition, intensive longitudinal data are particularly well-suited for examining transitions, sequences, and patterns, which are not easily illustrated with traditional analytic methods.
This session will present examples of innovative substantive research using many closely spaced assessments per unit of analysis. These examples will spur other researchers to analyze their own intensive longitudinal data in new and innovative ways.
1. Comparison of precarious job entry histories over time
Mr Ralf Dorau
(Federal Institute for Vocational Education and Training)
If you have a large dataset including many cohorts of job entry histories, computing similarities between all sequences and also clustering may not work with your computer hardware or statistical software. Furthermore clusters with many different states cannot be interpreted easily. Comparing different datasets is nearly impossible. In order to compare the occupational integration of job entry histories we have to develope some normative conditions to assign job entry histories to be integrated, precarious or decoupled. So we have developed an approach that includes the duration of continuous states and the temporal incidence of particular states.
2. Survey design based inferences for regression effects in non-stationary models for longitudinal count data
Dr Vandna Jowaheer
The estimation of regression effects in panel count data models is a challenging issue
under the finite population set up where the estimation is done
based on a sample of small number of clusters chosen from a finite population with
large number of unbalanced clusters. In this talk, I will use a two stage cluster
sampling design to construct the desired sample and examine the role of such sampling
designs on the estimation of the parameters of the generalized linear longitudinal
models based super-population for clustered count data.