Table 5.
Results of Simulations Designed to Mimic Conditions of HERS Examplea.
| Variable | β̂ (StdErr)h | 95% CI Coverage |
|---|---|---|
| Ideal Analysisb | ||
| Trichomoniasis (+ vs. −) | 0.65 (0.25) | 96.6% |
| Age (Years) | −0.05 (0.003) | 93.2% |
| Race (Black vs. Others) | 0.80 (0.19) | 96.4% |
| HIV Risk Cohort (IDU vs. Sexual) | 0.28 (0.18) | 95.0% |
| HIV Status (+ vs. −) | 0.25 (0.19) | 96.4% |
| Naïve Analysisc | ||
| Trichomoniasis (+ vs. −) | 1.54 (0.29) | 12.2% |
| Age (Years) | −0.02 (0.002) | 0 |
| Race (Black vs. Others) | 0.38 (0.18) | 37.2% |
| HIV Risk Cohort (IDU vs. Sexual) | 0.83 (0.18) | 12.8% |
| HIV Status (+ vs. −) | −0.42 (0.18) | 5.8% |
| Complete Analysis (nv=25%×n)d | ||
| Trichomoniasis (+ vs. −) | 0.61 (0.49) | 94.6% |
| Age (Years) | −0.05 (0.01) | 96.0% |
| Race (Black vs. Others) | 0.82 (0.34) | 95.0% |
| HIV Risk Cohort (IDU vs. Sexual) | 0.28 (0.33) | 94.6% |
| HIV Status (+ vs. −) | 0.23 (0.35) | 95.6% |
| Complete Analysis (nv=15%×n)e | ||
| Trichomoniasis (+ vs. −) | 0.58 (0.62) | 94.2% |
| Age (Years) | −0.05 (0.01) | 93.6% |
| Race (Black vs. Others) | 0.82 (0.40) | 94.2% |
| HIV Risk Cohort (IDU vs. Sexual) | 0.28 (0.36) | 95.0% |
| HIV Status (+ vs. −) | 0.24 (0.42) | 95.8% |
| Complete Analysis (nv=10%×n)f | ||
| Trichomoniasis (+ vs. −) | 0.50 (0.76) | 93.0% |
| Age (Years) | −0.05 (0.01) | 93.6% |
| Race (Black vs. Others) | 0.84 (0.45) | 94.0% |
| HIV Risk Cohort (IDU vs. Sexual) | 0.28 (0.48) | 93.6% |
| HIV Status (+ vs. −) | 0.26 (0.50) | 93.6% |
| Complete Analysis (nv=5%×n)g | ||
| Trichomoniasis (+ vs. −) | 0.23 (1.05) | 84.6% |
| Age (Years) | −0.05 (0.02) | 93.0% |
| Race (Black vs. Others) | 0.94 (0.58) | 94.4% |
| HIV Risk Cohort (IDU vs. Sexual) | 0.35 (0.62) | 95.2% |
| HIV Status (+ vs. −) | 0.29 (0.65) | 93.4% |
500 simulations. nm=690, nv=214. Data were generated from eqns. (1) and (18)–(20). True parameters: (β0=0.63, β1=0.64, β2=−0.05, β3=0.79, β4=0.28, β5=0.23, θ0=−3.27, θ1=2.92, θ2=1.34, θ3=0.36, θ4=0.91, θ5=−0.61, γ0=−3.43, γ1=2.48, δ0=−5.64, δ1=4.13, δ2=0.45, δ3=1.28).
ML estimates from eqn. (1).
ML estimates from eqn. (1) with (Y*, X*) replacing (Y, X).
ML estimates of primary parameters are obtained by jointly modeling eqn.s. (1) and (18)–(20). The internal validation sample accounts for 25% of the total sample. 500/500 replicates converged.
The internal validation sample accounts for 15% of the total sample. 499/500 replicates converged.
The internal validation sample accounts for 10% of the total sample. 497/500 replicates converged.
The internal validation sample accounts for 5% of the total sample. 485/500 replicates converged.
Empirical standard deviations across 500 estimates are reported in parenthesis.