9.1 Chi-Square Goodness-of-Fit Test

The chi-square goodness-of-fit test is a powerful tool for comparing observed data to expected distributions. It's widely used in genetics, quality control, and social sciences to uncover patterns and deviations in categorical data.

Calculating the chi-square statistic involves summing the squared differences between observed and expected frequencies. The test's interpretation relies on p-values or critical values, with considerations for sample size and data characteristics influencing its applicability and results.

Chi-Square Goodness-of-Fit Test Fundamentals

Purpose of chi-square goodness-of-fit test

• Assesses agreement between observed and expected frequencies compares categorical data to theoretical distributions
• Evaluates if sample data fits specific probability distribution reveals patterns or deviations
• Applied in genetics tests Mendelian inheritance ratios (3:1 dominant to recessive)
• Used in quality control analyzes defect rates in manufacturing (expected vs actual defects)
• Employed in social sciences examines survey response distributions (political party affiliations)

Calculation of chi-square statistic

• Chi-square test statistic formula sums squared differences between observed and expected frequencies
  • $\chi^2 = \sum_{i=1}^k \frac{(O_i - E_i)^2}{E_i}$
    • $O_i$ represents observed frequency for category i (actual counts)
    • $E_i$ denotes expected frequency for category i (theoretical counts)
    • $k$ indicates number of categories (groups or classes)
• Degrees of freedom calculation determines distribution shape
  • $df = k - 1 - m$
    • $k$ represents number of categories (groups or classes)
    • $m$ denotes number of parameters estimated from data (constraints)

Chi-Square Test Interpretation and Considerations

Interpretation of test results

• P-value approach compares calculated p-value to significance level
  • Reject null hypothesis if p-value < α (typically 0.05)
  • Indicates strength of evidence against null hypothesis
• Critical value approach uses chi-square distribution table
  • Reject null hypothesis if test statistic > critical value
  • Provides threshold for decision-making
• Effect size measures quantify magnitude of association
  • Cramer's V for categorical variables ranges from 0 to 1
  • Phi coefficient for 2x2 contingency tables measures correlation

Assumptions and limitations of test

• Assumptions ensure valid test application
  • Random sampling prevents bias in data collection
  • Independent observations ensure no influence between data points
  • Mutually exclusive categories prevent overlap in classifications
  • Expected frequencies ≥ 5 for each category maintains test reliability
• Limitations affect result interpretation
  • Sensitive to sample size influences statistical significance
    • Large samples may detect small, practically insignificant differences
    • Small samples may miss meaningful differences
  • Only tests for significant differences not nature or direction
  • Cannot be used for continuous data without categorization limits applicability
• Alternatives when assumptions violated ensure proper analysis
  • Fisher's exact test for small sample sizes (< 5 in any category)
  • G-test for likelihood ratio testing offers alternative approach