Real datasets are messy. Some cells are empty (missing values) and some values sit far from the rest (outliers). Notice both before you analyse.
Why it matters
Missing values shrink your sample and can bias results. Outliers drag the mean and inflate the SD.
Key ideas
- Report how many values are missing, and in which variables
- Ask why they are missing — random or systematic?
- Outlier rule of thumb: below Q1 − 1.5×IQR or above Q3 + 1.5×IQR
- First check for typing errors, then decide and report your decision
- Never delete a value just because it is inconvenient
- Median and IQR resist outliers; mean and SD do not
Real-life research example
In a BMI survey one participant has BMI = 240. That is a typing error for 24.0. Correcting the entry is right; silently dropping the person is not.
When to use
- Screen for missing values and outliers before every analysis
- Run a sensitivity check: results with and without the suspicious value
When NOT to use
- Don't fill missing values with the mean without justification
- Don't remove outliers to obtain a significant p-value
Mini example
Values: 4, 5, 5, 6, 7, 8, 40 Q1 = 5, Q3 = 7.5, IQR = 2.5 → upper fence = 7.5 + 3.75 = 11.25 So 40 is an outlier. Mean = 10.7 but median = 6.
Quick takeaway
Look first, then clean. Every cleaning decision must be reported.
Try it yourself
Type your own numbers below and add one extreme value. Which value is flagged? How far does the mean move compared with the median?
Model answer
The extreme value pulls the mean toward itself while the median barely moves. Report both, explain your decision about the value, and show a sensitivity check.