A dataset may contain accurate recorded values and still produce a misleading trend if the missing observations are not random. What disappears can matter as much as what remains.
Suppose slower cases are increasingly left unreported while faster cases continue to be recorded. The observed average will improve even if the underlying process stays the same. No existing value was altered; the composition of the visible sample changed.
Blank fields, delayed reports and dropped participants should be counted and described by period and group. A stable overall missing rate can still hide a problem if the gaps become concentrated among people with different outcomes.
Analysts may use imputation or sensitivity tests to explore plausible values, but these methods depend on assumptions. Results should show how the conclusion changes under reasonable alternatives rather than presenting one filled-in series as certain.
Before interpreting a trend, ask how many observations were expected, how many arrived and why the gaps occurred. A line based only on available records describes the observed sample, not automatically the population that should have been measured.
Original educational commentary by Breaking Click. This explainer is designed to help readers evaluate information more carefully.



