Interactive results site · Results · Datasets · Methods · Limits · Explanations
This project estimates where sex differences in score variability appear across 17 live datasets. The public README is generated from the same normalized bundle as the site pages, so counts and tables stay aligned.
The current public bundle includes NIH Collaborative Perinatal Project outputs for core cognition and growth trajectories. These CPP rows are visible in the dataset inventory and cell explorer, and currently remain method-limited rather than headline-eligible.
What "Headline claim", "Supporting evidence", "Provisional", and "Method-limited" mean
- Headline claim: Core confirmatory rows with the strongest survey-design handling and no fallback flags. Only these rows feed the main headline claim.
- Supporting evidence: Inferential rows kept for added age or domain coverage, but still kept separate from the main headline claim.
- Provisional: Rows that depend on fallback weights or other temporary inference paths. They stay visible, but they do not count as headline evidence.
- Method-limited: Rows with usable point estimates but weaker uncertainty estimates, usually because only simple effective-sample-size approximations were available.
- QA only: Rows retained for transparency and diagnostics only, not for inferential claims.
| Metric | Value |
|---|---|
| Headline-eligible confirmatory cells | 67 |
| Share male-greater | 85% |
| Median variance ratio | 1.15x |
| Mean variance ratio | 1.14x |
| Range | 0.10x to 1.46x |
| Datasets contributing | 7 |
- Strongest positive: Numeracy in PIAAC cycle 2, age 60-65 (VR 1.46x)
- Strongest counterexample: Reading achievement in ECLS-K:2011, age K (VR 0.10x)
- Supporting evidence: 263 inferential rows from NHANES, HRS, and PSID remain separate from the headline claim
| Dataset | Trait | Age | VR | CI low | CI high | Claim status |
|---|---|---|---|---|---|---|
| ECLS-K:2011 | Reading achievement | K | 0.10x | 0.10x | 0.10x | Headline claim |
| PIAAC cycle 2 | Numeracy | 60-65 | 1.46x | 1.21x | 1.76x | Headline claim |
| PIAAC cycle 2 | Numeracy | 30-34 | 1.44x | 1.13x | 1.84x | Headline claim |
| PIAAC cycle 2 | Adaptive problem solving | 45-49 | 1.41x | 1.02x | 1.95x | Headline claim |
| PIAAC cycle 2 | Numeracy | 45-49 | 1.38x | 1.00x | 1.91x | Headline claim |
| PIAAC cycle 2 | Adaptive problem solving | 35-39 | 1.38x | 0.99x | 1.93x | Headline claim |
| PIAAC cycle 2 | Literacy | 60-65 | 1.38x | 1.13x | 1.69x | Headline claim |
| ECLS-K:2011 | Reading achievement | 4 | 1.37x | 1.25x | 1.50x | Headline claim |
| Dataset | Claim status | Cells | With CI | Headline rows | Median VR | % Male-greater |
|---|---|---|---|---|---|---|
| CPP core cognition | Method-limited | 4 | 4 | 0 | 1.00x | 50% |
| CPP growth trajectories | Method-limited | 15 | 15 | 0 | 1.01x | 53% |
| ECLS-K:2011 | Headline claim | 32 | 26 | 23 | 1.18x | 85% |
| HRS public | Provisional | 182 | 91 | 0 | 0.97x | 40% |
| HSLS:09 | Headline claim | 2 | 2 | 2 | 1.14x | 100% |
| ICILS 2023 | Supporting evidence | 2 | 2 | 2 | 1.21x | 100% |
| NHANES selected cycles | Supporting evidence | 472 | 132 | 132 | 1.04x | 58% |
| NLSY79 Child and Young Adult | Provisional | 23 | 5 | 0 | 1.38x | 100% |
| NLSY79 main | Method-limited | 18 | 18 | 0 | 1.23x | 89% |
| NLSY97 main | Method-limited | 11 | 11 | 0 | 1.28x | 100% |
| NNYFS 2012 | Supporting evidence | 19 | 14 | 14 | 1.01x | 50% |
| PIAAC cycle 2 | Headline claim | 30 | 30 | 30 | 1.14x | 87% |
| PIRLS 2021 | Headline claim | 3 | 3 | 3 | 0.99x | 33% |
| PISA 2022 | Headline claim | 3 | 3 | 3 | 1.25x | 100% |
| PSID CDS / TAS | Provisional | 112 | 40 | 0 | 1.10x | 68% |
| TIMSS 2019 | Headline claim | 4 | 4 | 4 | 1.12x | 100% |
| TIMSS 2023 | Headline claim | 4 | 4 | 4 | 1.13x | 100% |
The generated bundle now exports strongest male-greater rows, strongest female-greater rows, closest-to-equal rows, largest-N rows, widest-CI rows, and per-dataset variance-ratio quantiles and histograms.
The pipeline computes sex-specific weighted variances within dataset-defined cells. Public counts and display tables are generated from the normalized cross-dataset table rather than hand-maintained page content.
Run python scripts/run_paper_bundle.py after the backend comparison build to regenerate the site bundle, public pages, and README together.
MIT