Mornin' Richard (at least it is here). I think your second paragraph pretty well sums up the problem: "That said, I think the WikiTree estimates are useful (and in this sense at least, accurate) as an assessment of the mean expected contribution from a given relative. That at least is pretty simple, right? Just based on the number of meioses in a given path."
Therein lies the rub, and the previous commentary about the Buffalo et al. paper: there is no established mean. WikiTree is using, I believe--and I've mentioned this a few times before but never had input to correct me--some flavor of the Coefficient of Relationship to calculate the percentages we see on the profiles. The CoR can't apply to the X chromosome because the CoR is based on 50% autosomal contribution from both the mother and the father. The expected contribution of the X chromosome can't be a simple divide-by-two.
Too, our testing/comparison services use half-identical regions for matching, so a female who would receive 50% of her X from her father and 50% from her father, as WikiTree correctly reports, would compare on HIRs as 100% to each. I think that sort of thing is more likely to cause confusion than would removing the percentages and just leaving the little xDNA icon to illustrate the inheritance path.
And when we try to diverge from the direct ancestral lines, it goes even farther astray from reality. We see it immediately with full brothers, whom WikiTree projects to share 100% of their X chromosomes...which almost never happens because with each ovum the mother's X has most likely gone through crossover from one to three times. Unless something like the Buffalo et al. curve-fitted simulations are used, I just don't think X chromosome sharing percentage projections are of any real use. The 50/50 assumption can't be used even as close as 1st cousin relationships.
Yep; I'm (sorta painfully) aware of the differing SNP overlaps between microarray tests. If you look at my informal Ancestry v2.0 comparison paper I linked to above, you'll see some specific data for those tests broken down in summary to the autosomes, X, Y (PAR and not), and mtDNA, with some SNP total summaries included of four revs of the Illumina default OmniExpress and GSA chips. Frankly, from the earliest days of popular microarray testing I think most people have felt that the most genealogy/population-relevant SNPs were carefully evaluated and selected for testing. That really isn't the case. And the fact that in the span of just a handful of years we're dealing with tests that have as little as 23% SNP overlap is a continuing challenge.
I've never looked specifically the X overlap among our popular microarray tests. Hm; may want to do that.
A quick aside: if you used something like WGSExtract to compile your "all SNP" GEDmatch upload, what you ended up with is a collection of SNPs for which there was at least one correspondent in the "template" commercial microarray results. In other words, you probably uploaded around 2.1 million SNPs to GEDmatch which they then "slimmed" to around 1.1 million. Those commercial "templates" should have combined for a total of about 53,000 X chromosome SNPs, which still is equivalent to only about 0.03% of the chromosome.
The dbSNP database has over 27 million SNPs cataloged on the X chromosome, and even at that we have to consider that it wasn't until last September that we saw the first telomere-to-telomere assembly of the human X (Miga et al., Nature), filling in the centromeric region and 29 previously known gaps. Those data won't hit our reference maps until the next major GRCh update, though...and the old GRCh37 reference we still use for most of genealogy had a number of corrections made in GRCh38. Genealogy is well behind the times there. The extra SNP count from my WGS helped me at GEDmatch, at least a bit, to reduce the number of probable X chromosome false matches. A number of matches fell off once there were more SNPs available to fill in the blanks.
As for relative effectiveness of imputation on the X, I really don't know. Above my pay grade. :-) That accuracy is predicated first and foremost by the quality and size of the cohort genomes in use at any given time and, particularly since none of our testing companies use the X only for matching (a minimum-threshold autosomal match must exist first), I have a feeling that the accuracy of imputation for the X lags behind a bit...at least when it comes to the genealogy tools we have available.
Diversity on the X chromosome is largely driven by the matrilineal line, and for that reason will vary in demography/geography from the autosomes. Some studies (like Arbiza et al., American Journal of Human Genetics, June 2014) indicate that the diversity seems to be fairly consistent within broad continental populations, but with the X diversity notably lower in both European and East Asian populations. So the autosomal imputation references carried over to the X at that level should be fairly accurate, and most accurate for European/East Asian cohorts. The flip side is that the lack of diversity also implies those populations will have a more difficult time making genealogically-relevant determinations about X inheritance as the generations go back in time.