Solved Exercises on Applications of Two-Way Tables

Master applications of two-way tables: conditional probability, chi-square tests, association analysis, and real-world statistical problems through these 5 detailed exercises.

Solution: Exercises 1 to 3
1 Conditional Probability Applications
Exercise 1
A survey of 500 students was conducted to determine the relationship between grade level and participation in extracurricular activities. The results are shown in the two-way table below:

Participates Does Not Participate Total
Freshman 80 45 125
Sophomore 90 35 125
Junior 100 25 125
Senior 110 15 125
Total 380 120 500

Calculate P(Participates|Junior), P(Junior|Participates), and determine if there is an association between grade level and participation.
Definition:

Conditional Probability: P(A|B) = P(A and B)/P(B), which is the probability of A occurring given that B has occurred.

Joint Probability: P(A and B) = (count in cell where A and B intersect)/(total count).

Marginal Probability: P(A) = (total for A)/(grand total).

Conditional Probability Calculation Method:
  1. Identify the condition (what we know happened)
  2. Identify the event we want to find the probability of
  3. Apply the formula: P(Event|Condition) = (count where both occur)/(count where condition occurs)
  4. For P(A|B): numerator = cell count for A and B, denominator = total for B
Step 1: Calculate P(Participates|Junior)

P(Participates|Junior) = (Number of Juniors who participate)/(Total number of Juniors)

= 100/125 = 0.8 or 80%

Step 2: Calculate P(Junior|Participates)

P(Junior|Participates) = (Number of Juniors who participate)/(Total number who participate)

= 100/380 ≈ 0.263 or 26.3%

Step 3: Calculate Marginal Probability

P(Participates) = Total who participate / Grand total

= 380/500 = 0.76 or 76%

Step 4: Analyze Association

Compare conditional probability to marginal probability:

  • P(Participates|Junior) = 80%
  • P(Participates) = 76%
  • Since 80% ≠ 76%, there appears to be an association
Step 5: Calculate All Conditional Probabilities

Freshman: 80/125 = 64%

Sophomore: 90/125 = 72%

Junior: 100/125 = 80%

Senior: 110/125 = 88%

P(Participates|Junior) = 0.8 (80%)
P(Junior|Participates) = 0.263 (26.3%)
There is a strong positive association between grade level and participation.
Final answer:

P(Participates|Junior) = 0.8 (80%) and P(Junior|Participates) = 0.263 (26.3%). There is a strong positive association between grade level and participation in extracurricular activities. Participation rates increase systematically from 64% for freshmen to 88% for seniors.

Applied rules:

• Conditional Probability Formula: P(A|B) = P(A ∩ B)/P(B)

• Association Detection: If P(A|B) ≠ P(A), then A and B are associated

• Independence Test: If P(A|B) = P(A) for all B, then A and B are independent

• Direction Matters: P(A|B) ≠ P(B|A) in general

2 Chi-Square Test for Independence
Exercise 2
A researcher wants to determine if there is an association between smoking status and exercise frequency. The results from a sample of 400 adults are shown in the two-way table below:

Regular Exercise Occasional Exercise No Exercise Total
Smoker 40 30 30 100
Non-smoker 120 80 100 300
Total 160 110 130 400

Perform a chi-square test of independence to determine if there is an association between smoking status and exercise frequency.
Definition:

Chi-Square Test of Independence: A statistical test used to determine if there is a significant association between two categorical variables.

Null Hypothesis (H₀): The two variables are independent (no association).

Alternative Hypothesis (H₁): The two variables are dependent (there is an association).

Chi-Square Test Formula:

χ² = Σ [(Observed - Expected)² / Expected]

Expected Count = (Row Total × Column Total) / Grand Total

Degrees of Freedom = (r - 1)(c - 1)

