9.2 Tests of Independence and Homogeneity

Chi-square tests help us understand relationships between categories. We use them to check if variables are independent or if proportions are the same across groups. These tests are crucial for analyzing survey data and making sense of categorical information.

The process involves setting up hypotheses, crunching numbers, and interpreting results. We can also measure how strong associations are using tools like Phi coefficient and Cramer's V. These methods give us a clearer picture of connections in our data.

Tests of Independence and Homogeneity

Tests of independence vs homogeneity

• Test of Independence examines relationship between two categorical variables within single population, null hypothesis assumes no association
• Test of Homogeneity compares proportions across multiple populations or groups, null hypothesis assumes equal proportions
• Key differences involve sampling method (random vs stratified), research question (association vs proportion comparison), and data structure (single vs multiple contingency tables)

Chi-square test for independence

• Steps: State hypotheses, construct contingency table, calculate expected frequencies, compute chi-square statistic, determine degrees of freedom, find p-value, interpret results
• Chi-square statistic formula: $\chi^2 = \sum \frac{(O - E)^2}{E}$ where O is observed frequency and E is expected frequency
• Degrees of freedom: $df = (r - 1)(c - 1)$ where r is number of rows and c is number of columns
• Interpretation compares p-value to significance level, rejecting null if p-value < significance level, concluding variable independence or association (smoking and lung cancer)

Chi-square test for homogeneity

• Steps mirror independence test: State hypotheses, organize data, calculate expected frequencies, compute chi-square statistic, determine degrees of freedom, find p-value, draw conclusions
• Uses same chi-square statistic and degrees of freedom formulas as independence test
• Conclusions drawn by comparing p-value to significance level, rejecting null if p-value < significance level, determining proportion homogeneity or differences across populations (voting preferences across age groups)

Association strength measures

• Phi coefficient measures association strength for 2x2 contingency tables
  • Formula: $\phi = \sqrt{\frac{\chi^2}{n}}$ where n is total sample size
  • Range from -1 to 1, values closer to ±1 indicate stronger association (gender and career choice)
• Cramer's V measures association strength for larger contingency tables
  • Formula: $V = \sqrt{\frac{\chi^2}{n(k-1)}}$ where k is smaller of number of rows or columns
  • Range from 0 to 1, values closer to 1 indicate stronger association (education level and income)
• Interpretation guidelines: 0.00-0.10 negligible, 0.10-0.30 weak, 0.30-0.50 moderate, 0.50-1.00 strong association
• Effect size measures provide sample size-independent comparisons across studies and practical significance alongside statistical significance