Use resistant summaries, robust regression, trimming, winsorization, and influence functions when contamination or heavy tails threaten conventional methods. Compare robust results with the original analysis transparently.
