Resources

A municipal transit authority analyzes rider satisfaction ratings on a scale from 11 to 1010. When evaluated overall, Bus Route A has an average rider satisfaction score of 7.27.2 across nA=1,000n_A = 1{,}000 daily surveyed riders, whereas Bus Route B has an average satisfaction score of 7.87.8 across nB=1,000n_B = 1{,}000 daily surveyed riders. However, the data is confounded by travel period (Peak Rush Hour vs. Off-Peak). On Route A, 800800 riders were surveyed during Peak hours (mean score =7.0= 7.0) and 200200 riders were surveyed during Off-Peak hours (mean score =8.0= 8.0). On Route B, 300300 riders were surveyed during Peak hours. Assuming that within each travel period Route A scored higher than or equal to Route B, solve backwards for the maximum possible Peak-hour mean satisfaction score yˉB,peak\bar{y}_{B,\text{peak}} for Route B given that its Off-Peak mean satisfaction score was yˉB,off=8.1\bar{y}_{B,\text{off}} = 8.1, and explain how this reverse scenario illustrates Simpson's Paradox.

Answer and steps

Final answer

yˉB,peak=7.1\bar{y}_{B,\text{peak}} = 7.1. Despite Route A outperforming Route B in the Off-Peak stratum (8.08.0 vs. 8.18.1 implies Route B is slightly higher off-peak, but during peak hours Route A achieves 7.07.0 while Route B achieves 7.17.1, or Route A outperforms within comparable weighted conditions), the aggregate mean for Route B (7.87.8) appears higher than Route A (7.27.2) because 70%70\% of Route B riders were surveyed during high-satisfaction Off-Peak hours compared to only 20%20\% for Route A, demonstrating Simpson's Paradox driven by unequal subgroup sampling weights.

Steps

01

Determine the Subgroup Counts for Route B

Route B has a total of nB=1,000n_B = 1{,}000 surveyed riders, divided into Peak and Off-Peak travel periods:

  • Peak riders: nB,peak=300n_{B,\text{peak}} = 300
  • Off-Peak riders: nB,off=1,000−300=700n_{B,\text{off}} = 1{,}000 - 300 = 700
02

Calculate the Peak-Hour Mean for Route B (yˉB,peak\bar{y}_{B,\text{peak}})

The overall mean rating yˉB\bar{y}_B is the weighted average of the subgroup means: yˉB=nB,peak⋅yˉB,peak+nB,off⋅yˉB,offnB\bar{y}_B = \frac{n_{B,\text{peak}} \cdot \bar{y}_{B,\text{peak}} + n_{B,\text{off}} \cdot \bar{y}_{B,\text{off}}}{n_B}

Substitute the given values into the equation: 7.8=300⋅yˉB,peak+700⋅8.11,0007.8 = \frac{300 \cdot \bar{y}_{B,\text{peak}} + 700 \cdot 8.1}{1{,}000}

Multiply both sides by 1,0001{,}000: 7,800=300⋅yˉB,peak+5,6707{,}800 = 300 \cdot \bar{y}_{B,\text{peak}} + 5{,}670

Subtract 5,6705{,}670 from both sides to find the total satisfaction points scored during Peak hours: 300⋅yˉB,peak=7,800−5,670=2,130300 \cdot \bar{y}_{B,\text{peak}} = 7{,}800 - 5{,}670 = 2{,}130

Solve for yˉB,peak\bar{y}_{B,\text{peak}}: yˉB,peak=2,130300=7.1\bar{y}_{B,\text{peak}} = \frac{2{,}130}{300} = 7.1


03

Understand the Mechanism of Simpson's Paradox

Simpson's Paradox occurs when a trend or comparison seen in aggregated data is altered or reversed when the data is separated into subgroups, caused by an underlying confounding variable.

  1. Subgroup Performance vs. Aggregate Performance:

    • Peak hours: Route A scored 7.07.0, while Route B scored 7.17.1.
    • Off-Peak hours: Route A scored 8.08.0, while Route B scored 8.18.1.
    • Within both travel periods, satisfaction ratings are relatively close (with Route B slightly higher by 0.10.1 points in each stratum).
    • Yet, in the aggregate, Route B (7.87.8) appears to drastically outperform Route A (7.27.2) by 0.60.6 points.
  2. The Confounding Variable (Subgroup Weights):

    • Travel period is a confounding variable because satisfaction is inherently higher during Off-Peak hours (8.08.0 to 8.18.1) than during Peak hours (7.07.0 to 7.17.1) across both routes.
    • For Route A, 80%80\% of riders (800/1,000800/1{,}000) were surveyed during low-satisfaction Peak hours, dragging its overall average down to 7.27.2.
    • For Route B, 70%70\% of riders (700/1,000700/1{,}000) were surveyed during high-satisfaction Off-Peak hours, pulling its overall average up to 7.87.8.

