Course contentsShow
← Data Science
Lesson 400 of 2,14511. Hypothesis Tests for Proportions and CategoriesFree lesson

Assumptions and Conditions for Proportion Tests

Verifying sample size requirements, independence, and the success-failure condition for valid inference.

Assumptions and Conditions for Proportion Tests

What you'll learn: You'll discover the critical checklist that determines whether your proportion test will produce trustworthy results.

Why Conditions Matter

Before running a proportion test (like the one-sample Z-test for proportions you just learned), you need to verify that your data meets certain requirements. Think of these as safety checks before takeoff—ignore them, and your statistical conclusions might crash.

The Three Key Conditions

1. Independence

Each observation must be independent of the others. Your sample should represent less than 10% of the population (the "10% condition"), or observations should be randomly sampled. If people influence each other's responses, independence fails.

2. Randomization

Data should come from a random sample or randomized experiment. Without randomization, your sample may be biased, and your inferences won't generalize.

3. Success-Failure Condition

This is the big one for proportions. You need at least 10 successes AND at least 10 failures in your sample:

  • np ≥ 10 (expected successes)
  • n(1-p) ≥ 10 (expected failures)

Where n is your sample size and p is the proportion under your null hypothesis.

Why the Success-Failure Condition Exists

The normal approximation we use for proportion tests only works well when you have enough data in both categories. With too few successes or failures, the sampling distribution becomes skewed, and the Z-test becomes unreliable.

Example: Testing if a coin is fair (p = 0.5) with n = 15 flips? You'd expect 7.5 successes and 7.5 failures—both below 10. The test isn't valid yet. Wait until you have at least 20 flips.

Key Takeaway: Before trusting any proportion test, verify independence, randomization, and that np ≥ 10 and n(1-p) ≥ 10—without these, your p-values are meaningless.