Outliers from entry errors

• A key was pressed twice, or the decimal point was typed one place off.
• A value was entered in the wrong field, such as a postal code in a quantity column.
• A unit was assumed, so a figure in cents was typed into a column of dollars.

• A single extreme value moves an average or a total far more than hundreds of ordinary rows.
• Charts rescale to fit the outlier and flatten everything else.
• Removing extremes automatically also removes real ones, which are often the rows that matter most.

• Sort the column and read both ends.
• Test against limits that come from the real world: hours in a week, a plausible human age, stock that cannot go below zero.
• Compare each value with others for the same person, product, or site. A value ten times its neighbors is worth a look.

• Flag first. Deciding that a value is an error is separate from deciding what it should have been.
• Correct only with evidence: the source document, a related field, or the person who entered it.
• Where no evidence exists, keep the value, mark it, and decide openly whether to include it in each analysis.
• Keep a record of every value that was changed and why.

• Mixed unitsValues in one column are measured in different units, sometimes with the unit written in the cell and sometimes not.
• Disguised missing valuesA cell is not empty, but what it holds means there is no value: N/A, a dash, 999, or the word none.
• Impossible datesA value is shaped like a date but names a day that does not exist or could not have happened.

TermsOutlier, Anomaly, Validation, Human review