For a paired-data test of the difference of means, the usual null hypothesis is that the mean of the paired differences is zero.

In symbols, if did_idi​ is the difference for each pair (for example, di=afteri−beforeid_i=\text{after}_i-\text{before}_idi​=afteri​−beforei​), then:

  • Null hypothesis: H0:μd=0H_0:\mu_d =0H0​:μd​=0 (on average, there is no difference between the two measurements in the population).
  • Alternative hypothesis (two-sided): Ha:μd≠0H_a:\mu_d \neq 0Ha​:μd​=0, or one-sided (μd>0\mu_d >0μd​>0 or μd<0\mu_d <0μd​<0) depending on the question.

In words: under the null, the “before” and “after” (or the two paired conditions) do not differ systematically; any observed difference is just random variation.

Quick Scoop on Paired Mean Tests

When you run a paired t-test, you are not directly comparing the two raw means; you are comparing the average of their **differences**.
  • Each subject (or unit) gives you a pair of measurements.
  • You compute a difference for each pair.
  • The test then asks: “Is the average of these differences zero in the population, or not?”

Think of a “before-and-after” weight-loss study: if there is truly no treatment effect, then the average (after − before) across all people should be zero.

So the core answer to:

“When testing the difference of means for paired data, what is the null hypothesis?”

is:

H0:μd=0H_0:\mu_d =0H0​:μd​=0, meaning the population mean difference between paired observations is zero.

TL;DR: For paired data, the null hypothesis is that the mean of the paired differences equals zero (no true change or effect).

Information gathered from public forums or data available on the internet and portrayed here.