1-Variance Test (chi-square test for the standard deviation)

Even with the mean on target, a large spread creates defects. Test whether the standard deviation (or variance) of the measurements differs from a tolerance value with the chi-square test, and see the spread with a confidence interval and 4 charts.

1Data Table

Paste a table copied from Excel into a cell (Ctrl+V). Click a column header (A, B…) before pasting to use the first line as column names. Double-click a header to rename it, right-click to insert/delete rows and columns, and formulas such as =A1*2 work too.

21-Variance Test Options

Reference value input

How to run a 1-variance test

  1. Copy the measurement column to test together with its header and paste it.
  2. Select the measurement column, choose whether the reference is a standard deviation or a variance and enter it. To check whether the spread is larger than the reference, set the alternative hypothesis to 'Variance > reference'.
  3. Press Analyze and check the p-value, the confidence interval and the interpretation in the results table.

Worked example with sample data

The sample data are 25 length measurements (mm) of the same part, and the allowed standard deviation is 0.05 mm. Loading the sample data sets the reference to a standard deviation of 0.05.

Chi-square rejection region from the 1-variance test of 25 part lengths (24 DF, observed χ² = 43.4)
Chi-square rejection region from the 1-variance test of 25 part lengths (24 DF, observed χ² = 43.4)

The sample standard deviation is 0.06724 mm (variance 0.004521), above the 0.05 limit. The chi-square statistic χ² = 24 × 0.004521 / 0.05² = 43.4 with 24 degrees of freedom gives a two-sided p-value of 0.01789, below 0.05, so the spread differs from (is larger than) the reference.

The 95% confidence interval of the standard deviation is 0.0525 to 0.09354 mm, so even its lower end is above 0.05. In the chi-square rejection region chart, the observed χ² = 43.402 falls inside the red area on the right.

The mean is 12.01 mm, so only the spread is the problem. The dot plot and histogram show one outlier by the IQR rule; if it is not a measurement error, look for causes that increase the spread, such as machine vibration or fixture condition.

How it is calculated

The test statistic is χ² = (n-1)s²/σ0², and the p-value comes from the chi-square distribution with n-1 degrees of freedom (s² is the sample variance, σ0² the reference variance). The two-sided p-value is 2×min(lower, upper tail probability).

The confidence interval divides (n-1)s² by chi-square quantiles; when the reference is entered as a standard deviation, the interval is shown in standard deviation units. For a one-sided test the other end is 0 or +∞.

What the result shows

A results table, the standard deviation with its confidence interval (reference as a dashed line), a dot plot and a histogram with the mean line, and the rejection region on the chi-square distribution with the observed χ².

Caution

The chi-square variance test assumes normally distributed data and is much more sensitive to non-normality than tests of the mean. Interpret the result with care if the histogram is strongly skewed or there are outliers.

How to read the 1-variance test result

  • If the p-value is below the significance level, the actual spread differs from the reference. Check whether the sample standard deviation is above or below the reference.
  • If the confidence interval of the standard deviation does not contain the reference, the difference is significant. The upper end shows how large the spread could be.
  • In the rejection region chart, the null hypothesis is rejected when the observed χ² falls in the red area.

Frequently asked questions

Should I enter the reference as a standard deviation or a variance?

Whichever is convenient. The test result (p-value) is the same; only the units of the confidence interval and the interpretation change.

How is this different from process capability?

Cp and Cpk compare the spread with the specification width, while the 1-variance test checks whether the spread is statistically different from a specific value.

How do I compare the spread of two groups?

Use the 2-variance test, or the test for equal variances for 3 or more groups.

Other chart tools