Far East Journal of Theoretical Statistics
Volume 16, Issue 2, Pages 177 - 204
(July 2005)
|
|
LARGE SAMPLE BIAS IN TWO-PHASE SAMPLING CAUSED BY EXTERNAL VALIDATION
Ming Ji (U. S. A.), William Huster (U. S. A.), Nebiyou Bekele (U. S. A.), Wesley Johnson (U. S. A.) and David Rocke (U. S. A.)
|
Abstract: In
this paper, we show that Tenenbein’s double
sampling estimate of prevalence is usually
biased if the validation is drawn from an
external population. An explicit large sample
bias formula was derived for Tenenbein’s
estimates when the double samples are from two
arbitrary populations. Based on the large
sample bias result, we identified three
possible sources of the bias which also lead
to matching conditions for the detection and
reduction of the bias. Two sources of biases
can be reduced by propensity score matching
and the remaining bias is a continuous
function of the difference between the two
sampling populations. Simulation results are
presented to demonstrate that the existence of
bias when any one of the three matching
conditions was violated. A case study on
estimating prevalence of endometrial
abnormality among postmenopausal women is
presented. Our conclusion is that external
validation can lead to bias in the classical
two-phase sampling estimates. Therefore,
population difference should be carefully
examined before applying traditional two-phase
sampling estimates with external validation. |
Keywords and phrases: screening tests, double sampling, external validation, bias, propensity score. |
|
Number of Downloads: 251 | Number of Views: 765 |
|