Mild departures from the model can lead to overestimating $K$

When there is real population structure, this leads to LD among unlinked loci and departures from Hardy-Weinberg proportions. Roughly speaking, this is the signal that is used by the structure algorithm. But some departures from the model can also lead to Hardy-Weinberg or linkage disequilibrium. These include inbreeding, and genotyping errors such as occasional, undetected, null alleles. Even in the absence of population structure, these types of factors can lead to a weak statistical signal for $K>1$. Beginning in Version 2, we have suggested that the correlated allele frequency model should be used as a default because it often achieves better performance on difficult problems, but the user should be aware that this may make it easier to overestimate $K$ in such settings than under the independent frequencies model. The next subsection discusses how to decide whether inferred structure is real.