Comprehensive Study Guide
Module 12: Chi-Square Tests
Table of Contents
1. Overview of Chi-Square Tests
Purpose: Chi-square tests analyze categorical data to determine if observed frequencies differ significantly from expected frequencies.
Three Types of Chi-Square Tests
- Goodness of Fit: Tests if sample data fits an expected distribution
- Independence: Tests if two categorical variables are related
- Homogeneity: Tests if multiple populations have the same distribution
When to Use Chi-Square
- You have categorical data (counts/frequencies)
- You want to compare observed vs. expected frequencies
- You need to test relationships between categorical variables
- Do NOT use for quantitative data (use t-tests, ANOVA instead)
2. Chi-Square Distribution
Properties
- Always positive: χ² ≥ 0 (cannot be negative)
- Right-skewed: Tail extends to the right
- Shape depends on df: More degrees of freedom → more symmetric
- Mean = df: The expected value equals the degrees of freedom
Key Characteristics
- Unlike the normal distribution, chi-square is NOT symmetric (except with very large df)
- We always use the RIGHT tail for rejection region
- Different df values create different distribution curves
3. Chi-Square Goodness of Fit Test
When to Use
- ONE categorical variable
- ONE sample
- Want to test if data fits an expected distribution
Step-by-Step Procedure
- State hypotheses
- H₀: The data follows the specified distribution
- Hₐ: The data does NOT follow the specified distribution
- Check conditions
- Random sampling
- Independent observations
- All expected frequencies ≥ 5
- Calculate expected frequencies
- E = n × p (for each category)
- Calculate test statistic
- χ² = Σ[(O - E)² / E]
- Find degrees of freedom
- df = k - 1
- Find p-value or critical value
- Make decision and conclude
Formulas for Goodness of Fit
Expected Frequency:
E = n × p
where n = sample size, p = expected proportion
Test Statistic:
χ² = Σ[(O - E)² / E]
Sum over all categories
Degrees of Freedom:
df = k - 1
where k = number of categories
Example
Problem: Test if a die is fair with 120 rolls
Expected: Each face should appear 120/6 = 20 times
df: 6 - 1 = 5
4. Chi-Square Test of Independence
When to Use
- TWO categorical variables
- ONE sample (classified by both variables)
- Want to test if variables are related
Step-by-Step Procedure
- State hypotheses
- H₀: The two variables are independent (no association)
- Hₐ: The two variables are dependent (associated)
- Check conditions (same as goodness of fit)
- Calculate expected frequencies
- E = (Row Total × Column Total) / Grand Total
- Calculate for EACH cell
- Calculate test statistic
- χ² = Σ[(O - E)² / E] over ALL cells
- Find degrees of freedom
- df = (r - 1)(c - 1)
- Find p-value or critical value
- Make decision and conclude
Formulas for Test of Independence
Expected Frequency (for each cell):
E = (Row Total × Column Total) / Grand Total
Test Statistic:
χ² = Σ[(O - E)² / E]
Sum over ALL cells in contingency table
Degrees of Freedom:
df = (r - 1)(c - 1)
where r = rows, c = columns
Example
Problem: Test if gender and political party are independent (one sample of 500)
Setup: 2×3 contingency table (2 genders × 3 parties)
df: (2-1)(3-1) = 2
5. Chi-Square Test of Homogeneity
When to Use
- ONE categorical variable
- MULTIPLE samples/populations
- Want to test if distributions are the same across groups
Step-by-Step Procedure
- State hypotheses
- H₀: The distribution is the same for all populations
- Hₐ: At least one population has a different distribution
- Check conditions (same as others)
- Calculate expected frequencies
- SAME formula as independence: E = (RT × CT) / GT
- Calculate test statistic
- SAME formula: χ² = Σ[(O - E)² / E]
- Find degrees of freedom
- SAME formula: df = (r - 1)(c - 1)
- Find p-value or critical value
- Make decision and conclude
Key Difference from Independence
Calculations are IDENTICAL, but study design differs:
- Independence: One sample, two variables
- Homogeneity: Multiple samples, one variable
Example
Problem: Compare satisfaction across 3 cities (100 sampled from each city)
Setup: 3×3 table (3 cities × 3 satisfaction levels)
df: (3-1)(3-1) = 4
6. Comparison of All Three Tests
| Feature | Goodness of Fit | Independence | Homogeneity |
|---|---|---|---|
| Variables | 1 categorical | 2 categorical | 1 categorical |
| Samples | 1 sample | 1 sample | 2+ samples |
| Research Question | Does data fit expected distribution? | Are variables related? | Same distribution across groups? |
| Example | Is a die fair? | Gender vs party preference? | Three cities' political views same? |
| Expected Frequency | E = n × p | E = (RT × CT) / GT | E = (RT × CT) / GT |
| Test Statistic | χ² = Σ[(O-E)²/E] | χ² = Σ[(O-E)²/E] | χ² = Σ[(O-E)²/E] |
| Degrees of Freedom | df = k - 1 | df = (r-1)(c-1) | df = (r-1)(c-1) |
7. Choosing the Right Test
Decision Process
- Is the data categorical?
