AUEB SEMINARS - 29/6/2016: Challenges in Modeling Rate Data with Time Factors – the Identifiability Problem with Linear Confounders
Tue 28 Jun 2016 - 16:04
ΚΥΚΛΟΣ ΣΕΜΙΝΑΡΙΩΝ ΣΤΑΤΙΣΤΙΚΗΣ – ΙΟΥΝΙΟΣ 2016
Wenjiang Fu
Department of Mathematics, University of Houston
Challenges in Modeling Rate Data with Time Factors – the Identifiability Problem with Linear Confounders
ΤΕΤΑΡΤΗ 29/6/2016 13:15-14:00
ΑΙΘΟΥΣΑ 607, 6ος ΟΡΟΦΟΣ,
ΚΤΙΡΙΟ ΜΕΤΑΠΤΥΧΙΑΚΩΝ ΣΠΟΥ∆ΩΝ
(ΕΥΕΛΠΙ∆ΩΝ & ΛΕΥΚΑ∆ΟΣ)
ΠΕΡΙΛΗΨΗ
In monitoring human behavior and disease development, event rate (disease incidence or mortality, crime, consumer consumption, investment and sales, etc) are often collected / archived by age and calendar years. The aim of modeling such type of data in a rectangular table is to estimate accurately the temporal trend in calendar year or compare rate in different years or groups. Although the data seems to be simple and straightforward, the statistical models, however, suffer from two major challenges in modeling such one rate per cell data. The first one is the identifiability problem in the regression model with linearly dependent covariates: age (rows), period (columns) and birth cohort (diagonals) are linearly dependent. The second one seems to avoid the identifiability problem using a summary rate approach for each period (column), but incurs another major problem – the direct age-standardization with a serious issue of selecting the reference age-structure for fair comparison. Although both problems have been extensively studied during the past few decades, major issues still remain.
In this talk, I will briefly introduce these twin-sister problems with US cancer mortality data, crime rate data and life insurance purchase data. I will then introduce the intrinsic estimator method and provide theoretical justification based on large sample theory to resolve the identifiability problem. I will then examine the method through simulation studies and real data in cancer studies, marketing research, and sociological studies. This work was partly supported by grants from the NIH/NCI of the US.
- AUEB SEMINARS - 15/4/2016: Modeling discrete-valued time series with abundant and heterogeneous zeros
- AUEB SEMINARS - 26/10/2016: Criteria for Bayesian model choice and implications on the n<<p problem
- AUEB Stats Seminars 20/10/2023: Joint modeling of longitudinal and competing-risk data using cumulative incidence functions accounting for failure cause misclassification by Christos Thomadakis (Adjunct Lecturer, Department of Statistics, AUEB)
- AUEB STATS SEMINARS 17/1/2017: Revisiting the Statistical Foundations of Panel Data modeling
- AUEB SEMINARS - 15/12/2016: Modeling Earthquake Recurrence Times using Physical Insights and Statistical Analysis
Permissions in this forum:
You cannot reply to topics in this forum
|
|