Association: A relationship between two categorical variables where knowing the value of one variable provides information about the value of the other.
Independence: No relationship exists between variables; conditional relative frequencies are similar across categories.
- Construct the two-way table with observed frequencies
- Calculate conditional relative frequencies for each category
- Compare conditional frequencies across categories
- Determine if frequencies are similar (independence) or different (association)
- Draw conclusions about the relationship
Given regular tutoring: P(Pass) = 80/120 = 0.667 or 66.7%
Given regular tutoring: P(Fail) = 40/120 = 0.333 or 33.3%
Given no regular tutoring: P(Pass) = 140/280 = 0.500 or 50.0%
Given no regular tutoring: P(Fail) = 140/280 = 0.500 or 50.0%
Regular tutoring: 66.7% pass, 33.3% fail
No regular tutoring: 50.0% pass, 50.0% fail
There's a significant difference in pass rates (66.7% vs 50.0%)
Students who attend tutoring regularly have a higher pass rate (66.7%) compared to those who don't (50.0%).
Since conditional relative frequencies differ significantly, there is an association between tutoring attendance and academic performance.
There is a positive association between attending tutoring regularly and passing.
Students who attend tutoring regularly have a 66.7% pass rate compared to 50.0% for those who don't attend regularly.
• Conditional Relative Frequency: Cell value ÷ Row/Column total
• Association Detection: Compare conditional frequencies across categories
• Independence: Conditional frequencies are similar across categories
Medical Association: A relationship between treatment/exposure and outcome variables in clinical studies.
Recovery Rate: The conditional probability of recovery given specific patient characteristics.
Given under 40: P(Recover) = 150/200 = 0.750 or 75.0%
Given under 40: P(Not Recover) = 50/200 = 0.250 or 25.0%
Given 40+: P(Recover) = 225/300 = 0.750 or 75.0%
Given 40+: P(Not Recover) = 75/300 = 0.250 or 25.0%
Under 40: 75.0% recover, 25.0% don't recover
40+: 75.0% recover, 25.0% don't recover
The conditional frequencies are identical!
Since the conditional relative frequencies are identical for both age groups, there is no association between age group and recovery rate.
Age does not appear to influence the treatment effectiveness in this study.
There is no association between age group and recovery rate.
Both age groups have identical recovery rates of 75.0%.
Age does not appear to influence treatment effectiveness.
• Conditional Relative Frequency: Cell value ÷ Row total
• Independence Detection: Identical conditional frequencies suggest independence
• Association Strength: Larger differences indicate stronger association
Admission Rate: The conditional probability of being admitted given gender.
Gender Bias: A systematic difference in treatment or outcomes based on gender.
Given male: P(Admitted) = 240/400 = 0.600 or 60.0%
Given male: P(Rejected) = 160/400 = 0.400 or 40.0%
Given female: P(Admitted) = 300/600 = 0.500 or 50.0%
Given female: P(Rejected) = 300/600 = 0.500 or 50.0%
Male: 60.0% admitted, 40.0% rejected
Female: 50.0% admitted, 50.0% rejected
There's a 10% difference in admission rates
Male applicants have a higher admission rate (60.0%) than female applicants (50.0%).
Since conditional relative frequencies differ, there is an association between gender and admission.
Male applicants are 10 percentage points more likely to be admitted.
This association might suggest potential gender bias in admissions.
Further investigation would be needed to determine if this difference is statistically significant.
There is an association between gender and admission.
Male applicants have a higher admission rate (60.0%) than female applicants (50.0%).
This suggests potential gender bias in the admission process.
• Conditional Relative Frequency: Cell value ÷ Row total
• Association Detection: Different conditional frequencies indicate association
• Implication Analysis: Consider broader implications of detected associations
Two-Way Table: A table displaying the frequency distribution of two categorical variables.
Conditional Relative Frequency: The relative frequency of one variable given a specific value of another variable.
Association: A relationship between variables where conditional frequencies differ across categories.
Independence: No relationship exists between variables; conditional frequencies are similar.
• Conditional Relative Frequency = Cell Value ÷ Row/Column Total
• Marginal Relative Frequency = Row/Column Total ÷ Grand Total
• Association exists when conditional frequencies differ significantly
• Independence exists when conditional frequencies are similar
Customer Satisfaction: The conditional probability of being satisfied given the service type.
Service Quality: Measured by satisfaction rates across different service modalities.
Dine-in: P(Satisfied) = 240/300 = 0.800 or 80.0%
Takeout: P(Satisfied) = 120/200 = 0.600 or 60.0%
Delivery: P(Satisfied) = 90/150 = 0.600 or 60.0%
Dine-in: 80.0% satisfied
Takeout: 60.0% satisfied
Delivery: 60.0% satisfied
Dine-in has significantly higher satisfaction (20% higher than other services)
There is a clear association between service type and customer satisfaction.
Dine-in customers are more satisfied than takeout or delivery customers.
Takeout and delivery have identical satisfaction rates.
Dine-in service has the highest satisfaction rate (80%)
Takeout and delivery services have lower but equal satisfaction rates (60%)
Service type significantly influences customer satisfaction
There is a strong association between service type and customer satisfaction.
Dine-in customers have the highest satisfaction rate (80%).
Takeout and delivery customers have identical satisfaction rates (60%).
