Paired t Test (paired samples t-test)

When the same operators, parts or samples are measured twice, looking only at the paired differences removes the item-to-item variation, so even a small improvement can be detected. The mean difference is t-tested against a reference (usually 0) and 6 panels show the direction of change and how well the two measurements agree.

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.

2Paired t Test Options

How to run a paired t test

  1. Copy the before and after columns so that each item is on the same row.
  2. Select before as the value 1 column and after as the value 2 column. The difference is calculated as 'after - before', so check the order.
  3. Press Analyze and check the mean difference, the p-value and, in the before-after chart, how many pairs increased or decreased.

Worked example with sample data

The sample data are the times (seconds) 12 operators needed for the same task before and after a work method improvement. Each row is one operator; before is used as value 1 and after as value 2.

Before-after chart from the paired t test of 12 operators' task times
Before-after chart from the paired t test of 12 operators' task times

The mean difference (after - before) is -3.317 s, so the task time dropped by 3.3 s on average. With a standard deviation of the differences of 2.643, t = -4.347 (11 degrees of freedom) and the p-value is 0.001161, so the improvement is significant.

In the before-after chart, 11 of the 12 lines slope downward and only one operator got slower (+2.0 s). The 95% confidence interval of the mean difference, -4.996 to -1.637 s, does not contain 0 either.

Ignoring the pairing and analyzing the same data with a 2-sample (Welch) t test gives a p-value of 0.05344 - not significant. The skill differences between operators (57.4 to 67.8 s before the change) hide the improvement, which is why before/after data must be analyzed with the paired t test.

How it is calculated

For each row the difference d = after - before is calculated (rows where either value is blank are skipped), then a 1-sample t test with n-1 degrees of freedom checks whether the mean of d equals the reference difference. The result is the same as SciPy's ttest_rel.

Bland-Altman plot

The x axis is the mean of the two measurements and the y axis is their difference. The solid middle line is the mean difference (bias) and the dashed lines are the limits of agreement at the mean ± 1.96 standard deviations - useful for comparing two measurement methods or seeing whether the change depends on the size of the value.

Difference from the 2-sample t test

Analyzing the same data with a 2-sample t test mixes the item-to-item variation into the error and inflates the p-value. Paired data should be analyzed with the paired t test to see the effect properly.

How to read the paired t test result

  • A significant result means the before-after difference is statistically meaningful. Check the sign of the mean difference to see whether it is an increase or a decrease (a negative after - before is a decrease).
  • If most lines in the before-after chart slope the same way, the improvement appeared evenly across the items.
  • If the QQ plot of the differences shows clear non-normality, also check with the Wilcoxon signed-rank test.

Frequently asked questions

Is it the same as Excel's T.TEST type 1?

Yes. Excel's T.TEST(before, after, 2, 1) is the two-sided paired t test p-value and matches the result here.

What happens to rows that do not pair up?

Rows where either value is blank or not a number are dropped entirely; only complete pairs are used.

Can I reverse the direction of the difference?

Swap the value 1 and value 2 columns and the difference becomes 'before - after' with only the sign reversed. The two-sided p-value stays the same.

Other chart tools