Question about extremely high Z-scores (Z = 100) in qpDstat results

[aaannaw]

Dear Author,

I am currently using qpDstat from the Admixtools package to calculate D-statistics among several rodent species. However, I have noticed that many of my results show Z-scores = 100, and most tests have |Z| > 3. I would like to ask whether this pattern is expected or indicates a problem in my setup.

Here are some details about my data and parameters:

Reference genome size: ~2.3 Gb

Number of SNPs used: 278518514，quite large

Parameter file: blgsize: 0.01

Outgroup: a relatively distant species

Many tests were run for all possible quartets (W, X, Y, Z), not only for topologies consistent with the species tree.

My questions are:

Is it normal to get Z = 100 for many tests, or does this indicate numerical saturation (e.g., SE(D) too small)?

Should I increase the block size (e.g., blgsize: 0.05 or larger) to avoid unrealistically small SE values?

Would it be more appropriate to limit the tests to quartets consistent with the species tree, instead of testing all possible combinations?

Could the high Z-scores result from using too distant an outgroup or from excessive divergence among species? In that case, would you recommend restricting the D-statistic tests within clades and choosing the nearest outgroup for each trio?

Any guidance on how to interpret these large Z-scores and how to adjust parameters or filtering strategies would be greatly appreciated.

Thank you very much for your time and for maintaining this excellent tool.

Best regards,
Na Wan

[bumblenick]

Probably this is OK. I clip Z scores to +- 100
I recommend downsampling to no more than 1-2M snps.

[aaannaw]

Hi，

Thank you very much for your reply and suggestions.

We have already completed the analysis using all available SNP data. However, I noticed that our D values are much smaller than those reported in other studies — most of ours are on the order of 10⁻⁴, whereas other papers often report values around 10⁻² or 10⁻¹.

Could you please let me know why this might be the case? Is it still reasonable to trust the results I obtained?

Thank you again for your help and advice.

Best regards,
Na Wan