• Conditional Relative Frequency: Cell value ÷ Row total
• Association Detection: Compare conditional frequencies across categories
• Pattern Recognition: Look for systematic differences in frequencies
Work Performance: Categorical variable with two levels (high performer, low performer).
Device Effectiveness: Measured by the conditional probability of high performance given device type.
Laptop: P(High Perf) = 210/300 = 0.700 or 70.0%
Tablet: P(High Perf) = 125/250 = 0.500 or 50.0%
Smartphone: P(High Perf) = 100/250 = 0.400 or 40.0%
Laptop: 70.0% high performers
Tablet: 50.0% high performers
Smartphone: 40.0% high performers
Clear trend: laptop > tablet > smartphone for high performance
There is a strong association between device type and work performance.
Laptop users have the highest rate of high performance (70.0%).
Smartphone users have the lowest rate of high performance (40.0%).
Performance decreases as device complexity decreases.
Laptop users are 30 percentage points more likely to be high performers than smartphone users.
This represents a substantial difference in productivity.
Organizations should consider device policies based on these findings.
There is a strong association between device type and work performance.
Laptop users have the highest high-performance rate (70%).
Smartphone users have the lowest high-performance rate (40%).
Device type significantly influences work performance.
• Conditional Relative Frequency: Cell value ÷ Row total
• Association Strength: Larger differences indicate stronger association
• Practical Significance: Consider real-world implications of findings
Complete Summary: Analyzing Association in Two-Way Tables
Two-Way Table: Displays frequency distribution of two categorical variables.
Conditional Relative Frequency: Proportion of one variable given a specific value of another.
Association: Relationship where conditional frequencies differ across categories.
Independence: No relationship exists when conditional frequencies are similar.
Association Detection: Compare conditional relative frequencies across categories.
Independence Criterion: Conditional frequencies are similar across all categories.
Strength of Association: Larger differences indicate stronger association.
Direction of Association: Identify which categories are more likely to co-occur.
Step 1: Construct two-way table with observed frequencies.
Step 2: Calculate conditional relative frequencies for each category.
Step 3: Compare conditional frequencies across categories.
Step 4: Determine if association exists based on frequency differences.
Step 5: Interpret results in context of the problem.
Strong Association: Large differences in conditional frequencies (>20%).
Moderate Association: Moderate differences in conditional frequencies (10-20%).
Weak Association: Small differences in conditional frequencies (<10%).
Independence: Negligible differences in conditional frequencies (<5%).
Association: A relationship between two categorical variables where the conditional relative frequencies of one variable differ depending on the value of the other variable.
Conditional Relative Frequency: The relative frequency of a specific category of one variable given a specific category of another variable.
Independence: Two variables are independent if the conditional relative frequencies of one variable are the same regardless of the value of the other variable.
Contingency Table: Another name for a two-way table showing the relationship between two categorical variables.
Marginal Frequencies: The totals for each row and column in a two-way table.
- Construct Table: Organize data into a two-way table format.
- Calculate Marginals: Find row and column totals.
- Compute Conditional Frequencies: Calculate P(A|B) for each category of B.
- Compare Frequencies: Look for differences in conditional frequencies.
- Determine Association: Decide if variables are associated or independent.
- Interpret Results: Explain the meaning of the association in context.
• Conditional Relative Frequency = Cell Value ÷ Row/Column Total
• Association exists when: P(A|B₁) ≠ P(A|B₂) for different values of B
• Independence exists when: P(A|B₁) ≈ P(A|B₂) for different values of B
• Strength of association increases with larger differences in conditional frequencies
Questions & Answers
Question: How do I know if the difference in conditional frequencies is big enough to claim there's an association? How do I determine if it's just due to chance?
Answer: This is a great question about statistical significance! At the high school level:
Rule of Thumb:
- Differences > 15-20% are usually meaningful
- Differences < 5% are typically not meaningful
- Differences 5-15% require careful consideration
Advanced Statistics:
Later, you'll learn about the chi-square test of independence, which mathematically determines if the observed differences are statistically significant or likely due to chance.
For now, focus on the practical significance: "Is this difference large enough to matter in real life?"
In our examples, a 10-16% difference in pass rates or satisfaction is meaningful in practice.
Question: In Exercise 2, the recovery rates were identical (75% for both age groups). But couldn't this still be due to chance? Maybe with a larger sample, we'd see differences?
Answer: Excellent critical thinking! You're absolutely right:
Sample Size Matters:
- With small samples, identical results might be due to chance
- With large samples, identical results are more likely to represent true independence
Confidence in Results:
The larger your sample size, the more confident you can be that identical conditional frequencies represent true independence rather than chance.
In Exercise 2, with 200 and 300 participants respectively, the identical results (75%) are more convincing than if we had only 10 participants in each group.
For now, we accept the apparent independence, but you're right to think about sample size effects.
Question: In Exercise 3, you found an association between gender and admission. Does this mean the university is discriminating based on gender?
Answer: This is a crucial distinction in statistics:
Association vs. Causation:
Association: Variables tend to occur together (correlation)
Causation: One variable directly causes changes in another
Confounding Variables:
The association might be explained by other factors like:
- Academic major preferences (some majors have different admission rates)
- Test scores or GPA differences by gender
- Application timing or other factors
Further Investigation Needed:
To determine if there's actual discrimination, researchers would need to control for other relevant factors using more sophisticated statistical methods.
Association is a clue that warrants investigation, but doesn't prove discrimination.