Course contentsShow
← Mathematics
Lesson 3104 of 3,12160. Statistics — ModelingFree lesson

Sign Test for Median

Testing hypotheses about a population median using the sign test with binomial distribution.

Sign Test for Median

What you'll learn: You'll discover how to test hypotheses about a population median without assuming your data follows a normal distribution, using only the signs of differences.

Why the Sign Test?

Sometimes your data doesn't meet the assumptions required for t-tests or other parametric methods—perhaps it's skewed, ordinal, or has outliers. The sign test is a nonparametric alternative that asks a simpler question: "Is the population median equal to some hypothesized value?"

How It Works

The sign test uses a beautifully simple idea:

  1. Hypothesize a median value (call it M₀). Your null hypothesis is H₀: median = M₀
  2. Subtract M₀ from each observation in your sample
  3. Count the signs: How many differences are positive? How many negative? (Ignore zeros)
  4. Think binomial: If the true median equals M₀, then each observation has probability 0.5 of being above or below it—just like flipping a fair coin!

Under H₀, the number of positive signs follows a binomial distribution with n = (number of non-zero differences) and p = 0.5.

The Test Procedure

Calculate your test statistic: the number of positive signs (or negative—your choice, just be consistent). Then find the p-value using the binomial distribution with p = 0.5. If this p-value is small (below your significance level), reject H₀.

Analogy: Imagine claiming a coin is fair. You flip it 20 times and get 16 heads. The sign test asks: "How likely is 16 or more heads if the coin really is fair?" That probability is your p-value.

When to Use It

The sign test works with any continuous distribution and is especially valuable when you have ordinal data, small samples, or severe outliers that would violate normality assumptions.

Key Takeaway: The sign test provides a distribution-free way to test hypotheses about a population median by treating differences from the hypothesized median as binomial trials with p = 0.5.