|
ABSTRACT Distinguishing Changes
in
Underlying Status from Changes in Measured Status
When longitudinal measurements imperfectly
represent intended
outcomes, observed changes may result from shifts in the distributions of
intended
outcomes, or from shifts in the distributions of measurements conditional
on intended outcomes. To exemplify the latter phenomenon: persons
reporting their health status may increasingly trust an interviewer as
acquaintance grows over the course of a study, hence be increasingly
willing to report poor health regardless of any underlying decline. This
paper proposes a latent variable framework and methodology for assessing
changes in underlying status versus changes in the measurement of status.
The framework applies to designs that collect multiple measures of an
intended outcome concurrently and track all the measures
longitudinally. Identification issues are considered. Results
validating the methodology
are reported. Methods are exemplified using a battery of self-reported
health items collected in a
7-round investigation of disability in older women. The paper's framework
recognizes that underlying and measured
change cannot be distinguished without strong modeling assumptions;
treating such assumptions as quantities to be varied in sensitivity
analyses, it is hoped that the proposed methods will inform the degree to
which underlying change can be inferred from changes in surrogates of
intended outcomes.
Return to 2001-2002 Schedule of Presentations | Return to Home Page |
|