A success rate divides successes by attempts. An overall rate weights each subgroup by its share of attempts; unequal group sizes matter.
standardization
Comparing groups under a stated common composition rather than their different observed mixes.
confounding
A difference linked to both the comparison and outcome that can provide an alternative explanation.
01 · TRY
Can the higher overall success rate identify the more effective method?
Compute all six rates. Explain why B leads overall while A leads within each difficulty group. Use the research widget to compare the observed mix with equal task weights and explain why standardization is not random assignment.
Specify the target task population and justify a standardization weight. Compare the observed estimate with at least two target mixes and explain unmeasured confounding and uncertainty.
Make a provisional claim
Loading saved evidence…
Inspect evidence and test an explanation
The investigation controls appear when your local evidence is ready.
04 · UNDERSTAND
Connect evidence and mechanism
When an overall result reverses within groups
An aggregate combines subgroup performance with subgroup composition. When methods receive different mixes of easy and hard tasks, an overall comparison can reverse both subgroup comparisons. This is often called Simpson’s paradox; the arithmetic is consistent because the weights differ.
Inspect the task mix
Method A: easy tasks 18 successes from 20 attempts; hard tasks 32 successes from 80 attempts. Total: 50 successes from 100 attempts.
Method B: easy tasks 64 successes from 80 attempts; hard tasks 6 successes from 20 attempts. Total: 70 successes from 100 attempts.
These are fictional teaching records, not clinical or population evidence.
Work through the rates
A: 18/20 = 90%, 32/80 = 40%, 50/100 = 50%. B: 64/80 = 80%, 6/20 = 30%, 70/100 = 70%. Equal task weights give A (90+40)/2 = 65% and B (80+30)/2 = 55%.
Compare and revise
Compute all six rates. Explain why B leads overall while A leads within each difficulty group. Use the research widget to compare the observed mix with equal task weights and explain why standardization is not random assignment.
Limits
Difficulty adjustment only addresses the recorded grouping. Other differences in participants, selection or measurement may remain; neither the original total nor the adjusted number alone proves a causal method effect.
Transfer
A new study labels difficulty after seeing whether each task succeeded. Can its adjusted comparison be interpreted in the same way?
These authentic 2023 and 2024 Current Population Survey annual estimates describe the U.S. civilian noninstitutional population age 16 and over. Labor-force participation is 100×labor force/population, not unemployment or employment. Counts are in thousands. Five disjoint bands cover 16+: 16–19, 20–24, 25–54, 55–64, 65+. Nested ages, sex groups and race groups are not added again.
Predict: if age-specific participation changes while the age mix also changes, must the overall rate move in the same direction? Keep the earlier fictional Simpson example separate: these real records need not reproduce every part of that reversal.
Observed band-derived rate uses each year's own population weights. Standardized rate uses the selected same five weights for both years: sum(weight×age-specific participation). Equal band weights do not mean equally many people at each single year of age; these bands have unequal widths and the 65+ band is open-ended.
Fixed 0–100% scale for observed band-derived and shared-weight rates; the full scale keeps small changes in perspective. Exact values follow in text.
2023: reconcile counts and weighting
Summed bands: population 266,942, labor force 167,117 thousand; band-derived participation=62.604236%. Shared-weight rate=62.604236%. Source total row reports population 266,942, labor force 167,116 thousand and published participation 62.6%.
Band sum minus published total: population 0, labor force 1 thousand. Counts and published rates are rounded; these discrepancies are preserved, not repaired or attributed to an invented revision. Derived precision is arithmetic, not additional survey precision.
Age
Population (000s)
Labor force (000s)
Derived rate%
Published rate%
Own weight%
Shared weight%
Contribution pp
16 to 19 years
17223
6360
36.927365
36.9
6.451963
6.451963
2.38254
20 to 24 years
22116
15769
71.30132
71.3
8.284946
8.284946
5.907276
25 to 54 years
127941
106541
83.27354
83.3
47.928389
47.928389
39.911666
55 to 64 years
41402
27251
65.820492
65.8
15.509736
15.509736
10.208585
65 years and over
58260
11196
19.217302
19.2
21.824966
21.824966
4.19417
2024: reconcile counts and weighting
Summed bands: population 268,572, labor force 168,107 thousand; band-derived participation=62.592899%. Shared-weight rate=62.83769%. Source total row reports population 268,571, labor force 168,106 thousand and published participation 62.6%.
Band sum minus published total: population 1, labor force 1 thousand. Counts and published rates are rounded; these discrepancies are preserved, not repaired or attributed to an invented revision. Derived precision is arithmetic, not additional survey precision.
Age
Population (000s)
Labor force (000s)
Derived rate%
Published rate%
Own weight%
Shared weight%
Contribution pp
16 to 19 years
17586
6485
36.875924
36.9
6.547965
6.451963
2.379221
20 to 24 years
21387
15293
71.506055
71.5
7.963228
8.284946
5.924238
25 to 54 years
128642
107522
83.582345
83.6
47.898515
47.928389
40.059671
55 to 64 years
41221
27167
65.905728
65.9
15.348212
15.509736
10.221804
65 years and over
59736
11640
19.485737
19.5
22.24208
21.824966
4.252755
2024−2023 observed band-derived change=-0.011337 percentage points. With shared weights, change=0.233454 percentage points. A percentage-point difference is not a percent change. This contrast shows sensitivity to weights; it does not identify a causal effect of aging, a policy or individual behavior.
Transfer: change the standard, retain the evidence
Compare the 2023 standard, 2024 standard and artificial equal standard. Explain why the standardized level changes, whether the change's sign is stable here, and why selecting a convenient standard after viewing results is poor reporting. Identify which within-band composition and survey issues remain even when broad age weights are fixed. Do not infer that every age band's rate increased.
Reasoning guide
Band-derived rates are 62.604236% in 2023 and 62.592899% in 2024: change −0.011337 pp. Using 2023 shares gives 62.604236% and 62.837690%, change +0.233454 pp. Using 2024 shares gives 62.359263% and 62.592899%, change +0.233636 pp. Equal-band rates are 55.308004% and 55.471158%, change +0.163154 pp. Shared weights give a positive difference across these standards, without proving a causal effect or statistical significance.
The published total rate is 62.6% in both years, a rounded headline. The broad 25–54 band can change internally, and annual January population-control updates affect cross-year comparisons. Standardization changes the descriptive question; it is not random assignment, an untouched test or a forecast of a population with different ages.
Primary sources, extraction and uncertainty
BLS 2023 annual Table 3 and BLS 2024 annual Table 3, retrieved 29 October 2026. All-sex/all-race TOTAL block only; public federal statistical data. Local extraction retains raw-source hashes, exact labels, source totals and canonical JSON hash. Source notes say population controls are updated annually; estimates are survey estimates, not population census counts. This selected table supplies no standard errors for our derived contrast; no significance claim or fabricated interval is made.
Copy the age-weighting report
Writing is temporary, 3,000 characters per field. Reset restores 2023 weights/blank writing, retaining guide/report exposure this mount; reload clears it. Authentic source data and completed writing do not establish independent mastery.
Worked example
A: 18/20 = 90%, 32/80 = 40%, 50/100 = 50%. B: 64/80 = 80%, 6/20 = 30%, 70/100 = 70%. Equal task weights give A (90+40)/2 = 65% and B (80+30)/2 = 55%.
