Limits

Descriptive, Not Causal

This project documents observed patterns in score variability. It does not and cannot establish why those patterns exist. Observed variance on a particular instrument in a particular population at a particular time reflects the combined influence of measurement properties, educational exposure, selection into the sample, and many other factors that cannot be separated with observational data alone.

The results do not support claims about innate capacity, biological determinism, or essential sex differences. They describe what public-use test scores look like, not what people are capable of.

Public-Use Data Constraints

All datasets used here are public-use files, which impose several constraints:

• Some survey design variables (PSU, stratum IDs) are masked or coarsened in public releases, limiting the precision of design-aware inference.
• Geographic identifiers are often suppressed, preventing sub-national analysis.
• Small-cell suppression rules may exclude the tails of the distribution, which are precisely the part of the distribution most relevant to variability differences.
• Some datasets require registration-based access, and their terms of use prohibit redistribution of the raw microdata.

These constraints are why the project uses an evidence-tier system: datasets with stronger design metadata get stronger labels, while those with weaker public-use infrastructure are marked as method-limited or provisional.

Bounded and Heaped Scales

Some measures have hard floor or ceiling effects. For example:

• The NHANES adult cognition screen (CFDDS) has a narrow scoring range that mechanically compresses variance, especially in older-adult age bands where many respondents cluster near the ceiling.
• NLSY Child/YA PIAT and PPVT scores can pile up at scale boundaries in certain age bands.
• Kindergarten reading scores in ECLS-K show floor effects at school entry.

Cells affected by bounded-scale issues are flagged as QA-only or carry QA flags so that they are not mistaken for clean estimates. However, even some inferential cells may have residual scale-boundary effects that are difficult to detect automatically.

Why Some Datasets Are Supporting Rather Than Headline

Not all 14 datasets contribute to the headline claim. The main reasons a dataset stays below headline quality:

Method-limited inference

The NLSY datasets and PSID CDS/TAS use simple effective-sample-size SE approximations instead of proper replicate weights. Their point estimates may be reasonable, but the uncertainty quantification is weaker than what the headline datasets provide.

Weight-related caveats

Some PSID TAS rows require alternate public weight fallback because the labeled cross-sectional weight is not available or not appropriate for that wave. These rows are marked provisional.

Secondary or exploratory priority

NHANES physical traits and NNYFS are classified as secondary or exploratory because physical traits (height, weight, grip strength) are a different domain from the cognitive and achievement traits that are the project’s primary focus. They broaden coverage but are kept structurally separate.

The ECLS-K Kindergarten Reading Reversal

By later grades, ECLS-K reading achievement shifts to male-greater variability (log VR around +0.17 to +0.31 by grades 2–4). This developmental reversal means:

• The variability pattern for a given trait can change substantially across age or grade within the same longitudinal sample.
• A claim that “males are more variable” in reading is age-dependent and instrument-dependent, not a stable universal.
• Any public summary that omits this counterexample would be misleading.

This is why the project design requires the ECLS-K kindergarten reading reversal to remain visible in all public-facing materials.

What This Project Does Not Claim

• No biological universality. Observed score variance on public-use instruments does not equal innate biological variability.
• No essentialist conclusions. The project does not claim that sex differences in variability are fixed, immutable, or reflective of fundamental capacities.
• No claim of stability across instruments. A result on one measure in one population at one age does not generalize to other measures, populations, or ages without explicit evidence.
• No claim that supporting evidence confirms the headline. Supporting and provisional datasets are retained for transparency and domain breadth, not because they reinforce the headline pattern.
• No pooling of incomparable instruments. The project does not combine raw scores from different instruments into a single pooled estimate.