The disproportionate sample weights across conditions distort the aggregate result, illustrating Simpson's Paradox.

Typical errors

OverviewACEpath Teacher's Diagnosis

01 The student treats the aggregate mean as an unweighted simple average of the peak and off-peak satisfaction scores, ignoring subgroup sample sizes.

Sample Student Work

Reviewed by ACEpath Teacher
Step-by-Step Review
Walk Through It with Your Teacher
Incorrect
  1. Step 1

    yB,peak+yB,off2=7.8  ⟹  yB,peak+8.12=7.8\frac{y_{B,\text{peak}} + y_{B,\text{off}}}{2} = 7.8 \implies \frac{y_{B,\text{peak}} + 8.1}{2} = 7.8

    Correct

    300⋅yˉB,peak+700⋅8.11,000=7.8\frac{300 \cdot \bar{y}_{B,\text{peak}} + 700 \cdot 8.1}{1{,}000} = 7.8

    You set up the aggregate mean as a simple arithmetic average of the two periods, dividing by 22. However, Route B has 300300 peak riders and 700700 off-peak riders out of 1,0001{,}000 total, so the overall mean must be a weighted average reflecting those unequal sample sizes.

  2. Step 2

    yB,peak+8.1=15.6  ⟹  yB,peak=15.6−8.1=7.5y_{B,\text{peak}} + 8.1 = 15.6 \implies y_{B,\text{peak}} = 15.6 - 8.1 = 7.5

    Correct

    300⋅yˉB,peak+5,670=7,800  ⟹  yˉB,peak=2,130300=7.1300 \cdot \bar{y}_{B,\text{peak}} + 5{,}670 = 7{,}800 \implies \bar{y}_{B,\text{peak}} = \frac{2{,}130}{300} = 7.1

    Your algebraic manipulation was executed cleanly, but the resulting value is incorrect because it inherited the unweighted equation from Step 1.

  3. Step 3

    Simpson's Paradox occurs because Route B has a higher overall average even though Route A scores better in subgroups.

    Correct

    Simpson's Paradox is driven by the confounding variable of travel period: satisfaction is naturally higher off-peak (8.08.0–8.18.1), and Route B surveyed 70%70\% of its riders off-peak compared to only 20%20\% for Route A, inflating Route B's aggregate score.

    You accurately stated the surface-level paradox, but you did not explain the underlying statistical mechanism: the disproportionate subgroup weights (confounding by travel period) that cause the reversal.

WHAT YOU DID WELL
Compliments from Your Teacher

You correctly identified the given aggregate score of 7.87.8 and the off-peak score of 8.18.1.

You solved your linear algebraic equation accurately without arithmetic slips.

You correctly recognized that Simpson's Paradox describes a reversal between aggregate and subgroup trends.

THE ERROR
Where Your ACEpath Teacher Spotted the Mistake
Unweighted Average for Stratified Data

You divided by 22 instead of weighting each subgroup mean by its sample size (300300 and 700700).

WHY IT HAPPENED
ACEpath Teacher's Diagnosis

You treated the aggregate mean as an unweighted simple average of stratum means, overlooking that combined means must be weighted by group sizes, which is the exact mechanism causing Simpson's Paradox.

REASONING PATH
Your ACEpath teacher’s suggested approach
Weighted Mean Formula
  • For subgroups of size n1n_1 and n2n_2 combining to N=n1+n2N = n_1 + n_2, the overall mean is yˉ=n1yˉ1+n2yˉ2N\bar{y} = \frac{n_1 \bar{y}_1 + n_2 \bar{y}_2}{N}.

  • Always multiply each stratum mean by its subgroup frequency before summing and dividing by the total sample size.

Explaining Simpson's Paradox
  • A complete explanation of Simpson's Paradox must name the confounding variable and identify how unequal distribution across subgroups shifts the overall average.

  • Highlight the difference in subgroup proportions between the two groups being compared.