Step 1: Calculate Expected Frequencies

Smoker + Regular Exercise: (100 × 160) / 400 = 40

Smoker + Occasional Exercise: (100 × 110) / 400 = 27.5

Smoker + No Exercise: (100 × 130) / 400 = 32.5

Non-smoker + Regular Exercise: (300 × 160) / 400 = 120

Non-smoker + Occasional Exercise: (300 × 110) / 400 = 82.5

Non-smoker + No Exercise: (300 × 130) / 400 = 97.5

Step 2: Calculate Chi-Square Components

Smoker + Regular: (40 - 40)² / 40 = 0

Smoker + Occasional: (30 - 27.5)² / 27.5 = 0.227

Smoker + No: (30 - 32.5)² / 32.5 = 0.192

Non-smoker + Regular: (120 - 120)² / 120 = 0

Non-smoker + Occasional: (80 - 82.5)² / 82.5 = 0.076

Non-smoker + No: (100 - 97.5)² / 97.5 = 0.064

Step 3: Calculate Chi-Square Statistic

χ² = 0 + 0.227 + 0.192 + 0 + 0.076 + 0.064 = 0.559

Step 4: Determine Degrees of Freedom

df = (rows - 1) × (columns - 1) = (2 - 1) × (3 - 1) = 2

Step 5: Compare to Critical Value

For α = 0.05 and df = 2, critical value ≈ 5.99

Since 0.559 < 5.99, we fail to reject H₀

Step 6: State Conclusion

At the 5% significance level, there is insufficient evidence to conclude that there is an association between smoking status and exercise frequency.

χ² = 0.559 < 5.99 (critical value)
We fail to reject the null hypothesis. No significant association found.
Final answer:

With χ² = 0.559 and df = 2, we fail to reject the null hypothesis at α = 0.05. There is insufficient evidence to conclude that there is a significant association between smoking status and exercise frequency. The variables appear to be independent.

Applied rules:

• Expected Frequency: E = (Row Total × Column Total) / Grand Total

• Chi-Square Statistic: χ² = Σ[(O - E)²/E]

• Degrees of Freedom: df = (r-1)(c-1)

• Decision Rule: Reject H₀ if χ² > critical value

3 Medical Study Application
Exercise 3
A medical study was conducted to determine if a new drug is effective in treating a particular disease. The results from 300 patients are shown in the two-way table below:

Recovered Did Not Recover Total
Received Drug 110 40 150
Received Placebo 70 80 150
Total 180 120 300

Calculate the relative risk and determine if the drug is effective.
Definition:

Relative Risk (RR): The ratio of the probability of an outcome in the exposed group to the probability of the same outcome in the unexposed group.

Relative Risk Formula: RR = P(outcome|exposed) / P(outcome|unexposed)

Interpretation: RR = 1 (no association), RR > 1 (increased risk), RR < 1 (decreased risk/protection).

RR = \frac{P(\text{Recovery}|\text{Drug})}{P(\text{Recovery}|\text{Placebo})}
Relative Risk Formula
Step 1: Calculate Recovery Probabilities

P(Recovery|Drug): 110/150 = 0.733 or 73.3%

P(Recovery|Placebo): 70/150 = 0.467 or 46.7%

Step 2: Calculate Relative Risk

RR = P(Recovery|Drug) / P(Recovery|Placebo)

RR = 0.733 / 0.467 = 1.57

Step 3: Calculate Additional Measures

P(No Recovery|Drug): 40/150 = 0.267 or 26.7%

P(No Recovery|Placebo): 80/150 = 0.533 or 53.3%

Reduction in risk: 53.3% - 26.7% = 26.6%

Step 4: Calculate Odds Ratio

Odds of recovery with drug: 110/40 = 2.75

Odds of recovery with placebo: 70/80 = 0.875

Odds Ratio: 2.75 / 0.875 = 3.14

Step 5: Interpret Results

Since RR = 1.57 > 1, the drug increases the probability of recovery.

Patients receiving the drug are 1.57 times more likely to recover than those receiving placebo.

The odds ratio of 3.14 indicates even stronger association in the odds scale.

RR = 1.57 > 1
The drug is effective in increasing recovery rates.
Final answer:

The relative risk is 1.57, indicating that patients receiving the drug are 1.57 times more likely to recover than those receiving placebo. The drug is effective, with recovery rates increasing from 46.7% with placebo to 73.3% with the drug (a 26.6 percentage point improvement).

