Rob van Hest Capture-recapture Methods in Surveillance - RePub ...
Rob van Hest Capture-recapture Methods in Surveillance - RePub ...
Rob van Hest Capture-recapture Methods in Surveillance - RePub ...
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Estimat<strong>in</strong>g <strong>in</strong>fectious disease <strong>in</strong>cidence<br />
estimates f<strong>in</strong>d<strong>in</strong>g their way <strong>in</strong>to the scientific literature. The role of the f 1/f 2 ratio <strong>in</strong> the<br />
agreement or disagreement between three-source log-l<strong>in</strong>ear capture-<strong>recapture</strong> and<br />
truncated model estimates for the number of <strong>in</strong>fectious disease patients, especially when a<br />
parsimonious log-l<strong>in</strong>ear model is selected, should be subject of further mathematical or<br />
statistical studies.<br />
Appendix<br />
Equations for the truncated population estimators<br />
Truncated b<strong>in</strong>omial model: est(N) = obs(N) + (f 1) 2 /3f 2<br />
Truncated Poisson mixture model: est(N) = obs(N)/[1–exp(-2f 2 /f 1)]<br />
Truncated Poisson heterogeneity model: est(N) = obs(N) + (f1) 2 /2f 2<br />
Equiprobability<br />
If the truncated b<strong>in</strong>omial model is true, i.e. if the sources are <strong>in</strong>dependent and<br />
equiprobable with probability of captur<strong>in</strong>g any case = p, our estimator (f1) 2 /3f2 is correct<br />
<strong>in</strong> the sense that the expected number of unlisted cases is given by<br />
3 ( Ef<br />
1 )<br />
E f 0 Nq =<br />
3Ef<br />
2<br />
2<br />
.<br />
= (1)<br />
If we <strong>in</strong>troduce a small departure from equiprobability so that the list probabilities are (p<br />
− h, p, p + h) <strong>in</strong>stead of (p, p, p), the estimation error can be def<strong>in</strong>ed as<br />
( Ef<br />
1 )<br />
g( h,<br />
p)<br />
= − Ef<br />
0.<br />
(2)<br />
3Ef<br />
Differentiat<strong>in</strong>g with respect to h, we f<strong>in</strong>d that<br />
g<br />
2<br />
2<br />
∂g<br />
∂ g 2N(<br />
1 − p)<br />
0,<br />
p)<br />
= ( 0,<br />
p)<br />
= 0;<br />
( 0,<br />
p)<br />
= , (3)<br />
2<br />
∂h<br />
∂h<br />
3p<br />
( 2<br />
so that we overestimate, at least for small h. The same happens if we consider an<br />
asymmetrical departure, (p − h, p, p). In that case,<br />
and there is aga<strong>in</strong> an overestimate.<br />
g<br />
2<br />
∂g<br />
∂ g 2N(<br />
1 − p)<br />
0,<br />
p)<br />
= ( 0,<br />
p)<br />
= 0;<br />
( 0,<br />
p)<br />
= , (4)<br />
2<br />
∂h<br />
∂h<br />
9p<br />
( 2<br />
2<br />
153