Hypotheses for Comparing Two Population Proportions

From a Policy Question to Hypotheses

In Free Response Practice on Two-Proportion Intervals, you practiced keeping the group order clear when comparing two population proportions. A hypothesis test begins with a related comparison, but its first task is different: state the population claim that will be examined and the alternative claim that would count as evidence against it.

Suppose a fictional city is considering a policy that would add evening bus service. A survey asks residents in two districts whether they support the proposal. Before analyzing survey results, a researcher needs hypotheses about the population proportions of residents who support the policy. The hypotheses are not statements about the proportions observed in the survey; they describe the populations the survey is intended to represent.

The hypotheses must use the same two populations and the same definition of “success.” In this example, success means that a resident supports the proposed evening bus service. Keep the group order fixed: define Group 1 first and Group 2 second, then write the difference as Group 1 minus Group 2.

The Null Hypothesis: No Difference in the Populations

Let \(p_1\) be the true proportion of all residents in District East who support the proposed evening bus service. Let \(p_2\) be the true proportion of all residents in District West who support it. The null hypothesis says these population proportions are equal:

The same null hypothesis can be written using a difference:

These are equivalent statements: if the proportions are equal, subtracting one from the other gives zero. In words, the null says that the true proportion of District East residents who support the policy is the same as the true proportion of District West residents who support it. It does not say that either proportion is zero, or that nobody supports the policy.

The equality belongs in the null hypothesis for the standard two-proportion test. The null gives a specific no-difference model against which the observed sample results can later be assessed. In What a P-Value Really Measures, you learned that a p-value is calculated assuming the null hypothesis is true. For this comparison, that means using a model in which \(p_1=p_2\).

Three Alternatives: Any Difference or a Direction

The alternative hypothesis depends on the question being asked. There are three common choices. Select the one that matches the research question, not simply the one that matches the direction of the sample results.

The first alternative is two-sided: differences in either direction are relevant. The other two alternatives are one-sided: only a difference in the stated direction counts as evidence for the alternative. In difference form, these alternatives are \(p_1-p_2\ne0\), \(p_1-p_2>0\), and \(p_1-p_2<0\), respectively.

The words “different,” “higher,” and “lower” refer to the population proportions. If \(p_1-p_2>0\), District East has the higher population support proportion. If \(p_1-p_2<0\), District West has the higher one. Writing the difference form can make that direction especially easy to check.

When the Research Question Names a Direction

A directional alternative is appropriate when the research question specifically asks whether one population proportion is higher or lower than the other. The order of the parameters determines the inequality sign. With \(p_1\) for East and \(p_2\) for West, “East is higher” corresponds to \(p_1>p_2\). If you instead defined Group 1 as West, the same verbal claim would require \(p_1<p_2\).

Choose the direction from the research question or the claim being investigated, ideally before examining the sample results. Do not switch to a one-sided alternative just because the observed sample proportion in one group happens to be larger. The alternative identifies in advance which outcomes count as evidence against the null, a point developed in P-Values for One-Sided Versus Two-Sided Tests.

How Group Order Controls the Sign

The choice of Group 1 and Group 2 is flexible, but once you choose an order, maintain it in the parameter definitions, symbolic hypotheses, and written statements. This is the same discipline used when interpreting \(p_1-p_2\) in the earlier tutorials on comparing two proportions.

For example, if \(p_1\) is East and \(p_2\) is West, then “East has a higher proportion” is \(p_1>p_2\), or \(p_1-p_2>0\). If the definitions are reversed so \(p_1\) is West and \(p_2\) is East, the same claim becomes \(p_1<p_2\), or \(p_1-p_2<0\). The real-world claim has not changed; only the subtraction order has.

A useful self-check is to read the inequality aloud using the parameter definitions. For \(p_1<p_2\), say “the true proportion in Group 1 is less than the true proportion in Group 2.” If that sentence does not match the research question, revise the sign or the group order before continuing.

Common Mistakes and AP Exam Tips

• Writing hypotheses about the sample: Expressions involving \(\hat{p}_1\) and \(\hat{p}_2\) describe sample statistics. Hypotheses concern the population parameters \(p_1\) and \(p_2\).
• Leaving out the shared outcome: Define each proportion as the proportion with the same characteristic, such as supporting the same policy. A comparison is unclear if the two proportions measure different outcomes.
• Using the wrong sign for a direction: Check which group is \(p_1\). With East as Group 1, “East is higher” means \(p_1>p_2\), while “East is lower” means \(p_1<p_2\).
• Using a one-sided alternative for a general difference question: “Do the proportions differ?” permits either direction, so use \(p_1\ne p_2\). A greater-than or less-than alternative answers a more specific question.
• Choosing the alternative after seeing the sample results: Do not use the observed ordering alone to select a directional alternative. Match the alternative to the research question or claim.
• Putting inequality in the null: For the standard two-proportion test, state equality in \(H_0\), such as \(p_1=p_2\), and put the proposed difference in \(H_a\).
• Forgetting the context in words: “The proportions are different” is incomplete if the reader cannot tell which populations and outcome are being compared. Name both groups and the shared characteristic.

Key Takeaway

Writing hypotheses is a matter of translating the population question precisely. Define the two proportions for one shared outcome, use equality for the null, and let the wording of the research question determine whether the alternative is two-sided or directional. Sample results may later provide evidence, but they do not replace the population hypotheses.