Learn Without Walls

Hypothesis Tests for Proportions

Learn how to test claims about population proportions using the normal approximation

Lesson Objectives

By the end of this lesson, you will be able to:

1. When to Test Proportions

We test hypotheses about population proportions when our data is:

Real-World Examples

  • Quality Control: "Less than 5% of products are defective"
  • Politics: "More than 50% of voters support the candidate"
  • Medicine: "The cure rate is 80%"
  • Education: "At least 70% of students pass the exam"
  • Marketing: "30% of customers will buy the new product"

Key Notation

Symbol Meaning
p Population proportion (parameter, unknown)
p₀ Hypothesized population proportion (claimed value in H₀)
(p-hat) Sample proportion (statistic, calculated from data)
n Sample size
x Number of successes in sample

Sample Proportion

p̂ = x / n

Where x = number of successes, n = sample size

2. Conditions for Proportion Tests

IMPORTANT: Check These Conditions First!

Before conducting a hypothesis test for proportions, verify these conditions:

Required Conditions:

  1. Random Sample: The sample must be randomly selected from the population
  2. Independence: Observations must be independent
    • If sampling without replacement: n ≤ 0.05N (sample is at most 5% of population)
  3. Normal Approximation (Success-Failure Condition):
    np₀ ≥ 10 AND n(1 - p₀) ≥ 10

    This ensures the sampling distribution of p̂ is approximately normal so we can use the z-distribution.

Example 1: Checking Conditions

Scenario: A company claims 25% of customers prefer their new product. We survey 200 customers.

Check conditions:

  • Random sample: (assume survey was random)
  • Independence: (200 is less than 5% of all customers)
  • Normal approximation:
    np₀ = 200(0.25) = 50 ≥ 10
    n(1-p₀) = 200(0.75) = 150 ≥ 10

Conclusion: All conditions are met. We can proceed with the test!

3. Test Statistic for Proportions

Test Statistic for One-Sample Proportion Test

z = (p̂ - p₀) / √(p₀(1-p₀)/n)

Where:
p̂ = sample proportion
p₀ = hypothesized population proportion (from H₀)
n = sample size

Important Note: When calculating the standard error for hypothesis testing, we use p₀ (the hypothesized proportion from H₀), NOT p̂. This is different from confidence intervals, where we use p̂.

Hypothesis Forms

Test Type Null Hypothesis Alternative Hypothesis
Two-tailed H₀: p = p₀ Hₐ: p ≠ p₀
Right-tailed H₀: p = p₀ Hₐ: p > p₀
Left-tailed H₀: p = p₀ Hₐ: p < p₀

4. Examples: Hypothesis Tests for Proportions

Example 2: Right-Tailed Test (Critical Value Approach)

Problem: A politician claims that more than 60% of voters support a new policy. A poll of 400 voters shows 256 support it. Test at α = 0.05.

Step 1: State hypotheses

  • H₀: p = 0.60 (60% support, as claimed)
  • Hₐ: p > 0.60 (more than 60% support) — right-tailed test
  • α = 0.05

Step 2: Check conditions

np₀ = 400(0.60) = 240 ≥ 10
n(1-p₀) = 400(0.40) = 160 ≥ 10
Conditions met!

Step 3: Calculate sample proportion and test statistic

p̂ = x/n = 256/400 = 0.64

z = (p̂ - p₀) / √(p₀(1-p₀)/n)
z = (0.64 - 0.60) / √(0.60(0.40)/400)
z = 0.04 / √(0.24/400)
z = 0.04 / √0.0006
z = 0.04 / 0.0245
z = 1.63

Step 4: Find critical value

Right-tailed test, α = 0.05 → critical value = 1.645

Step 5: Make decision

z = 1.63 < 1.645 (not in critical region)

Decision: Fail to reject H₀

Step 6: Conclusion

There is not sufficient evidence at the 0.05 significance level to support the politician's claim that more than 60% of voters support the policy. While the sample showed 64% support, this difference could be due to sampling variability.

Example 3: Left-Tailed Test (P-Value Approach)

Problem: A manufacturer claims their defect rate is 3%. Quality control inspects 500 items and finds 10 defects. Test if the defect rate is less than claimed (α = 0.01).

