Topic 7.9 Notes – Carrying Out a Test for the Difference of Two Population Means

The Two-Sample t-Test for μ₁ − μ₂

This test is used when you have:

• Two independent groups
  • Independent random samples
  • or two groups from a randomized experiment
• A quantitative response variable
• A claim about μ₁ − μ₂ (usually whether it equals 0)

The big question is:

Is the observed difference in sample means large enough, relative to natural variability, to suggest the population means are different?

If $H_0$ is true and $μ_1 = μ_2$, then any difference we see in $\bar{x}_1 - \bar{x}_2$ should just be random sampling variability.

Hypotheses in Context

You must define the parameters clearly.

• $H_0: μ_1 - μ_2 = 0$
• $H_a: μ_1 - μ_2 \neq 0$, $> 0$, or $< 0$

Example setup:

Suppose we compare mean exam scores of students who studied with flashcards vs. those who used practice tests.

• $μ_1$: true mean exam score for students using flashcards
• $μ_2$: true mean exam score for students using practice tests

Then:

• $H_0: μ_1 - μ_2 = 0$
• $H_a: μ_1 - μ_2 \neq 0$

Be precise. On FRQs, vague definitions of μ₁ and μ₂ lose points fast.

The Test Statistic

Here is the formula from the formula sheet:

$$t = \frac{(\bar{x}_1 - \bar{x}_2) - (μ_1 - μ_2)}{\sqrt{\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}}}$$

Under the null (difference = 0), this simplifies to:

$$t = \frac{\bar{x}_1 - \bar{x}_2}{\sqrt{\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}}}$$

Break it down:

• Numerator → observed difference − expected difference (usually 0)
• Denominator → standard error of the difference

$$\text{SE} = \sqrt{\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}}$$

So the t-statistic measures:

How many standard errors the observed difference is from 0.

Large $|t|$ means stronger evidence against $H_0$.

Why a t-Distribution

The sampling distribution of

$$\frac{(\bar{x}_1 - \bar{x}_2) - (μ_1 - μ_2)}{\sqrt{\frac{s_1^2}{n_1} + \frac{s_2^2}{n_2}}}$$

is approximately t-distributed, assuming:

• Independent random samples or random assignment
• Nearly normal populations or large samples

The t-model accounts for extra uncertainty because we use sample standard deviations $s_1, s_2$, not population σ’s.

Degrees of Freedom

Technology gives the exact df. On your calculator (2-SampTTest), it’s reported automatically.

The df will fall between:

• smaller of $n_1 - 1$ and $n_2 - 1$
• and $n_1 + n_2 - 2$

If you must approximate by hand on the AP exam:

$$\text{df} = \text{smaller sample size} - 1$$

Always report:

• t statistic
• df
• p-value

Students often forget df on FRQs. That costs points.

Visualizing the p-Value

Suppose you compute $t = 2.4$ for a two-sided test.

Here’s what that means visually. The shaded regions in both tails beyond $-2.4$ and $2.4$ represent the p-value:

Two-sided p-value on a t-distribution (df = 10)

The p-value is the probability, assuming $H_0$ is true, of getting a difference in sample means at least as extreme as the one observed.

That phrase “assuming $H_0$ is true” must appear in your interpretation.

Correct interpretation example:

If the true mean exam scores for the two study methods are equal, the probability of observing a difference in sample means as large as the one found (or larger) is 0.018.

It is not the probability that $H_0$ is true.

Decision and Conclusion

Formal decision rule:

• If $p \le α$ → Reject $H_0$
• If $p > α$ → Fail to reject $H_0$

Never say “accept $H_0$.”

Full conclusion structure:

1. Compare p to α
2. State reject/fail to reject
3. State evidence about $H_a$
4. Refer to population means in context

Example ending:

Since p = 0.018 is less than α = 0.05, we reject $H_0$. There is convincing evidence that the true mean exam scores differ between students who used flashcards and those who used practice tests.

Notice it’s about population means, not sample means.

How This Supports a Research Question

The result of the test becomes your statistical reasoning.

If the research question was:

Do the two study methods lead to different average exam scores?

Your test provides the evidence to answer that question about the populations those students represent.

You are never proving a difference. You are assessing whether the sample provides convincing evidence of one.