162 
Fishery Bulletin 99(1) 
Table 3 
Average point estimates- 
—Bayes mode, Bayes mean 
and conditional maximum 
likelihood (CML) estimate 
— and their standard 
errors (in parentheses) for 25 simulated samplings of four stock mixtures composed of harbor porpoises from the Gulf of Maine- 
Bay of Fundy, Gulf of St. 
Lawrence, Newfoundland, and West Greenland. The haplotypes of stock mixtures 
were drawn from the 
original baseline samples (0%) or augmented baseline samples for which half (50%) or all (100%) of the missing haplotypes were 
replaced by singletons. Baseline and stock-mixture sample sizes were those reported by Rosel et al. (1999) and analyzed earlier in 
this section. Reported proportions do not necessarily 
sum to 1.0 because they are 
rounded. 
Stock mixture 
Gulf of Maine-Bay 
Gulf of 
and estimator 
of Fundy 
St. Lawrence 
Newfoundland 
West Greenland 
Stock mixture 1 : 0% 
0.95 
0.01666 
0.01666 
0.01666 
Bayes mode 
0.91 (0.07) 
0.03 (0.02) 
0.02 (0.00) 
0.04 (0.07) 
Bayes mean 
0.76 (0.17) 
0.07 (0.11) 
0.09 (0.12) 
0.08 (0.12) 
CML mean 
0.82 (0.09) 
0.04 (0.06) 
0.07 (0.08) 
0.06 (0.08) 
Stock mixture 2: 50% 
0.95 
0.01666 
0.01666 
0.01666 
Bayes mode 
0.85 (0.15) 
0.04(0.07) 
0.07(0.11) 
0.04 (0.07) 
Bayes mean 
0.71 (0.20) 
0.09 (0.13) 
0.11 (0.13) 
0.09 (0.13) 
CML mean 
0.68 (0.13) 
0.09 (0.10) 
0.14 (0.09) 
0.10 (0.09) 
Stock mixture 3: 100% 
0.95 
0.01666 
0.01666 
0.01666 
Bayes mode 
0.72 (0.30) 
0.08 (0.19) 
0.04 (0.06) 
0.16 (0.27) 
Bayes mean 
0.59(0.24) 
0.12 (0.17) 
0.12 (0.16) 
0.17 (0.20) 
CML mean 
0.56 (0.13) 
0.12 (0.13) 
0.13 (0.11) 
0.19 (0.14) 
Stock mixture 4: 0% 
0.25 
0.25 
0.25 
0.25 
Bayes mode 
0.25(0.26) 
0.18 (0.29) 
0.21 (0.22) 
0.36 (0.34) 
Bayes mean 
0.27 (0.22) 
0.23 (0.22) 
0.24 (0.21) 
0.26 (0.23) 
CML mean 
0.30 (0.14) 
0.21 (0.14) 
0.24(0.13) 
0.25 (0.15) 
tion. Their method depended on the fixed condition of the 
locus in the upriver population: removal of about 25% of 
such mixture individuals resulted in the remainder meet- 
ing Hardy- Weinberg equilibrium. This approach was dif- 
ficult to generalize to the other loci, most of which were 
Highly variable. Further, the approach could not provide 
a complete description of the genetic composition of the 
lower-river population. To use the information better, all 
loci available for each type of genetic data were analyzed 
to provide a Bayes posterior distribution of the population 
proportions and their allele (allozymes and microsatellites) 
or haplotype RFs (mtDNA). Each type of genetic data was 
treated separately in order to examine the consistency of 
population composition estimates from independent data. 
The stock-mixture prior distributions for the baseline 
characters required some change to accommodate the sin- 
gle-population baseline. As is routine, loci were assumed 
to have been inherited independently, and their alleles 
were in Hardy-Weinberg equilibrium for either popula- 
tion. However, the baseline prior parameters (/3s) for each 
genetic character of the upstream population were un- 
informative: their sum, the baseline prior “sample size,” 
equaled just 1, and each equaled the inverse of the number 
( J h ) of HAGs. (empirical Bayes or pseudo-Bayes methods 
for computing the prior parameters were not applicable 
with a single baseline population.) With the baseline sam- 
ple counts for locus h denoted as y h = iy hl , ■ ■ ■ ,y/,ji t Y, the 
stock-mixture prior (or baseline posterior) of HAG RFs for 
the upstream population was 
n 
n{ Qu P I ) = IT ^y h x + y hJll +J~ h 
Notice that the stock-mixture prior “sample size” was the 
unit-augmented actual sample size, n h + 1, where n h =^y h/ . 
The corresponding downstream population stock-mix- 
ture prior reflected the even greater uncertainty in that 
population’s characteristics by a downstream stock-mix- 
ture prior “sample size” equal to only 1, yet with average 
HAG RFs closely approximating those from the counts, x h = 
( x hv . . . ,x hJh Y, seen in the stock-mixture sample, viz. 
n 
X li 1 + ^ h 
x i,j Ii +t V 
+1 X 
x hj + 1 
The prior for population proportions, n(p), was the stan- 
dard Dirichlet, D( 0.5, 0.5). Three chains were generated, 
beginning from diverse upriver population proportions of 
0.95, 0.50, and 0.05 (Fig. 3). The chain lengths for allo- 
zymes and microsatellites were 10,000 samples with the 
first 5000 discarded as burn-in. The posterior sample com- 
prised the 15,000 samples from their second halves. The 
chain lengths for mtDNA were 100,000 samples and the 
posterior sample comprised the 150,000 samples from 
their second halves. 
