Thresholds for homozygous genotypes using NGS

[andtill]

When performing variant calling from NGS data, one commonly use (for example) 0.2-0.8 for calling a heterozygous and 0-0.2 or 0.8-1 for calling a homozygous genotype. Allelic imbalance for heterozygous genotypes I understand. But why do one get homozygous calls that are not (or nearly) 0 or 1? We have run some dilution series of DNA with known genotypes and observe homozygous calls having balance value of 0 or 1 when having enough DNA and values like ~0.1 or ~0.9 when having little amount of input DNA.

Any thoughts why "wrong" alleles are amplified and sequenced when having small amount of DNA?

[gringer]

Shot noise becomes more of an issue at low coverage. When the number of template sequences to sample (or amplify) is small, each individual sequence contributes more to the estimated allele frequency.