일치 쌍에 대한 부호 검정

Use the sign test when the underlying population distribution is unknown or cannot be assumed normal. This nonparametric test is particularly valuable for small sample sizes and paired data where you cannot verify normality assumptions. It compares pre- and post-treatment measurements on the same subjects to determine if median values differ significantly.

Subtract the second variable value from the first variable for each matched pair. Record the sign of each difference as positive, negative, or zero. Exclude all zero differences from the analysis, as they do not contribute to the test conclusion. Count the remaining positive and negative signs to determine your test statistic.

The null hypothesis posits that the median difference between matched pairs is zero, implying no treatment effect. The alternative hypothesis suggests the median difference is not equal to zero, indicating a significant treatment effect. The test determines whether sufficient evidence exists to reject the null hypothesis at your chosen significance level.

The test statistic is the count of the less frequent sign (positive or negative) among the paired differences. When sample size is less than 25, you compare this count directly against a critical value from a statistical table at your significance level, typically 0.05.

If the test statistic is less than or equal to the critical value, the result is statistically significant, and you reject the null hypothesis at 95% confidence. This indicates a significant difference in median values between the paired samples. If the test statistic exceeds the critical value, there is insufficient evidence to reject the null hypothesis.

The sign test analyzes only the direction of differences, not their size, making it robust to outliers and extreme values. This approach simplifies calculations and makes the test applicable when you cannot assume specific distributional properties. By focusing on signs rather than magnitudes, the test remains valid even with skewed or non-normal data distributions.

The sign test is simple to conduct, requires no normality assumptions, and works effectively with small sample sizes. It is particularly useful when data are ordinal or when the underlying distribution is unknown. The test's simplicity and versatility make it a practical choice for comparing pre- and post-intervention measurements in research studies.