He, uHe, and FIS outside of CI range.

[Rowen Monks]

Hello all,

My estimates for He, uHe, and FIS are at times lower than the lower confidence interval. This is the code I am using and a single line of the output. I am using the loci default.

hetsfis.bca.9999p <- dartR.base::gl.report.heterozygosity(filtereddata_gl, nboots = 9999, plot.display= T, method = "pop", CI.type = "norm", error.bar = "CI")

pop
n.Ind
n.Loc
n.Loc.adj
polyLoc
monoLoc
Ho
HoLCI
HoHCI
He
HeLCI
HeHCI
uHe
uHeLCI
uHeHCI
FIS
FISLCI
FISHCI
4
18.51293
7734
1
3979
3755
0.142549
0.139359
0.14576
0.141718
0.142752
0.148716
0.145652
0.146914
0.15293
0.027818
0.034294
0.073064

Thank you for your help!

[Jose Luis Mijangos]

Hi,

When getting confidence intervals (CI) using bootstrapping, if the resampling distribution is skewed or it's not captured by a normal approximation (i.e. CI.type = "norm"), the CI may not include the mean. Try using for CI.type "perc" or "bca".

I noticed Ho’s CI contains the mean, but He/FIS don’t. This could be due to departures from Hardy-Weinberg proportions such as a Wahlund effect. I’d suggest running gl.diagnostics.hwe and reading the function documentation as well as the references within. 

For a nice illustration of this CI not centred on the mean behaviour, have a look at Figure 6 in:
Hesterberg, T. C. (2015). What Teachers Should Know About the Bootstrap. The American Statistician, 69(4), 371–386.

It would be a nice idea to include a plot of the distribution of the bootstraps to the function to help diagnose this issue. You could send me your dataset, or a subset, and I can have a closer look and implementing the distribution plot. 

Happy to help more and if you can, please post what you find back here so others can benefit too.

Cheers,
Luis