CORRECTED PATH
ACEpath Teacher's Demonstration
  1. Determine subgroup counts for Route B: nB,peak=300n_{B,\text{peak}} = 300 and nB,off=1,000−300=700n_{B,\text{off}} = 1{,}000 - 300 = 700.

  2. Set up the weighted mean equation: 300⋅yˉB,peak+700(8.1)1,000=7.8\frac{300 \cdot \bar{y}_{B,\text{peak}} + 700(8.1)}{1{,}000} = 7.8.

  3. Multiply by 1,0001{,}000 and simplify: 300⋅yˉB,peak+5,670=7,800300 \cdot \bar{y}_{B,\text{peak}} + 5{,}670 = 7{,}800.

  4. Subtract 5,6705{,}670: 300⋅yˉB,peak=2,130300 \cdot \bar{y}_{B,\text{peak}} = 2{,}130.

  5. Solve for the peak mean: yˉB,peak=2,130300=7.1\bar{y}_{B,\text{peak}} = \frac{2{,}130}{300} = 7.1.

Final answer

yˉB,peak=7.1\bar{y}_{B,\text{peak}} = 7.1

NEXT STEPS
Final Takeaways from Your Teacher
  • Review how subgroup sample sizes determine the weights in combined means.

  • Practice identifying the confounding variable in stratified survey data.

OverviewACEpath Teacher's Diagnosis

02 The student sets up the weighted average correctly but makes a division error by dividing by 3 instead of 0.3, yielding an implausible rating.

Sample Student Work

Reviewed by ACEpath Teacher
Step-by-Step Review
Walk Through It with Your Teacher
Partially correct
  1. Step 1

    0.30×yB,peak+0.70×8.1=7.8  ⟹  0.30×yB,peak+5.67=7.8  ⟹  0.30×yB,peak=7.8−5.67=2.130.30 \times y_{B,\text{peak}} + 0.70 \times 8.1 = 7.8 \implies 0.30 \times y_{B,\text{peak}} + 5.67 = 7.8 \implies 0.30 \times y_{B,\text{peak}} = 7.8 - 5.67 = 2.13

  2. Step 2

    yB,peak=2.133=0.71y_{B,\text{peak}} = \frac{2.13}{3} = 0.71

    Correct

    yˉB,peak=2.130.30=7.1\bar{y}_{B,\text{peak}} = \frac{2.13}{0.30} = 7.1

    You divided by 33 instead of the coefficient 0.300.30 (or 310\frac{3}{10}), which shifted the decimal point by one place and produced 0.710.71 instead of 7.17.1.

  3. Step 3

    Unequal subgroup weights create Simpson's Paradox because Route B surveyed 70% off-peak versus Route A surveying only 20% off-peak.

WHAT YOU DID WELL
Compliments from Your Teacher

You correctly calculated the peak and off-peak weights (0.300.30 and 0.700.70) for Route B.

You accurately set up and simplified the linear equation for the weighted mean.

You clearly explained how disproportionate sampling across conditions produces Simpson's Paradox.

THE ERROR
Where Your ACEpath Teacher Spotted the Mistake
Decimal place value division error

You divided 2.132.13 by 33 rather than by 0.300.30.

WHY IT HAPPENED
ACEpath Teacher's Diagnosis

Dropping the leading decimal point when dividing by a decimal coefficient shifts the quotient by a factor of 1010, producing an answer outside the reasonable range of the problem scale (11 to 1010).

REASONING PATH
Your ACEpath teacher’s suggested approach
Isolate with the Exact Coefficient
  • When solving c⋅x=dc \cdot x = d, always divide both sides by cc, preserving all decimal places: x=dcx = \frac{d}{c}.

  • Divide 2.132.13 by 0.300.30, which is equivalent to 21.33=7.1\frac{21.3}{3} = 7.1.

Contextual Feasibility Check
  • A weighted average of two values must lie strictly between the minimum and maximum subgroup values.

  • Since the aggregate is 7.87.8 and the off-peak mean is 8.18.1, the peak mean must be close to 7.87.8, so a score of 0.710.71 on a 11–1010 scale is impossible.

CORRECTED PATH
ACEpath Teacher's Demonstration
  1. Divide both sides by 0.300.30: yˉB,peak=2.130.30\bar{y}_{B,\text{peak}} = \frac{2.13}{0.30}

  2. Multiply the numerator and denominator by 1010 to clear decimals: 21.33=7.1\frac{21.3}{3} = 7.1

Final answer

yˉB,peak=7.1\bar{y}_{B,\text{peak}} = 7.1

NEXT STEPS
Final Takeaways from Your Teacher
  • Always perform a reality check: ensure subgroup means lie within the valid scale range (11 to 1010) and bracket the overall mean.

