I am using triad.census to conduct a triad.census on vast networks constituted by references. The census is done for time/area subsamples of the whole reference networks so that I end up running 900 different triad.censuses. The number
of nodes in each of the sub-networks can get up to 20,000 nodes.
In the results I get from this exercise, the count of the 003-triads (null triads) are often negative. I am not sure what causes this negative count. In principle, everything seems to be right as I cross-checked a small samples of the 900
censuses I did with results from Pajek – and the number are equivalent except the negative 003-triad counts.
Is there any explanation to this phenomenon? Any hint is welcome!
Stefan