Topic 4.10 Notes – Carrying Out a Test for the Difference Between Two Population Means

What a Two-Sample t-Test for a Difference in Means Does

This test compares two independent groups. The groups might be two populations from random samples, or two treatments from a randomized experiment.

• The variable must be quantitative.
• The parameter you care about is the difference in population means, $\mu_1 - \mu_2$.
• The sample statistic is the difference in sample means, $\bar{x}_1 - \bar{x}_2$.

A usual null hypothesis says there is no difference:

• $H_0:\mu_1-\mu_2=0$
• same idea as $H_0:\mu_1=\mu_2$

The alternative depends on the question:

• $H_a:\mu_1-\mu_2\ne 0$
• $H_a:\mu_1-\mu_2>0$
• $H_a:\mu_1-\mu_2<0$

Because the population standard deviations are unknown, this uses sample standard deviations $s_1$ and $s_2$, so it is a t-test.

One fast warning that shows up all the time on tests: this is only for independent groups. If the data are paired, like before/after or twins, you turn each pair into a difference and do a one-sample t-test on those differences.

Also, the subtraction order must stay fixed. If you define $\mu_1-\mu_2$, every statistic and conclusion must match that order.

When the Test Is Appropriate

Both groups have to meet the conditions. One solid group does not rescue a bad one.

Randomization

You need either:

• two independent random samples, or
• a randomized experiment with random assignment to treatments

On the AP exam, saying “the samples are independent” is too vague by itself. Say how the data were collected.

Independence

For random samples taken without replacement, check the 10% condition for both groups:

• $n_1 \le 0.10N_1$
• $n_2 \le 0.10N_2$

That condition is not needed in a randomized experiment.

Sample data condition

The t-procedure is okay if:

• both populations are approximately normal, or
• both sample sizes are at least 30, or
• with small samples, both groups show no strong skewness and no outliers on separate graphs

That is why side-by-side boxplots are so helpful here. You are checking each group for overall shape and any clear outliers before using the test.

Boxplots showing skewness and outliers

Carrying Out the Test

A full test has a standard flow.

1. Define the parameters in context
   Example: $\mu_1$ = the mean battery life for all phones using Brand A, and $\mu_2$ = the mean battery life for all phones using Brand B.

2. Write hypotheses using population means, never sample means.

3. Name the procedure
   Two-sample t-test for a difference between population means.

4. Compute the standard error

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

   The variances add because the samples are independent. This is unpooled, so on a calculator use pooled = No.

5. Compute the test statistic

   $$t=\frac{(\bar{x}_1-\bar{x}_2)-0}{\sqrt{\frac{s_1^2}{n_1}+\frac{s_2^2}{n_2}}}$$

   Example with $\bar{x}_1=18.6, s_1=4.2, n_1=24$ and $\bar{x}_2=15.1, s_2=3.6, n_2=21$:

   $$SE=\sqrt{\frac{4.2^2}{24}+\frac{3.6^2}{21}}\approx 1.163$$

   $$t=\frac{18.6-15.1}{1.163}\approx 3.01$$

   A positive $t$ means group 1’s sample mean is above group 2’s.

6. Get df from technology
   In AP Stats, df is usually not $n_1+n_2-2$.

7. Find the p-value from the correct tail(s).

   This depends on your alternative hypothesis. A one-sided test puts all the area in one tail. A two-sided test splits the area between both tails, like the comparison shown here.

One-tailed vs. two-tailed t-tests

Interpreting the p-Value and Making the Decision

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

In context, that means assuming the two population means are equal.

If $p\text{-value} \le \alpha$, reject $H_0$.
If $p\text{-value} > \alpha$, fail to reject $H_0$.

Your conclusion must:

• mention population means
• match the direction of $H_a$
• use AP wording like “convincing statistical evidence”

Example conclusion:
“There is convincing statistical evidence that the population mean battery life for phones using Brand A differs from the population mean battery life for phones using Brand B.”

Random samples let you generalize to populations. Random assignment lets you talk about cause and effect. Statistical significance does not automatically mean the difference is important in real life.

Common Mistakes to Catch Fast

• Using this test for matched pairs data
• Switching from $\mu_1-\mu_2$ to $\bar{x}_2-\bar{x}_1$
• Writing hypotheses with $\bar{x}_1$ and $\bar{x}_2$ instead of $\mu_1$ and $\mu_2$
• Using pooled procedures or automatically using $df=n_1+n_2-2$
• Using the wrong standard error formula
• Picking the wrong tail for the p-value
• Saying “accept $H_0$”
• Saying a nonsignificant result proves the means are equal
• Writing a conclusion about the samples instead of the population means