molecalculator

Moles & Molar Mass

Normality Test for Paired t Test: How to Check

Calculator No. 01-8646 · Updated · Formula and assumptions shown below

The calculator on this page is switched off while it is checked. The formula is below, and the worked example shows how to do the sum by hand.

For study and lab planning. The result follows from the numbers you enter and the formula shown. Check it against your method before you weigh, mix or dilute anything, follow the safety data sheet and your supervisor, and do not use it for clinical, dosing or medication work.

The paired t test assumes that the differences between paired observations are approximately normally distributed. You check this with a normality test applied to the difference scores, not to the raw data of each group separately.

Formula

The paired t test statistic is calculated as:

t = (mean difference) / (standard deviation of differences / √n)

where:

  • mean difference (d) is the average of the paired differences (each pair subtracted one way), in the same units as the measurements
  • standard deviation of differences (s) is the sample standard deviation of those differences, in the same units
  • n is the number of pairs (not the total number of observations)
  • degrees of freedom (df) = n − 1

The p-value is obtained from the t-distribution with n − 1 degrees of freedom. The calculator above computes the t statistic and the two-tailed p-value. It does not perform the normality test itself; it helps you see the effect of sample size and variability on the test result.

Worked Example

Suppose you measure blood pressure before and after a 4-week exercise program in 20 participants. The mean of the before-minus-after differences is 1.5 mmHg, and the standard deviation of those differences is 2.0 mmHg.

Step 1: Compute the standard error of the mean difference:

SE = s / √n = 2.0 / √20 = 2.0 / 4.4721 = 0.4472

Step 2: Compute the t statistic:

t = d / SE = 1.5 / 0.4472 = 3.3541

Step 3: Degrees of freedom = n − 1 = 19.

Step 4: For a two-tailed test, the p-value is the probability of getting a t value at least as extreme as 3.3541 with 19 degrees of freedom. Using a t-table or software, p ≈ 0.0033.

Since p < 0.05, you would reject the null hypothesis of no mean difference. The calculator gives the same t and p for the default inputs.

Assumptions

  • The differences are independent of each other (each pair is independent).
  • The differences are approximately normally distributed. This matters most for small samples (n < 30).
  • The scale of measurement is at least interval (equal intervals).
  • The test is two-tailed unless specified otherwise.

Why Check Normality on the Differences

The paired t test is mathematically equivalent to a one-sample t test on the difference scores. The normality assumption applies to those differences, not to the original two variables. For example, if you measure reaction time before and after caffeine, the raw reaction times might be skewed, but the differences (after − before) could still be bell-shaped. Conversely, the raw data could look normal while the differences are skewed. Always compute the difference for each pair first, then examine that single list of numbers.

This distinction is a common source of confusion. Many textbooks and software packages automatically compute the differences when you run a paired t test, and the normality check they offer is on those differences. If you test the raw data of each group separately, you are checking the wrong assumption.

How to Perform a Normality Test

There are several ways to assess normality of the differences. The most common formal tests are:

  • Shapiro–Wilk test: powerful for small to moderate samples (n < 50). It tests the null hypothesis that the data come from a normal distribution.
  • Kolmogorov–Smirnov test: less powerful than Shapiro–Wilk, and it requires specifying the mean and standard deviation. Not recommended for small samples.
  • Anderson–Darling test: a modification of Kolmogorov–Smirnov that gives more weight to the tails. Often used in quality control.
  • D’Agostino–Pearson test: based on skewness and kurtosis. Works well for moderate to large samples.

In practice, most statistical software (R, SPSS, Minitab, Python’s scipy) offers the Shapiro–Wilk test by default. For a quick visual check, you can also create a histogram, a boxplot, or a Q–Q plot of the differences. A Q–Q plot that roughly follows a straight diagonal line suggests normality.

Sample Size and the Central Limit Theorem

Normality of the differences matters most when the sample size is small. The paired t test is based on the t distribution, which assumes the sample mean of the differences is normally distributed. Thanks to the Central Limit Theorem, the sampling distribution of the mean approaches normality as n increases, even if the underlying data are not normal. For n ≥ 30, the t test is generally robust to moderate violations of normality. For n < 15, the test can be unreliable if the differences are strongly skewed or have outliers.

