> Estimates are that around 16% of children are abused, so if you're only finding 5% it means you're substantially worse at choosing households to investigate than if you were picking at random.Sampling iid would be unconstitutional and also obviously politically infeasible. So, the only people who are investigated are people for whom there is probable cause or at least an accusation.
Once you're sampling from a population for whom you have either probable cause or at least an accusation, the group under suspicion probably not an iid sample of the overall population and may even be skewed in the wrong direction. Eg, a physical abuser is more likely to conceal a bruise than the parent of a particularly careless child.
Additionally, suppose for sake of argument that abusers are in fact over-represented in the accused group. The behavior of both abusers and the abused confounds the statistical analysis. Consider alcoholism as an analogy (but also very correlated underlying pathology). Just like with alocholism, there are "high functioning" abusers who are able to avert suspicion and, in case of suspicion, manipulate the people who they are abusing.
So we are sampling from a constrained population in which abusers may in fact be under-represented, and then again within that group there are confounding issues.
None of this is to say that we should abandon constitutional rights, of course. But in a country with strong protections for the accused we might actually expect certain types of crimes to be more difficult to investigate and prosecute than a "just iid sample" approach.
(It's also possible that the estimates are wrong, and in both directions. It's additionally possible that the 16% statistic refers to a different definition of abuse than the 5% number. Etc. etc.)