  • When dividing by a decimal like 0.300.30, convert it to a fraction (310\frac{3}{10}) or multiply the numerator and denominator by 1010 first to avoid dropping powers of 1010.

OverviewACEpath Teacher's Diagnosis

03 The student solves for the peak mean correctly but attributes Simpson's Paradox to extreme outliers or respondent bias rather than disproportionate stratum weights.

Sample Student Work

Reviewed by ACEpath Teacher
Step-by-Step Review
Walk Through It with Your Teacher
Partially correct
  1. Step 1

    300⋅yB,peak+700⋅8.11,000=7.8  ⟹  300⋅yB,peak+5,670=7,800  ⟹  300⋅yB,peak=2,130  ⟹  yB,peak=7.1\frac{300 \cdot y_{B,\text{peak}} + 700 \cdot 8.1}{1{,}000} = 7.8 \implies 300 \cdot y_{B,\text{peak}} + 5{,}670 = 7{,}800 \implies 300 \cdot y_{B,\text{peak}} = 2{,}130 \implies y_{B,\text{peak}} = 7.1

  2. Step 2

    This demonstrates Simpson's Paradox because extreme outlier scores from dissatisfied peak-hour commuters pulled Route A's overall rating down. The disparity is caused by measurement bias among rush hour riders.

    Correct

    Simpson's Paradox occurs because travel period is a confounding variable with disproportionate subgroup weights. Route B has 70%70\% of its sample in the higher-satisfaction Off-Peak stratum (8.18.1), pulling its overall average up to 7.87.8. In contrast, Route A has 80%80\% of its sample in the lower-satisfaction Peak stratum (7.07.0), dragging its overall average down to 7.27.2. The overall gap (7.87.8 vs. 7.27.2) is an artifact of unequal subgroup sample sizes, not outliers or measurement bias.

    Simpson's Paradox is a mathematical consequence of combining groups with unequal sample weights across strata, not an artifact of outliers or measurement errors.

WHAT YOU DID WELL
Compliments from Your Teacher

You correctly identified the subgroup sample sizes for Route B (300300 peak and 700700 off-peak riders).

You accurately set up the weighted average equation and solved for yˉB,peak=7.1\bar{y}_{B,\text{peak}} = 7.1.

THE ERROR
Where Your ACEpath Teacher Spotted the Mistake
Misidentifying the Mechanism of Simpson's Paradox

You attributed the reversal to outliers and measurement bias instead of unequal subgroup sample weights.

WHY IT HAPPENED
ACEpath Teacher's Diagnosis

Confusing structural confounding (unequal distribution across subgroups with different baseline means) with data quality issues (such as outliers, unrepresentative respondents, or measurement error). In Simpson's Paradox, every individual measurement can be perfectly accurate, but the aggregate average is dominated by the stratum that makes up the majority of each group.

REASONING PATH
Your ACEpath teacher’s suggested approach
Mathematical Origin of Simpson's Paradox
  • An aggregate mean is yˉ=w1yˉ1+w2yˉ2\bar{y} = w_1 \bar{y}_1 + w_2 \bar{y}_2, where w1,w2w_1, w_2 are stratum proportions.

  • Check whether group weights (wiw_i) differ significantly between the comparison groups before suspecting outliers.

Identifying Confounding Variables
  • A confounding variable is associated with both the grouping factor (route) and the outcome (satisfaction).

  • Verify if one route is sampled primarily during high-performing conditions while the other is sampled primarily during low-performing conditions.

CORRECTED PATH
ACEpath Teacher's Demonstration
  1. Compare stratum proportions: Route A is 80%80\% Peak (800/1,000800/1{,}000) and 20%20\% Off-Peak (200/1,000200/1{,}000), while Route B is 30%30\% Peak (300/1,000300/1{,}000) and 70%70\% Off-Peak (700/1,000700/1{,}000).

  2. Compare within-stratum performance: Route B leads Route A by just 0.10.1 points in both strata (7.17.1 vs. 7.07.0 Peak; 8.18.1 vs. 8.08.0 Off-Peak).

  3. Explain the aggregate distortion: Route B's overall average is heavily weighted by the high Off-Peak score (70%70\% at 8.18.1), whereas Route A's overall average is heavily weighted by the lower Peak score (80%80\% at 7.07.0), inflating Route B's aggregate lead to 0.60.6 points.

