Chi-Square Tests & Categorical Data Analysis: Tests, Assumptions & Applications

Key Concepts & Definitions

1. Categorical Data

• Data classified into categories (nominal/ordinal) rather than numerical values.
• Example: Gender (Male/Female), Opinion (Agree/Disagree), Product Preference (Brand A/B/C).

2. Chi-Square (χ²) Tests

A family of non-parametric tests used to analyze categorical data by comparing observed frequencies with expected frequencies under a null hypothesis.

---

Types of Chi-Square Tests

1. Chi-Square Goodness-of-Fit Test

Purpose: Tests if a sample matches a population distribution. Formula: $${\chi }^2=\sum \frac{(O_i-E_i{)}^2}{E_i}$$ where:

• $O_i$ = Observed frequency
• $E_i$ = Expected frequency (calculated under $H_0$)

Assumptions:

• Categorical data (nominal/ordinal).
• Expected frequency $E_i\ge 5$ in at least 80% of cells (if violated, combine categories).
• Independent observations.

Example: A die is rolled 60 times. Observed frequencies: 10, 8, 12, 9, 11, 10. Test if fair ($H_0$: $p=\frac{1}{6}$ for each face).

Steps:

1. Calculate $E_i$: $E_i=\frac{60}{6}=10$ for each face.
2. Compute ${\chi }^2$: $${\chi }^2=\frac{(10-10{)}^2}{10}+\frac{(8-10{)}^2}{10}+\cdots +\frac{(10-10{)}^2}{10}=0.8$$
3. Degrees of Freedom (df): $k-1=6-1=5$ (where $k$ = number of categories).
4. Critical Value: From χ² table, ${\chi }_{0.05,5}^2=11.07$.
5. Decision: Since $0.8<11.07$, fail to reject $H_0$ (die is fair).

---

2. Chi-Square Test of Independence

Purpose: Tests if two categorical variables are independent (no association). Formula: Same as goodness-of-fit, but $E_i$ is calculated using row/column totals.

Assumptions:

• Same as goodness-of-fit.
• Variables are cross-classified (contingency table).

Example: Test if gender (Male/Female) and smoking status (Smoker/Non-smoker) are independent in a sample of 200 people.

Steps:

1. Calculate $E_{ij}$: $$E_{ij}=\frac{(\text{Row Total})\times (\text{Column Total})}{\text{Grand Total}}$$

   • $E_{\text{Male,Smoker}}=\frac{100\times 60}{200}=30$
   • $E_{\text{Female,Non-smoker}}=\frac{100\times 140}{200}=70$

2. Compute ${\chi }^2$: $${\chi }^2=\frac{(40-30{)}^2}{30}+\frac{(60-70{)}^2}{70}+\frac{(20-30{)}^2}{30}+\frac{(80-70{)}^2}{70}=10$$

3. df: $(r-1)(c-1)=(2-1)(2-1)=1$ (where $r$ = rows, $c$ = columns).

4. Critical Value: ${\chi }_{0.05,1}^2=3.841$.

5. Decision: Since $10>3.841$, reject $H_0$ (gender and smoking status are dependent).

---

Comparison Table: Goodness-of-Fit vs. Test of Independence

---

Advantages & Limitations

Advantages:

• Non-parametric: No assumptions about population distribution.
• Versatile: Works for nominal/ordinal data.
• Simple Interpretation: Rejecting $H_0$ suggests a meaningful pattern.

Limitations:

• Sample Size Dependency: Requires $E_i\ge 5$ (use Fisher’s exact test otherwise).
• No Causality: Only tests association, not cause-effect.
• Sensitive to Small Samples: May lack power with few observations.

---

Applications in Real World

1. Market Research:
   • Test if customer preferences differ by age group (e.g., soft drink vs. energy drink).
2. Quality Control:
   • Check if defects in products vary by production shift.
3. Medical Studies:
   • Determine if a disease is associated with smoking status.
4. Social Sciences:
   • Analyze voting patterns by demographic groups.

---

Worked Example: Past Exam Question (Adapted)

Question: A survey of 200 students classified by gender and preference for online vs. offline classes:

Test at 5% significance if gender and class preference are independent.

Solution:

1. Calculate $E_{ij}$:

   • $E_{\text{Male,Online}}=\frac{100\times 90}{200}=45$
   • $E_{\text{Female,Offline}}=\frac{100\times 110}{200}=55$

2. Compute ${\chi }^2$: $${\chi }^2=\frac{(50-45{)}^2}{45}+\frac{(50-55{)}^2}{55}+\frac{(40-45{)}^2}{45}+\frac{(60-55{)}^2}{55}=1.27$$

3. df: $(2-1)(2-1)=1$.

4. Critical Value: ${\chi }_{0.05,1}^2=3.841$.

5. Decision: $1.27<3.841$ → Fail to reject $H_0$ (no significant association).

---

Exam Tip

What Examiners Look For:

1. Correct Hypothesis Formulation:
   • Always state $H_0$ and $H_1$ clearly (e.g., "$H_0$: Variables are independent").
2. Expected Frequency Calculation:
   • Show all $E_i$ values (partial credit if missing).
3. Chi-Square Formula Application:
   • Use the correct formula and compute each term separately.
4. Degrees of Freedom:
   • Remember:
     • Goodness-of-fit: $k-1$
     • Test of independence: $(r-1)(c-1)$
5. Decision Rule:
   • Compare ${\chi }^2$ with critical value or use $p$-value (if provided).
6. Interpretation:
   • Avoid vague statements like "reject $H_0$". Instead:
     • ✅ "There is significant evidence to suggest an association between [variables]."
     • ❌ "The null hypothesis is wrong."

Common Mistakes to Avoid:

• Ignoring Assumptions: If $E_i<5$, mention it and suggest combining categories.
• Incorrect df: Double-check $(r-1)(c-1)$ for contingency tables.
• Arithmetic Errors: Recompute ${\chi }^2$ carefully (examiners deduct for calculation mistakes).
• Misinterpreting Results: A rejected $H_0$ means association exists, not causation.

Quick Revision Checklist:

• Can I define chi-square tests and their types?
• Do I know how to calculate expected frequencies?
• Can I compute ${\chi }^2$ and determine df correctly?
• Can I interpret results in context (e.g., "smoking and disease are associated")?
• Do I recognize when to use goodness-of-fit vs. test of independence?

---

End of Note