Why Is Your Variance So Large? Understanding What the Number Really Means

By Published Last updated

Bar chart of 40, 42, 41, 43, 44 and one outlier of 90, with dashed lines at the mean with the outlier (50) and without it (42).

You ran the calculation twice and the variance still looks alarmingly big. Before assuming you made a mistake, it is worth knowing that a large variance often reflects something real in the data. This guide covers the usual causes and how to investigate them.

Worked Example: One Outlier

Five readings: 40, 42, 41, 43, 44.

  • The mean is 210 / 5 = 42.
  • Deviations: −2, 0, −1, 1, 2. Squared: 4, 0, 1, 1, 4. The sum of squares is 10.
  • Sample variance: 10 / 4 = 2.5.

Now add a sixth reading of 90, so the data is 40, 42, 41, 43, 44, 90.

  • The mean is 300 / 6 = 50.
  • Deviations: −10, −8, −9, −7, −6, 40. Squared: 100, 64, 81, 49, 36, 1600. The sum of squares is 1930.
  • Sample variance: 1930 / 5 = 386.

One value raised the variance from 2.5 to 386, more than 150 times as large. The single squared deviation of 1600 is 83% of the total.

Common Causes

Extreme Values

As the example shows, values far from the mean contribute disproportionately, because the squaring step turns a large distance into a very large number.

A Genuinely Wide Range

If the data naturally spans a wide range, from very small to very large values, the variance will be large. That is not an error; it is what the data looks like.

Mixed Groups

Combining distinct groups, such as measurements from two different machines or from adults and children, inflates the variance compared with the variance inside each group. If the data has two clusters, analyse them separately.

Large Units

Variance is in squared units, so its size depends heavily on scale. Measurements in grams have a variance one million times larger than the same measurements in kilograms. A big number may simply reflect small units.

Distance Counts Quadratically

A value 10 units from the mean contributes 100 to the sum of squares; a value 40 units away contributes 1600. Doubling the distance quadruples the contribution. This is why a few moderate outliers can outweigh dozens of values close to the mean.

How to Investigate

  1. Sort the data and look at the smallest and largest values.
  2. Check any unusual values against the original records. Is 90 a real reading, or a typo for 39?
  3. Look for two or more clusters that might be different groups.
  4. Recalculate with and without the suspicious values and report both if you cannot decide.

Comparing Variances Across Different Units

Do not compare raw variances between data sets measured in different units or on very different scales. Converting to standard deviation does not fix that either, because the standard deviation keeps the units. Use the coefficient of variation (standard deviation divided by the mean) instead.

Use the Standard Deviation to Get a Feel for It

The square root of the variance is in the original units and is easier to judge. For the six readings above, √386 ≈ 19.65, so the readings typically sit about 20 units from their mean of 50. Seeing that next to the data makes it obvious that the 90 is doing most of the work.

Summary

A large variance usually points to something real: an outlier, a wide natural range, mixed groups or simply large units. Look at the individual values rather than just the final number. The variance calculator shows each squared deviation, so you can see at a glance which values contribute most.

Back to Blog