Presence of too many rare alleles in germplasm populations will cause false positive results in Association Mapping analysis using SSR markers and make it less accurate. Opinion needed.
I think, for GWAS it is not recommended to use the SSR markers. First of all because their frequency is significantly less in the genome compared to SNPs. Locating the markers that are in LD with the trait of your interest will be feasible usually with SNP markers. However, during association analysis, I believe we consider the minor allele frequency into account to address your question of rare alleles. Hope this answer partially helps.
The major problem with using SSR markers (or any other marker type) for association mapping with germplasm populations is that the germplasm populations themselves are seldom genetically uniform, and you need to know something about allelic frequencies in order to accurately assess genetic diversity or perform accurate association mapping. Therefore, you need to genotype a representative sample of the population (or heroically and probably incorrectly assume that the single plant you genotype is representative of the population) and estimate allelic frequencies before you can use such marker data for association mapping. SNPs also have rare alleles -- and these contribute essentially nothing to the association analysis (or a diversity analysis) because of their rareness -- unless you are working with very large populations that have been very precisely phenotyped (seldom affordable in plant breeding, although HTP methods are getting more affordable). The major limitation with SSRs is that there are smaller numbers of them so they do not provide as comprehensive of coverage as SNPs. However, SSR loci are more likely to be multi-allelic than SNPs (and even perhaps SNP haplotypes), in the genomic regions where they are present they can be quite effectvely used for association mapping. My personal preference would be to use some combination of genomic-SSRs, genomic-indels, and SNP-haplotypes for association mapping -- and with appropriate bioinformatics your may well be able to sample all of these with a genotyping-by-sequencing approach.
SSR is also suitable for association mapping,some paper abbout association mapping use SSR to analysis, especially in the species which haven't been sequencing
SSR markers are discrete. For association mapping, you are finding out the association of specific traits being figured out by microsatellite markers (SSR). It is better and wiser to use SSR markers as it caters for rare alleles as well. The bonus is you have a true revelation of both dominant and rare alleles for assessment. The minor disadvantage is what you have pointed out: 'the expression of rare alleles' which might influence 'false positive'. You'll find in most literatures that association mapping is done using SSR markers and is widely published and accepted.