TABLE 1

Comparison of MLEs to the true parameter values for simulated data sets

ModelMean llMLEsllTruthaMax llMLEsllTruthb% MLEs = truthcCoverage of multidimensional 95% C.I.'sdCoverage of one- dimensional 95% C.I.'se% points outside 95% C.I.'sf
Ancient growth
cM0.6313.473.0100.099.8794.79
Recent growth
cM0.4373.0922.6100.099.9398.84
Bottleneck
cM0.3636.3147.499.899.5299.87
Hotspotsg0.5053.4317.0100.099.8099.65
cM, 136.685179.070.00.08.3699.98
0.6236.5426.499.699.4099.72
Clark's phasing algorithmh0.7794.9019.8100.099.9699.43
Ascertainment bias1.99812.6514.694.097.6499.86
• a The average overall data sets of the log-likelihood at the MLEs minus the log-likelihood of the true demographic parameters.

• b The maximum distance between the log-likelihood at the MLEs and the log-likelihood of the true demographic parameters.

• c The proportion of data sets where the MLEs for all parameters were the true demographic parameters.

• d The proportion of data sets where the true parameter values were <3.9 or <5.5 log-likelihood units from the MLEs, for the growth and bottleneck models, respectively.

• e The proportion of data sets where the true value of each parameter was <1.92 log-likelihood units from the MLE using the profile log-likelihood curve, averaged over three or five parameters for the growth and bottleneck models, respectively.

• f The fraction of grid points (see results) having a log-likelihood >3.9 or >5.5 log-likelihood units, for the growth and bottleneck models, respectively, from the MLEs.

• g Each window has five recombination hotspots, but for the whole window cM.

• h Haplotype phase was inferred in the test data sets and simulations to estimate using Clark's phasing algorithm (see methods).