Table 3. Performance on 25 published simulated datasets in populations of size 50 [12].
Slow Clock simulations | Fast Clock simulations | |||
---|---|---|---|---|
Prior information | Naïve a | Informative b | Naïve a | Informative b |
MCMC sampling | ||||
Continuous parameter samples (95% interval of ESS) | ||||
μ | 168; 578 | 401; 951 | 62; 570 | 90; 916 |
mG | 259; 936 | 1948; 4531 | 798; 1652 | 2999; 12748 |
mS | 42; 145 | 43; 87 | 199; 1737 | 170; 461 |
r | 189; 357 | 220; 369 | 43; 143 | 184; 613 |
tinf | 166; 1532 | 222; 559 | 216; 2264 | 295; 2796 |
phylogenetic tree topology c | 580; 1764 | 1501; 4331 | 104; 972 | 83; 796 |
Infectors (% Fisher’s exact tests accepted) | ||||
between chains | 95.8% | 97.4% | 92.5% | 97.6% |
autocorrelation | 94.6% | 96.4% | 87.7% | 96.0% |
Parameter inference (95% interval of posterior medians) | ||||
log10(μ) | -4.85; -4.67 | -4.95; -4.78 | -3.26; -3.15 | -3.20; -3.16 |
mG | 2.1; 5.0 | 3.7; 5.7 | 4.2; 6.1 | 4.7; 6.1 |
mS | 6.5; 10.2 | 11.2; 12.6 | 9.7; 13.9 | 11.4; 12.6 |
r | 0.75; 1.3 | 0.36; 0.77 | 0.82; 2.1 | 0.30; 1.2 |
Tree inference | ||||
Infection times (coverage: % of 95% CIs containing the true value) | ||||
76.1% | 97.8% | 94.6% | 94.2% | |
Infection time bias (median) | 3.97 days | 0.10 days | 0.25 days | 0.01 days |
95% interval of medians | (-3.20; 8.80) | (-1.98; 2.28) | (-6.49; 5.86) | (-2.00; 1.97) |
Infectors (number correct/number identified) | ||||
Edmonds’ | 29.0/49.3 | 30.8/49.3 | 32.8/49.3 | 45.5/49.3 |
MPC | 25.4/49.3 | 29.9/49.3 | 30.8/49.3 | 45.5/49.3 |
≥50% support | 12.9/14.4 | 25.2/30.9 | 23.4/28.9 | 45.4/48.7 |
≥80% support | 3.0/3.1 | 18.5/19.7 | 4.4/5.0 | 40.9/42.1 |
Results are based on two MCMC chains of 25,000 samples each. The mean outbreak size was 49.3 cases; ESS, effective sample size; CI, credible interval; MPC, maximum parent credibility.
a aS = 3, μS = 1, σS = ∞
b aS = 144, μS = 12, σS = 1
c path-distance approximate ESS [23]