Applied rules:

• Relative Risk: RR = P(outcome|exposed) / P(outcome|unexposed)

• Effectiveness: RR > 1 indicates beneficial effect when outcome is favorable

• Interpretation: RR of 1.57 means 57% higher probability of favorable outcome

• Clinical Significance: Consider both statistical and practical significance

Key Concepts and Statistical Measures for Two-Way Table Applications
P(A|B) = \frac{P(A \cap B)}{P(B)} = \frac{\text{Cell Count}}{\text{Row/Column Total}}
Conditional Probability Formula
Chi-Square
\chi^2 = \sum \frac{(O-E)^2}{E}
Test for Independence
Expected Freq.
E = \frac{RT \times CT}{GT}
Expected Count Formula
Relative Risk
RR = \frac{P(\text{Outcome}|\text{Exposed})}{P(\text{Outcome}|\text{Unexposed})}
Effectiveness Measure
Key definitions:

Two-Way Table: A contingency table displaying the frequency distribution of two categorical variables, showing the intersection of categories from both variables.

Marginal Distribution: The distribution of one variable in the table, calculated by summing across the other variable (row totals or column totals).

Conditional Distribution: The distribution of one variable given a specific value of the other variable, calculated as proportions within rows or columns.

Association: A relationship between two variables where the conditional distribution of one variable depends on the value of the other variable.

Independence: Two variables are independent if knowing the value of one variable provides no information about the value of the other variable.

Two-Way Table Analysis Methodology:
  1. Data Organization: Construct a two-way table with appropriate row and column labels
  2. Calculate Marginal Totals: Sum rows and columns to get totals
  3. Determine Marginal Proportions: Calculate overall proportions for each category
  4. Calculate Conditional Proportions: Calculate proportions within each row/column
  5. Compare Distributions: Compare conditional distributions to marginal distributions
  6. Statistical Testing: Perform chi-square test for formal independence assessment
  7. Measure Association Strength: Calculate relative risk, odds ratio, or phi coefficient
  8. Interpret Results: Draw conclusions about the nature and strength of association
Tip 1: Always identify which variable is explanatory and which is response when analyzing association.
Tip 2: Look for patterns in the data before performing calculations - visual inspection often reveals associations.
Tip 3: When conditional probabilities vary significantly across categories, there is likely an association.
Tip 4: Remember that association does not imply causation - look for potential confounding variables.
Tip 5: For small samples, consider using Fisher's exact test instead of chi-square test.
Common errors: Confusing marginal and conditional probabilities, misinterpreting chi-square results, assuming causation from association, using chi-square with small expected frequencies.
Exam preparation: Practice constructing two-way tables, calculating conditional probabilities, interpreting chi-square results, distinguishing association from causation.
Essential Formulas and Rules:

• Conditional Probability: P(A|B) = P(A∩B)/P(B) = (Cell Count)/(Row/Column Total)

• Expected Frequency: E = (Row Total × Column Total)/Grand Total

• Chi-Square Statistic: χ² = Σ[(O - E)²/E]

• Degrees of Freedom: df = (r-1)(c-1) where r = rows, c = columns

• Independence Test: P(A|B) = P(A) if and only if A and B are independent

• Relative Risk: RR = P(outcome|exposed)/P(outcome|unexposed)

• Odds Ratio: OR = (a/c)/(b/d) where a,b,c,d are cell counts in 2×2 table

Solution: Exercises 4 to 5
4 Market Research Application
Exercise 4
A market research firm surveyed 800 consumers to determine if there is an association between income level and brand preference for smartphones. The results are shown in the two-way table below:

Brand A Brand B Brand C Total
Low Income 80 120 100 300
Medium Income 120 100 80 300
High Income 100 80 120 300
Total 300 300 300 900

Perform a comprehensive analysis to determine if there is an association between income level and brand preference.
Definition:

Market Segmentation Analysis: Using two-way tables to identify patterns in consumer preferences across different demographic groups.

Brand Loyalty: The tendency of consumers to consistently purchase the same brand over time.

Step 1: Calculate Marginal Distributions

Overall Brand Preferences:

  • Brand A: 300/900 = 33.3%
  • Brand B: 300/900 = 33.3%
  • Brand C: 300/900 = 33.3%

Income Distribution:

  • Low Income: 300/900 = 33.3%
  • Medium Income: 300/900 = 33.3%
  • High Income: 300/900 = 33.3%
Step 2: Calculate Conditional Distributions

Brand Preferences by Income Level:

  • Low Income: A (26.7%), B (40.0%), C (33.3%)
  • Medium Income: A (40.0%), B (33.3%), C (26.7%)
  • High Income: A (33.3%), B (26.7%), C (40.0%)
Step 3: Compare Conditional to Marginal

Clear differences exist between conditional and marginal distributions:

  • Low income prefers Brand B (40% vs 33.3% overall)
  • Medium income prefers Brand A (40% vs 33.3% overall)
  • High income prefers Brand C (40% vs 33.3% overall)
Step 4: Perform Chi-Square Test

Calculate expected frequencies:

  • Low Income + Brand A: (300 × 300) / 900 = 100
  • Medium Income + Brand B: (300 × 300) / 900 = 100
  • High Income + Brand C: (300 × 300) / 900 = 100

χ² = Σ[(O - E)²/E] = 40, df = (3-1)(3-1) = 4

Critical value at α = 0.05: 9.49

Since 40 > 9.49, we reject H₀

Step 5: Analyze Business Implications

The association has significant marketing implications:

  • Brand A should target medium-income consumers
  • Brand B should focus on low-income markets
  • Brand C should emphasize high-income demographics
Strong association exists between income level and brand preference.
Each income level shows distinct brand preference patterns.
Final answer:

There is a strong association between income level and smartphone brand preference (χ² = 40, p < 0.001). Low-income consumers prefer Brand B (40%), medium-income prefer Brand A (40%), and high-income prefer Brand C (40%). This pattern suggests income-based market segmentation is effective for smartphone marketing.

Applied rules:

• Association Assessment: Compare conditional distributions to marginal distributions

• Statistical Testing: Use chi-square test to confirm significance

• Business Application: Translate statistical findings into actionable insights

• Pattern Recognition: Look for consistent directional changes across categories

5 Educational Policy Analysis
Exercise 5
An educational researcher studied the relationship between teaching method and student performance. The results from 600 students are shown in the two-way table below:

High Performance Average Performance Low Performance Total
Traditional Method 45 105 100 250
Interactive Method 85 115 50 250
Blended Method 70 90 40 200
Total 200 310 190 700

Conduct a comprehensive analysis to evaluate teaching method effectiveness and make policy recommendations.
Definition:

Educational Effectiveness Analysis: Using statistical methods to evaluate the impact of different teaching approaches on student outcomes.

Performance Metrics: Quantitative measures used to assess educational outcomes and teaching effectiveness.

Step 1: Calculate Marginal Distributions

Overall Performance Levels:

  • High Performance: 200/700 = 28.6%
  • Average Performance: 310/700 = 44.3%
  • Low Performance: 190/700 = 27.1%
Step 2: Calculate Conditional Distributions

Performance by Teaching Method:

  • Traditional: High (18.0%), Average (42.0%), Low (40.0%)
  • Interactive: High (34.0%), Average (46.0%), Low (20.0%)
  • Blended: High (35.0%), Average (45.0%), Low (20.0%)
Step 3: Calculate Effectiveness Ratios

High Performance Rates:

  • Traditional: 45/250 = 18.0%
  • Interactive: 85/250 = 34.0%
  • Blended: 70/200 = 35.0%

Low Performance Rates:

  • Traditional: 100/250 = 40.0%
  • Interactive: 50/250 = 20.0%
  • Blended: 40/200 = 20.0%
Step 4: Perform Chi-Square Analysis

Expected frequencies for each cell:

  • Traditional + High: (250 × 200) / 700 = 71.4
  • Interactive + High: (250 × 200) / 700 = 71.4
  • Blended + High: (200 × 200) / 700 = 57.1

After calculating all expected frequencies:

χ² = 34.7, df = (3-1)(3-1) = 4

Critical value at α = 0.05: 9.49

Since 34.7 > 9.49, we reject H₀

Step 5: Make Policy Recommendations

Based on the analysis:

  • Interactive and Blended methods show superior performance
  • Traditional method has highest failure rate (40% low performance)
  • Interactive and Blended methods reduce failure rate to 20%
  • Recommend transitioning from Traditional to Interactive or Blended
Strong association exists between teaching method and student performance.
Interactive and Blended methods are significantly more effective than Traditional.
Final answer:

There is a strong association between teaching method and student performance (χ² = 34.7, p < 0.001). The Interactive (34% high performance) and Blended (35% high performance) methods are significantly more effective than the Traditional method (18% high performance). The Traditional method also has the highest failure rate (40% low performance) compared to Interactive and Blended methods (20% each). Educational policy should favor Interactive and Blended teaching methods.