Final answer

yˉB,peak=7.1\bar{y}_{B,\text{peak}} = 7.1. Simpson's Paradox is illustrated because the large aggregate difference (7.87.8 vs. 7.27.2) is driven by unequal sampling weights across the confounding travel periods (70%70\% Off-Peak for Route B vs. only 20%20\% for Route A), rather than a large difference in route performance within either period.

NEXT STEPS
Final Takeaways from Your Teacher
  • Review the definition of Simpson's Paradox as a weighting artifact of confounding variables rather than data error.

  • Practice calculating weighted averages with identical subgroup means but shifted subgroup proportions to see how aggregate trends invert.

Lock it in

Similar question

01

A clinic tests recovery rates under Treatment 1 and Treatment 2. Treatment 1 treats 4040 mild cases with an 80%80\% recovery rate and 1010 severe cases with a 40%40\% recovery rate. Treatment 2 treats 1010 mild cases with an 80%80\% recovery rate and 4040 severe cases with a 40%40\% recovery rate. Calculate the aggregate recovery rate for each treatment and identify the confounding variable.

Answer

Treatment 1 aggregate rate: 40(0.80)+10(0.40)50=32+450=72%\frac{40(0.80) + 10(0.40)}{50} = \frac{32 + 4}{50} = 72\%. Treatment 2 aggregate rate: 10(0.80)+40(0.40)50=8+1650=48%\frac{10(0.80) + 40(0.40)}{50} = \frac{8 + 16}{50} = 48\%. The confounding variable is disease severity (mild vs. severe), because severity affects recovery and is disproportionately assigned between treatments.

02

In a company with two divisions, Division X has 200200 entry-level workers with an average salary of $40,000\text{\char36}40{,}000 and 5050 managers with an average salary of $100,000\text{\char36}100{,}000. Division Y has 5050 entry-level workers averaging $42,000\text{\char36}42{,}000 and 200200 managers averaging $102,000\text{\char36}102{,}000. Within each job level, Division Y pays higher salaries. Compute the overall average salary for each division to determine whether Simpson's Paradox occurs.

Answer

Division X overall mean: 200(40,000)+50(100,000)250=13,000,000250=$52,000\frac{200(40{,}000) + 50(100{,}000)}{250} = \frac{13{,}000{,}000}{250} = \text{\char36}52{,}000. Division Y overall mean: 50(42,000)+200(102,000)250=22,500,000250=$90,000\frac{50(42{,}000) + 200(102{,}000)}{250} = \frac{22{,}500{,}000}{250} = \text{\char36}90{,}000. Yes, Simpson's Paradox occurs in appearance because Division Y looks vastly higher overall ($90,000\text{\char36}90{,}000 vs $52,000\text{\char36}52{,}000) mainly because 80%80\% of its staff are managers, compared to only 20%20\% in Division X.

03

A university department admits students into two majors, Arts and Engineering. Across the department, 1,2001{,}200 female and 1,2001{,}200 male applicants apply. In Arts, the female admission rate is 60%60\% and the male admission rate is 55%55\%. In Engineering, the female admission rate is 25%25\% and the male admission rate is 20%20\%. If 1,0001{,}000 women apply to Arts and 200200 to Engineering, while 300300 men apply to Arts and 900900 to Engineering, find the overall admission rate for each gender and explain how Simpson's Paradox manifests.

Answer

Overall female admission rate: 1,000(0.60)+200(0.25)1,200=600+501,200=6501,200≈54.17%\frac{1{,}000(0.60) + 200(0.25)}{1{,}200} = \frac{600 + 50}{1{,}200} = \frac{650}{1{,}200} \approx 54.17\%. Overall male admission rate: 300(0.55)+900(0.20)1,200=165+1801,200=3451,200=28.75%\frac{300(0.55) + 900(0.20)}{1{,}200} = \frac{165 + 180}{1{,}200} = \frac{345}{1{,}200} = 28.75\%. Simpson's Paradox occurs because within both majors females have a higher acceptance rate (60%>55%60\% > 55\% and 25%>20%25\% > 20\%), and the overall gap is magnified to 54.17%54.17\% vs 28.75%28.75\% because women predominantly applied to the high-admit major (Arts, 83.3%83.3\% of female applicants) while men applied predominantly to the low-admit major (Engineering, 75%75\% of male applicants).

Loading workspace…