If your sample is small and the differences are clearly non-normal, you have two options:

  • Use the Wilcoxon signed-rank test, a nonparametric alternative to the paired t test. It tests whether the median difference is zero, and it does not require normality.
  • Use a permutation test or bootstrap paired test, which make fewer assumptions about the distribution.

These alternatives are less powerful than the t test when the data are truly normal, but they are safer when normality fails.

Common Mistakes and Misconceptions

  • Testing normality on each group separately: as noted, the assumption is about the differences, not the raw data.
  • Ignoring outliers: a single extreme difference can make the normality test fail. Investigate outliers before deciding whether to transform or use a nonparametric test.
  • Using normality tests as a pass/fail gate: formal tests have low power with small samples and high power with large samples. With large n, a significant Shapiro–Wilk test may detect a trivial deviation that does not affect the t test. Always combine the test with a visual check.
  • Forgetting that the t test is robust: moderate skewness or kurtosis does not invalidate the test, especially with n > 30. The test is more sensitive to outliers than to mild non-normality.
  • Transforming the data unnecessarily: if the differences are skewed, a log or square-root transformation might help, but it changes the interpretation. The t test on transformed differences tests the mean of the transformed differences, not the original mean.

What to Do If Normality Fails

If the normality test rejects the null hypothesis (p < 0.05), first look at the Q–Q plot and histogram. If the deviation is mild and n is moderate (say 20–30), the t test may still be valid. If the deviation is severe (e.g., heavy tails, strong skew, or clear outliers), consider:

  • Removing or correcting data entry errors, not outliers that are legitimate.
  • Applying a transformation such as log, square root, or Box–Cox to the differences.
  • Switching to the Wilcoxon signed-rank test.
  • Using a bootstrap paired t test, which does not assume normality.

Document your choice in your analysis report. Many journals expect you to state how you checked the normality assumption and what you did if it was violated.

Frequently Asked Questions

Do I need to test normality for a paired t test if my sample size is large?
For n ≥ 30, the t test is quite robust to non-normality due to the Central Limit Theorem. You can still run a normality test, but a significant result with a large sample often indicates a small deviation that won’t change your conclusion. Focus more on outliers and the visual plot.

What is the best normality test for small samples?
The Shapiro–Wilk test is generally recommended for small samples (n < 50) because it has good power. For very small samples (n < 10), no normality test is very reliable, so rely on domain knowledge and the Q–Q plot.

Can I use the paired t test if the differences are not normal?
Yes, if the sample size is large enough (n ≥ 30) and the deviation is not extreme. For small samples, the test may have inflated type I error or reduced power. In that case, use a nonparametric alternative like the Wilcoxon signed-rank test.

What is the difference between testing normality of raw data vs. differences?
The paired t test compares the mean difference to zero. The normality assumption applies to the population of differences. Testing each group separately can mislead you because the differences can be normal even if the raw data are not, and vice versa.

How do I check normality in Excel?
Excel does not have a built-in Shapiro–Wilk test, but you can create a histogram or use the Data Analysis Toolpak to get descriptive statistics. For a formal test, you would need to use a statistical add-in or another program like R or Python.

Is the Wilcoxon signed-rank test a good replacement?
Yes, it is the most common nonparametric alternative. It tests whether the median of the differences is zero, and it only requires that the differences are symmetric. It is less powerful than the t test when the data are normal, but it is safer when normality is violated.

Related

Formula used
in n | Sample size | - | 20 | 3
in d | Mean difference | - | 1.5 | 0
in s | SD of differences | - | 2.0 | 0.001
out t | t statistic | - | d / (s / sqrt(n)) | 4
out df | Degrees of freedom | - | n - 1 | 0
out p | p-value (two-tailed) | - | 2 * (1 - tcdf(abs(t), df)) | 4

Each line is one input (in, sel, list) or one result (out: label, unit, formula, decimals). How the calculators work

Method: the standard formula for this calculation with the constants and conversion factors written on the page, rounded as each result states. See method and constants, the disclaimer or report an error.

More in this section

The whole Moles & Molar Mass section →