How Do AncestryDNA cM Numbers Compare?

With the mirrors to the other world at AncestryDNA finally cracking open a little some comparison between their results and those from Family Tree DNA, 23andMe, and GEDmatch can be made (if you've been watching the BBC series Strange and Norrell you'll understand that reference).

 I only counted segments 5 cMs and over when comparing.

Starting the comparison with a double 5th cousin who has tested with all 3 companies:
This cousin has results at the 3 major companies plus GEDmatch. In this case AncestryDNA seems to have removed the 9 cM segment on Chromosome 6. This same segment was removed from two other relatives of this match, also. From totals around 24 cMs with GEDmatch and the other two companies we go down to a total of 13.404 at AncestryDNA.

AncestryDNA total shared 13.399 cM Total

Family Tree DNA 23.41 cM Total

 

GEDmatch cM Total 25.2

23andMe cM Total 24.3

 

Moving on to this match who has tested with the 3 major companies:
In this case AncestryDNA cut the cM total from 25 to 26 cM's to 6.748. This would mean, if Ancestry is right, this may not be a true cousin match. I haven't established a relationship in this, case so far. So they may be right in this case?


AncestryDNA Total 6.748

Family Tree DNA Total 25.27
 



23andMe Total 26.2

Moving on to a 4th Cousin:
The particular match is a known, confirmed, 4th cousin. AncestryDNA predicts he is a 5th to 8th cousin. AncestryDNA totals are much smaller with only 6.748 total, while 23andMe says we share 32 cMs. I tend to believe the higher number.

AncestryDNA 4th cousin match total 6.748 cMs

 

23andMe 4th cousin match total 32 cM's

 

I've compared a couple more matches, which basically follow along these same lines. I haven't found any cM totals predicted to be closer than 8 cM's, when comparing with AncestryDNA. Some comparisons are 20 cMs different. I've made notes for all of the matches I've been able to identify at AncestryDNA. Looking at the totals for some of these matches I'm surprised at how low the cM totals are. I believe if we compared elsewhere the totals would be higher. I don't know about how the phasing and filtering is working for anyone else? It's not working well for me. I believe the best approach to matching is to provide testers with unphased, unfiltered matches who share a 7 cM segment or more, and with at least 700 SNPs. I don't feel the current matching system is helpful enough in eliminating bad matches, because there are many matches sharing cM totals under 7 (many sharing only 5 cMs total). AncestryDNA's system also tends to remove some good matches. Losing good matches, and having close cousins downgraded, in order to eliminate a few bad matches isn't worth the trade off.