Kirk Borne

I don’t know if it’s the most common case, but this is one I was most concerned with when I saw it —> using imputation simply to fill in the gaps in time series data, without addressing the real questions as to why do those gaps even exist, and what other data sources are available during those same time gaps that might elucidate what else might have been happening in those missing periods. Thought experiment — for example — think of a case where there is a gap in a sequence of surveillance camera images that “coincidentally” occurred when there was other suspicious activity reported at that same time in the local neighborhood. Another case I see is in customer journey modeling where gaps are filled in through data imputation (i.e., with assumptions) not with facts. That’s just my own experience, not a full review of the topic.