Statistical Methods for Rates and Proportions 3rd ed

Textbook for the course: EHSC-GA 2045 Categorical Data Analysis (Fall 2011)



Title: Statistical Methods for Rates and Proportions 3rd ed
Authors: Joseph L. Fleiss, Bruce Levin, Myunghee Cho Paik
Edition: 3
Finished Date
Rating: 4
Language: English
Genres: Data Science, Statistics
Level: entry
Publishers: Wiley-Interscience
Publication Date: 2003-09-05
ISBN: 978-0471526292
Format: Pdf
Pages: 800
Download: Pdf

Ch 14

regression models for matched samples

1.    direct and indirect parametric modeling of matched-sample data  

questions

Is it valid to analyze data from a matched case-control study

as if they had been collected prospectively,
that is, to ignore the retrospective design and
to fit a parametric logistic regression model for P(Y|X,Z)
to estimate the disease-exposure log odds ratio?

unmatched case-control studies

use logistic regression

as if the data had been collected prospectively

  • consistent estimates and std errors
    • all coefficients
    • except the intercept term
    • assume
      • a correctly specified population disease prevalence model
      • unbiased sampling for cases and controls

paired case-control studies

use logistic regression

as if the data had been collected prospectively

  • no such justification
    • biased log odds ratio
    • grossly misleading coefficients for other terms in the model
    • the bias is negligible or extreme
    • paper
      • Levin and Paik (2001)

conditional logistic regression

  • do not suffer from the above potential bias
  • strong recommendation to use it

matched-sample prospective studies

biased