Biology Reference
In-Depth Information
sampling error swamps the genetic signature of genetic drift in the case of
LD- N e estimates 56 or when little to no pedigree structure is found in the
SA- N e method. Of the 26 estimates, only 13 ( Table 8.1 , shaded rows) gave
values that had bounded confidence intervals for both estimators. When
looking at the other 13 estimates, it appears that several of the LD- N e
estimates had upper confidence limits when the SA- N e method did not.
However, it is important to note that these LD- N e estimates (white rows
and marked in Table 8.1 ) were sensitive for the allele frequency cutoff
such that other cutoff values returned an infinity upper bound (data not
shown). In contrast, LD- N e estimates in the shaded rows had upper bound
confidence intervals regardless of allele frequency cutoff. Thus, there was
congruence between the two methods in returning estimates with
uncertainty in the upper confidence limits for the same household-by-
year samples. Thirteen of the 13 estimates with uncertainty in the upper
confidence limits (white rows) had n
21, whereas 11 of the 13 estimates
with bounded confidence intervals had an n
21 ( Table 8.1 ). Small
sample sizes will only provide bounded confidence intervals if the true N e
is small (
>
50), which is likely the case for houses 014_2002 and 092_2000
( Table 8.1 ). The reason is that the larger the true N e and the smaller the
sample size, the less likely one is to find related individuals in the sample
(Table 2 in Waples and Waples 68 ). Thus, if small sample sizes yield esti-
mates with unbounded confidence limits it is difficult to ascertain
whether the true N e is large or whether it is small, but a larger sampling
error is to blame. If one wants to detect populations that have a true N e of
500
1000, sample sizes need to be around 50 with about 20 polymorphic
loci. 56 It appears my current data set was able to get bounded confidence
limits with n
e
40 because true N e of each subpopulation was likely
much less than 500. Several studies 56 e 59 have used simulations to address
sampling, thus I refer readers to these papers for a discussion of appro-
priate samples sizes and number of loci to use in relation to types of
questions one might ask with N e estimates.
Interestingly, almost all point estimates range in the mid tens to low
hundreds. Even the unbounded confidence interval estimates, which still
can give some indication of the lower bounds of N e , tend to show low N e
point estimates. From here on, however, I will restrict my analyses and
discussion to the 11 estimates that had n
22
e
¼
>
21 ( Table 8.2 ). Even though
houses 014_2002 ( n
11) had estimates with
bounded confidence intervals, I removed them from subsequent analyses
to avoid bias. Bias may originate because I would be omitting the other
houses with n
¼
13) and 092_2000 ( n
¼
21 that potentially really do have larger effective sizes,
but could not get an accurate estimate due to small sample size. There was
a high correlation between the point estimates of the two estimators
( r
11; Table 8.2 ). These data show good congruence
between the two estimators and give me high confidence I am getting
0.894, p
0.0002, n
¼
¼
¼
Search WWH ::




Custom Search