- No → Use t-test, ANOVA, or regression (NOT chi-square)
- Yes → Continue
- How many categorical variables?
- One variable → Goodness of Fit OR Homogeneity (check #3)
- Two variables → Independence OR Homogeneity (check #3)
- How many samples?
- One sample → Goodness of Fit (if 1 variable) or Independence (if 2 variables)
- Multiple samples → Homogeneity
Quick Decision Rules
- Testing if data matches a theory/model? → Goodness of Fit
- One sample classified by TWO variables? → Independence
- Comparing MULTIPLE groups on ONE variable? → Homogeneity
- Random assignment to treatment groups? → Usually Homogeneity
8. Conditions and Assumptions
Required Conditions (ALL Tests)
- Random Sampling or Random Assignment
- Ensures representative data
- If violated: Cannot generalize to population
- Independence of Observations
- Each observation must be independent
- If violated: p-values are unreliable
- All Expected Frequencies ≥ 5
- MOST CRITICAL CONDITION
- Check EVERY expected cell
- If violated: Combine categories, collect more data, or use Fisher's exact test (for 2×2)
- Categorical Data (Counts)
- Must use frequencies, NOT proportions or percentages
9. Interpreting Results
Understanding the Test Statistic
- Small χ²: Observed frequencies close to expected → likely fail to reject H₀
- Large χ²: Observed frequencies far from expected → likely reject H₀
- Perfect fit would give χ² = 0 (never happens with real data)
Making Decisions
Using Critical Value:
- If χ² > critical value → REJECT H₀
- If χ² ≤ critical value → FAIL TO REJECT H₀
Using p-value:
- If p-value < α → REJECT H₀
- If p-value ≥ α → FAIL TO REJECT H₀
Writing Conclusions
Template for Goodness of Fit:
"At the [α] significance level, there is [sufficient/insufficient] evidence to conclude that the data does not follow the [specified distribution]."
Template for Independence:
"At the [α] significance level, there is [sufficient/insufficient] evidence to conclude that [variable 1] and [variable 2] are associated."
Template for Homogeneity:
"At the [α] significance level, there is [sufficient/insufficient] evidence to conclude that the distribution of [variable] differs across [populations]."
Important Reminder
Association ≠ Causation!
Even if a chi-square test shows two variables are associated, this does NOT prove that one causes the other. There may be confounding variables or the relationship may be indirect.
10. Common Mistakes to Avoid
Top 10 Mistakes
- Using proportions instead of counts
- WRONG: Entering 0.35 (35%)
- RIGHT: Entering 35 (the count)
- Not checking expected frequency condition
- Always calculate expected values BEFORE computing χ²
- Check that ALL are ≥ 5
- Confusing independence and homogeneity
- Look at study design, not just the table
- Using wrong degrees of freedom
- Goodness of fit: k - 1
- Independence/Homogeneity: (r-1)(c-1)
- Trying to use one-tailed test
- Chi-square is ALWAYS right-tailed
- Using chi-square for quantitative data
- Chi-square is for categories only!
- Claiming causation from association
- Association does not imply causation
- Forgetting to sum over ALL cells
- χ² must include contribution from every cell
- Not stating conclusion in context
- Always relate back to the original problem
- Rounding too early
- Keep extra decimals in calculations
- Round only final answer
Study Tips
- Make flashcards for the three test types and when to use each
- Practice calculating expected frequencies - this is key!
- Memorize the df formulas for each test
- Always check the expected frequency condition
- Work through complete examples step-by-step
- Practice identifying which test to use from word problems
- Keep the quick reference sheet handy during practice