These techniques are quite sensitive to the distribution of the data under study. For instance, if the numbers being processed by the LogLog counter aren't uni

These techniques are quite sensitive to the distribution of the data under study. For instance, if the numbers being processed by the LogLog counter aren't uniformly distributed, the cardinality estimates you get from it could be arbitrarily biased.