Applied rules:

• Effectiveness Measurement: Compare success rates across different interventions

• Policy Translation: Convert statistical findings to actionable recommendations

• Statistical Significance: Use formal tests to validate observed differences

• Practical Importance: Consider both statistical and practical significance

Detailed Summary: Applications of Two-Way Tables
P(A|B) = \frac{P(A \cap B)}{P(B)}
Conditional Probability
Key definitions:

Two-Way Table: A contingency table displaying the frequency distribution of two categorical variables, showing the intersection of categories from both variables. Provides a foundation for analyzing relationships between variables.

Contingency Table: Another term for two-way table, emphasizing that the table shows how one variable is contingent upon another.

Marginal Distribution: The distribution of one variable in the table, calculated by summing across the other variable (row totals or column totals). Represents the overall distribution of each variable independently.

Conditional Distribution: The distribution of one variable given a specific value of the other variable, calculated as proportions within rows or columns. Shows how one variable behaves under specific conditions of the other.

Association: A relationship between two variables where the conditional distribution of one variable depends on the value of the other variable. Indicated when conditional probabilities differ from marginal probabilities.

Independence: Two variables are independent if knowing the value of one variable provides no information about the value of the other variable. Occurs when P(A|B) = P(A) for all values of B.

Complete Two-Way Table Analysis Process:
  1. Data Collection: Gather data on two categorical variables of interest
  2. Table Construction: Organize data into a two-way table with clear row and column labels
  3. Calculate Totals: Compute row totals, column totals, and grand total
  4. Calculate Marginal Proportions: Determine overall proportions for each category
  5. Calculate Conditional Proportions: Determine proportions within each row/column
  6. Visual Inspection: Look for patterns in the data before formal testing
  7. Association Assessment: Compare conditional distributions to marginal distributions
  8. Statistical Testing: Perform chi-square test or other appropriate tests
  9. Effect Size Calculation: Calculate relative risk, odds ratio, or other measures
  10. Interpretation: Draw conclusions about the relationship between variables
  11. Application: Translate findings into practical recommendations
Tip 1: Always identify the explanatory and response variables before analysis.
Tip 2: Look for patterns visually before performing calculations - this helps validate results.
Tip 3: When conditional probabilities differ significantly from marginal probabilities, association likely exists.
Tip 4: Remember that association does not imply causation - consider potential confounding variables.
Tip 5: For small samples (expected frequencies < 5), consider Fisher's exact test instead of chi-square.
Tip 6: Always report both statistical significance and practical significance of findings.
Common errors: Confusing marginal and conditional probabilities, misinterpreting chi-square results, assuming causation from association, using chi-square with small expected frequencies, failing to consider practical significance alongside statistical significance.
Exam preparation: Practice constructing two-way tables, calculating conditional probabilities, interpreting chi-square results, distinguishing association from causation, translating statistical findings into practical recommendations.
Essential Formulas and Rules:

• Conditional Probability: P(A|B) = P(A∩B)/P(B) = (Cell Count)/(Row/Column Total)

• Expected Frequency: E = (Row Total × Column Total)/Grand Total

• Chi-Square Statistic: χ² = Σ[(O - E)²/E]

• Degrees of Freedom: df = (r-1)(c-1) where r = rows, c = columns

• Independence Test: P(A|B) = P(A) if and only if A and B are independent

• Relative Risk: RR = P(outcome|exposed)/P(outcome|unexposed)

• Odds Ratio: OR = (a/c)/(b/d) where a,b,c,d are cell counts in 2×2 table

• Association Strength: Larger differences between conditional and marginal probabilities indicate stronger associations

Two-Way Table Analysis Workflow
📊
Step 1: Data Organization
Arrange data in two-way table with explanatory variable as rows and response variable as columns
Rows: Explanatory Variable
Columns: Response Variable
Step 2: Calculate Totals
Compute row totals, column totals, and grand total
Row Total = Σ(Cell Counts in Row)
Column Total = Σ(Cell Counts in Column)
Step 3: Marginal Probabilities
Calculate overall proportions for each category
P(Category) = Category Total / Grand Total
Step 4: Conditional Probabilities
Calculate proportions within each row/column
P(A|B) = Cell Count / Row/Column Total
Step 5: Association Test
Compare conditional to marginal probabilities
If P(A|B) ≠ P(A), association exists
Step 6: Statistical Test
Perform chi-square test for independence
χ² = Σ[(O - E)²/E], df = (r-1)(c-1)
Analysis Process
1
Organize Data
2
Calculate Totals
3
Marginal Probs
4
Conditional Probs
5
Compare
6
Stat Test
Two-Way Table Structure
Explanatory Variable (Rows)
↓
Response Variable (Columns)
→
Intersection = Cell Count
Association exists when conditional ≠ marginal!

