Table 5.

Major assumptions of and questions relating to model-based clustering and their implications in population genetic studies using microsatellites

Method	Assumption or question	References	Related issues	References
In general	Individual runs are stochastic and may settle on local optima	Gilbert et al. (2012)	Ensure that a sufficient number of steps and runs have been performed	Gilbert et al. (2012)
BAPS and STRUCTURE	Hardy–Weinberg equilibrium within populations	Pritchard et al. (2000), Corander et al. (2004)	No inbreeding; if suspected, use InStruct	Gao et al. (2007)
			Individuals are not related by direct descent; related individuals should be removed prior to analysis	Anderson and Dunham (2008), Rodríguez-Ramilo and Wang (2012)
BAPS	Gametic linkage equilibrium within populations. Tight (BAPS) or loose (STRUCTURE) linkage allowed for one model only	Falush et al. (2003), Corander and Tang (2007)	Use of linkage model requires data be haploid, or phased data from a diploid or tetraploid	Corander and Tang (2007)
STRUCTURE			Sufficient number of markers should be unlinked. Phasing optional for diploids, required for polyploids	Falush et al. (2003), Kaeuffer et al. (2007)
BAPS and STRUCTURE	Is the information content of the dataset sufficiently high?	–	Population structure: Incomplete lineage sorting may confound inference when using as many 50 microsatellite loci	Orozco-terWengel et al. (2011)
			Admixture: If few microsatellite loci are used, the sample should include a significant number of pure individuals	Pritchard et al. (2000), Vaughan et al. (2009)
STRUCTURE and NewHybrids			Hybrid identification: reliable only for many loci (>24–50), especially when differentiation is low	Vähä and Primmer (2006), Fitzpatrick (2012)
STRUCTURE	Recessive allele model: null allelesare due to polymorphism	Falush et al. (2007)	Not appropriate for data that are missing due to experimental error	Falush et al. (2007)
STRUCTURE	Prior population1 and correlated allele frequency models allow detection of weak structure	Falush et al. (2003), Hubisz et al. (2009)	Using standardized values, performance of STRUCTURE and BAPS declines at standardized FST values of 0.28. 97% accuracy was attained only at FST = 0.39	Latch et al. (2006)
BAPS	Two models that incorporate population information a priori are available	Corander et al. (2006, 2003)

BAPS, Bayesian analysis of population structure.

¹The two prior population models, LOCPRIOR and USEPOPINFO, should be used for weak and strong structure, respectively.