An order with a quantity of 50,000 in a database where typical orders are 1–100. A customer with a credit score of 0 in a range of 300–850. A date of 1901 in a field that should contain recent transactions. These are outliers — and some of them are data errors masquerading as valid values.
An outlier is a value that deviates significantly from the rest of the values in a column. In the context of data quality, outliers are important because they're often signals of data entry errors, system glitches, or import problems — not legitimate extreme values.
Types of Outliers in Data Quality
Entry errors — A quantity of 10000 where the user meant 100. An age of 200 where they entered two fields in the wrong order. These are clearly wrong.
System defaults masquerading as values — A birth date of "01/01/1900" is often a system default for "unknown date," not an actual customer born in 1900.
Unit mismatch errors — Revenue entered in thousands rather than actual values, causing some records to appear 1,000x larger than others.
Sohovi profiles your datasets for quality issues in minutes — see what's broken before it breaks your pipeline — try Sohovi free.
Legitimate extreme values — Some outliers are real: a customer who genuinely placed a $500,000 order, or an employee who genuinely has been with the company for 40 years. Context determines whether an outlier is an error or a fact.
How to Detect Outliers Without Code
Sort the column and inspect the extremes — Sort ascending and descending to see the top and bottom values. Are the extremes plausible? This is the fastest manual method.
Look at min and max values — When profiling a column, the minimum and maximum values immediately reveal whether the range is plausible for the field's purpose.
Check value distribution — A histogram of values shows whether data clusters normally or has unexpected spikes at specific values (like 0 or 9999) that might indicate defaults or placeholders.
Apply business rules — Define plausible ranges for key numeric fields and flag any records outside those ranges. Age between 0–120. Revenue greater than 0. Order date not in the future.
Sohovi shows min/max values and value distribution for every numeric and date column when you profile a CSV — instantly surfacing outliers without writing any code or formulas.
What to Do With Outliers
Investigate before removing. An outlier might be:
- A data entry error (fix it)
- A legitimate value (keep it)
- A system default (null it)
- An indicator of a process problem (trace and fix the source)
Never automatically delete outliers. Instead, flag them, investigate the source, and document your decision.
