Skip to main content
. 2020 Sep 24;30(2):425–439. doi: 10.1177/0962280220958735

Table 3:

Example TB diagnostic trial analysed with each GEE method.


Working correlation matrix
Independent Exchangeable
Odds ratio estimate 1.65 1.51
Standard error (percentage increase compared to uncorrected standard error)a
 Uncorrected 0.1684 0.1115
 FG 0.2056 (+22%) 0.1196 (+7%)
 KC 0.1891 (+12%) 0.1948 (+75%)
 MD 0.2198 (+30%) 0.1395 (+25%)
 MBN 0.1956 (+16%) 0.1422 (+28%)
 MW 0.2818 (+67%) 0.1865 (+67%)
 KC approximation 0.1921 (+14%) 0.1245 (+12%)
Degrees of Freedom
 Pan and Wallb 70.1 17.9
 Fay and Graubardb 7.6 10.5
 Clusters minus parameters 5 5
 Cluster periods minus parameters 103 103
 Cluster periods minus parameters with  clusters regarded as parameters 89 89
(95% CI) p-value
 Uncorrected (1.19, 2.30) p = 0.003 (1.22, 1.88) p < 0.001
 FG + DFFG (1.02, 2.66) p = 0.04 (1.16, 1.97) p = 0.006
 FG + DFC-P (0.97, 2.80) p = 0.06 (1.11, 2.06) p = 0.02
 KC approx. + DFFG (1.06, 2.58) p = 0.03 (1.15, 1.99) p = 0.007
 KC approx. + DFC-P (1.01, 2.71) p = 0.05 (1.10, 2.08) p = 0.02

FG: Fay and Graubard; KC: Kauermann and Carroll; MD: Mancl and DeRouen; MBN: Morel, Bokossa, and Neerchal; MW: Mackinnon and White; GEE: Generalised estimating equations.

aStandard errors given on the log odds ratio scale.

bUsing Fay and Graubard standard error.