Step 1: State hypotheses

  • H₀: p = 0.03 (defect rate is 3%)
  • Hₐ: p < 0.03 (defect rate is less than 3%) — left-tailed test
  • α = 0.01

Step 2: Check conditions

np₀ = 500(0.03) = 15 ≥ 10
n(1-p₀) = 500(0.97) = 485 ≥ 10
Conditions met!

Step 3: Calculate test statistic

p̂ = 10/500 = 0.02

z = (0.02 - 0.03) / √(0.03(0.97)/500)
z = -0.01 / √(0.0291/500)
z = -0.01 / 0.00763
z = -1.31

Step 4: Find p-value

For left-tailed test: p-value = P(Z < -1.31)

From z-table: p-value = 0.0951

Step 5: Make decision

p-value (0.0951) > α (0.01)

Decision: Fail to reject H₀

Step 6: Conclusion

There is not sufficient evidence to conclude the defect rate is less than 3%. The observed 2% defect rate could reasonably occur by chance if the true rate is 3%.

Example 4: Two-Tailed Test

Problem: A university claims 70% of graduates find jobs within 6 months. A survey of 300 recent graduates finds 195 employed. Test if the actual rate differs from the claim (α = 0.05).

Step 1: State hypotheses

  • H₀: p = 0.70 (employment rate is 70%)
  • Hₐ: p ≠ 0.70 (employment rate differs from 70%) — two-tailed test
  • α = 0.05

Step 2: Check conditions

np₀ = 300(0.70) = 210 ≥ 10
n(1-p₀) = 300(0.30) = 90 ≥ 10

Step 3: Calculate test statistic

p̂ = 195/300 = 0.65

z = (0.65 - 0.70) / √(0.70(0.30)/300)
z = -0.05 / √(0.21/300)
z = -0.05 / 0.0265
z = -1.89

Step 4: Decision (both approaches)

Critical value: Two-tailed, α = 0.05 → ±1.96

|z| = 1.89 < 1.96 → Fail to reject H₀

P-value: For z = -1.89, two-tailed p-value = 2(0.0294) = 0.0588

0.0588 > 0.05 → Fail to reject H₀

Step 5: Conclusion

There is not sufficient evidence to conclude the employment rate differs from the claimed 70%. Although only 65% of the sample were employed, this could be due to sampling variability (p = 0.0588 is close to, but above, α = 0.05).

Check Your Understanding

Question 1: A candy company claims 20% of their candies are red. In a bag of 100 candies, you find 15 red ones. Check if the normal approximation conditions are met.

Check conditions:

np₀ = 100(0.20) = 20 ≥ 10
n(1-p₀) = 100(0.80) = 80 ≥ 10

Conclusion: Yes, conditions are met. We can use the normal approximation (z-test) for proportions.

Question 2: Calculate the test statistic: n = 200, x = 50, p₀ = 0.20

Step 1: Calculate p̂

p̂ = x/n = 50/200 = 0.25

Step 2: Calculate z

z = (p̂ - p₀) / √(p₀(1-p₀)/n)
z = (0.25 - 0.20) / √(0.20(0.80)/200)
z = 0.05 / √(0.16/200)
z = 0.05 / √0.0008
z = 0.05 / 0.0283
z = 1.77

Question 3: For a two-tailed test at α = 0.05, if z = -2.15, what is your decision?

Critical values: ±1.96

Comparison: |z| = |-2.15| = 2.15 > 1.96

Decision: Reject H₀

Reasoning: The test statistic falls in the critical region (more extreme than ±1.96), so we have sufficient evidence to reject the null hypothesis at the 0.05 significance level.

Summary

  • Use proportion tests for categorical data (yes/no, success/failure)
  • Check conditions: Random sample, independence, and np₀ ≥ 10 AND n(1-p₀) ≥ 10
  • Test statistic: z = (p̂ - p₀) / √(p₀(1-p₀)/n)
  • Use p₀ (hypothesized value) in standard error, not p̂
  • Critical values come from standard normal (z) distribution
  • Always interpret results in the context of the real-world problem
← Previous: Tests for Means Next: Practice Problems →