I am wondering if someone could tell me what PLINK uses as the reference allele in a GWAS based on linear regression. Some documentation on the site leads me to think it is consistently modeling A2 as the reference. In my data, A2 is not always the major allele. I am finding some examples when the results are opposite of what you would expect (e.g., major allele increases risk; minor allele decreases risk). I realize this is possible, though not common, and would love to make sure I am not missing something. Thanks.
By default, PLINK forces the major allele to A2 (unless you specify --keep-allele-order), and then its association analysis commands report effects of A1 alleles.
If you use --keep-allele-order and identify incorrect results for some major A1 alleles, let me know and I'll try to fix the issue in future PLINK 1.9 builds.