Table 1. Multinomial log-likelihoods comparing the fit of various models to the observed SFS derived from Sanger sequencing data and used by Gutenkunst et al. (2009) for their inference (SFSs in Figure 4).

Model	Multinomial LL	∆LL (Model − Data)
CEU
Data to data	−21,546	0
Gutenkunst	−21,555	−9
SMC++	−21,599	−53
MSMC 2-haplotype	−21,698	−152
MSMC 8-haplotype	−21,816	−270
MSMC 4-haplotype	−22,760	−1214
CHB
Data to data	−20,154	0
Gutenkunst	−20,202	−48
SMC++	−20,277	−123
MSMC 8-haplotype	−20,343	−188
MSMC 2-haplotype	−20,370	−216
MSMC 4-haplotype	−21,411	−1257
YRI
Data to data	−29,630	0
Gutenkunst	−29,647	−17
SMC++	−29,779	−150
MSMC 2-haplotype	−30,003	−373
MSMC 8-haplotype	−31,282	−1652
MSMC 4-haplotype	−32,976	−3346

Data to data denotes the best log-likelihood possible when replacing the proportions predicted by the model with the observed proportions from the SFS used in Gutenkunst et al.’s (2009) study (see Supplementary Note 4 in File S1). Gutenkunst denotes the model inferred by Gutenkunst et al. (2009) fit to the observed SFS. SMC++ denotes the model inferred by Terhorst et al. (2017) using a combined whole genome and SFS approach. MSMC 2-, 4-, and 8-haplotype denote the demographic models inferred by Schiffels and Durbin (2014) using MSMC on two, four, and eight haplotypes, respectively.