Questions & Answers

Question: How do I distinguish between marginal and conditional probabilities in two-way tables? What's the difference and why does it matter?

Answer: Great question! The distinction is crucial:

  • Marginal Probability: The overall probability of a category regardless of other variables. Calculated as (Row/Column Total)/(Grand Total). Example: P(Student participates) = Total participants / Total students.
  • Conditional Probability: The probability of a category given that we know another condition is true. Calculated as (Cell Count)/(Row/Column Total). Example: P(Student participates | Junior) = Juniors who participate / Total Juniors.

The difference matters because:

  • Marginal probabilities show overall patterns
  • Conditional probabilities show relationships between variables
  • If conditional ≠ marginal, there's an association between variables
  • Conditional probabilities help us make predictions based on known information

Question: When should I use relative risk versus odds ratio? What's the difference and when is each appropriate?

Answer: Both measure association strength but in different contexts:

  • Relative Risk (RR): Compares probabilities directly. RR = P(outcome|exposed) / P(outcome|unexposed). Best for cohort studies where you can calculate actual probabilities.
  • Odds Ratio (OR): Compares odds. OR = (odds in exposed) / (odds in unexposed). Better for case-control studies where probabilities aren't directly calculable.

Key differences:

  • RR ranges from 0 to ∞: RR = 1 (no association), RR < 1 (protective), RR > 1 (risk factor)
  • OR approximates RR when outcome is rare, but diverges when outcome is common
  • OR is symmetric - OR of A given B equals OR of B given A

Use RR when you can calculate actual probabilities, OR for case-control studies.

Question: How do I know if my chi-square test results are reliable? What conditions must be met?

Answer: The chi-square test has important assumptions for reliability:

  • Expected Frequency Rule: At least 80% of cells should have expected frequency ≥ 5, and no cell should have expected frequency < 1
  • Independence: Observations must be independent of each other
  • Categorical Data: Both variables must be categorical
  • Random Sampling: Data should come from random sampling

To check expected frequencies:

  • For each cell: Expected = (Row Total × Column Total) / Grand Total
  • Count how many cells violate the rule (expected < 5)
  • If more than 20% of cells have expected < 5, consider alternatives

Alternatives for small expected frequencies:

  • Fisher's Exact Test (for 2×2 tables)
  • Combine categories if meaningful
  • Use simulation-based methods

Question: What does it mean when we say "association does not imply causation"? How can I identify potential confounding variables?

Answer: This fundamental principle means that just because two variables are associated doesn't mean one causes the other. Examples:

  • Ice cream sales and drowning incidents are associated (both increase in summer), but ice cream doesn't cause drowning
  • Shoe size and reading ability are associated in children, but shoe size doesn't cause better reading

To identify confounding variables:

  • Ask: What other variables might affect both the explanatory and response variables?
  • Look for temporal sequence: Does the "cause" occur before the "effect"?
  • Consider lurking variables: Third variables that explain the association
  • Think about biological/social mechanisms: Is there a plausible pathway?

To establish causation, you typically need experimental studies (randomized controlled trials) rather than observational studies.

Question: How do I interpret the strength of an association in categorical data? Are there measures for strength like correlation coefficients?

Answer: Yes, there are several measures of association strength for categorical data:

  • Phi Coefficient (φ): For 2×2 tables, ranges from -1 to +1, similar to correlation coefficient
  • Cramer's V: For larger tables, ranges from 0 to 1, where 0 = no association, 1 = perfect association
  • Relative Risk/Odds Ratio: For 2×2 tables, magnitude indicates strength (closer to 1 = weaker association)
  • Percent Change: Compare conditional percentages to marginal percentages

Guidelines for Cramer's V interpretation:

  • 0.1 = Small association
  • 0.3 = Medium association
  • 0.5 = Large association

For practical interpretation, focus on the magnitude of differences between conditional and marginal distributions. Larger differences indicate stronger associations.