📊

Understanding Two-Sided P-Values in Statistics

May 26, 2025

Stat Quest: Calculating P-values

Introduction

  • Focus is on calculating two-sided p-values.
  • One-sided p-values are mentioned briefly, with caution.

Coin Flip Example

  • Hypothesis: Coin is not special; it's like a normal coin (null hypothesis).
  • P-value: Small p-value suggests rejecting the null hypothesis.
  • Probability Calculation:
    • Four possible outcomes when flipping a coin twice.
    • Probability of two heads: 1/4 = 0.25.
    • P-value includes equally rare or more extreme outcomes.
    • P-value for two heads: 0.5 (0.25 + 0.25).
  • Conclusion: Fail to reject the hypothesis (p-value > 0.05).

Importance of Rare or Extreme Events

  • Adding equally rare or rarer events ensures special findings are genuinely unusual.
  • Comparison with rare flowers analogy.

Example: Four Heads and One Tail

  • Calculation involves all possible outcomes with five flips.
  • P-value: Sum of probabilities for the observed and equally rare or more extreme outcomes.
  • Conclusion: Fail to reject the hypothesis (p-value = 0.375).

Distributions and Continuous Variables

  • Example: Heights of Brazilian women.
  • Use of statistical distributions for continuous variables.
  • P-values calculated using the area under the curve.
  • Example: Measuring a height of 142 cm or between 155.4 and 156 cm.
    • Calculation involves more extreme values.

One-sided P-values

  • Example scenario: Testing a drug's effect on recovery times.
  • One-sided p-value focuses on a specific direction of change.
  • Caution: Can be misleading if not used correctly.

Summary

  • P-value consists of:
    1. Probability of observed event.
    2. Probability of equally rare events.
    3. Probability of rarer or more extreme events.
  • Two-sided p-value considers both directions of unusual change.
  • One-sided p-value should be used cautiously.

Conclusion

  • Encouragement to subscribe or support the channel.

Note: Aim for a p-value less than 0.05 to reject the null hypothesis.