01 · Purpose & principles
A comparison tool, not a verdict
Campus Compass is a framework for comparing aggregate student political attitudes. It does not describe every student, judge any orientation, or infer an individual’s beliefs. Every value in the current product is synthetic demonstration data.
Axis labels describe direction, not desirability.
Economic and social attitudes remain separate.
No individual response is published.
02 · Axis definitions
Two normalized axes, from −10 to +10
Inputs are coded so that negative economic values indicate stronger support for redistribution, public provision, and labor protections. Negative social values indicate more progressive positions across the included cultural and civil-liberty issues. These labels apply only to the measured issue set.
03 · Data pipeline
Six repeatable publication gates
Immutable, checksummed source snapshots retain original column names and retrieval metadata.
Exact names, aliases, and reviewed fuzzy matches connect sources to canonical institutions.
Question codebooks transform survey-specific scales into common, directional issue measures.
Survey weights are applied; cells below the privacy threshold are suppressed.
Issue-weighted axis scores, confidence intervals, source shares, and quality tiers are calculated.
Range, missingness, sample-size, outlier, temporal, and cross-source checks gate publication.
04 · Scoring model
Configuration-driven weighted means
Each survey response is transformed through a versioned question mapping. Questions roll into issue scores; issue scores roll into axes. Missing issues are excluded and remaining weights are renormalized.
issue_scoreᵤᵢ = Σ(wq × transformed_responseᵤq) / Σwq
axis_scoreᵤₐ = clamp( Σ(wi × issue_scoreᵤᵢ) / Σwi, −10, +10 )Shrinkage for small samples. Direct university estimates are partially pooled toward a peer-group mean. The shrinkage weight depends on effective sample size and measurement reliability.
05 · Confidence & quality
Precision is earned, not implied
The 0–100 confidence score combines effective sample size, recency, issue coverage, source reliability, weighting quality, imputation, model uncertainty, and cross-source agreement. It is an evidence indicator—not the probability that a point is “correct.”
| Tier | Typical evidence | Display treatment |
|---|---|---|
| A | Strong, recent direct data | Solid point, narrow interval |
| B | Good direct or high-quality blend | Solid or ringed point |
| C | Moderate blended evidence | Ringed point, visible flags |
| D | Mostly estimated | Hollow point, wide interval |
| E | Weak estimate | Hollow, low opacity, warning |
06 · Estimation
Modeled values remain visibly modeled
When direct campus survey data is unavailable, the designed estimator uses interpretable, regularized models trained only on institutions with publishable direct data. Candidate predictors include county context, region, setting, institutional type, enrollment, selectivity, and peer similarity.
- Cross-validation reports MAE, RMSE, R², calibration, and subgroup performance.
- Prediction intervals—not survey standard errors—express modeled uncertainty.
- Estimated scores use hollow points and an explicit warning in every detail view.
- A failed validation threshold yields insufficient data, never a forced estimate.
07 · Validation
Automated checks plus human review
Publication blocks on invalid score ranges, suppressed samples, missing provenance, incompatible methodology links, ambiguous institution matches, schema drift, and failed statistical validation. Large changes and conflicting sources enter a research review queue.
08 · Known limitations
What the compass cannot tell you
- Campus averages do not describe individual students or the diversity within a university.
- Issue selection and question wording influence both axes.
- Local context is not a substitute for student opinion.
- Historical comparisons can break when questionnaires or methodologies change.
- The current scores are fictional demonstrations and must not be cited as research findings.
09 · Version history
Every release remains auditable
Initial synthetic release: two axes, eight issue dimensions, direct/blended/estimated evidence states, confidence intervals, and reproducible configuration.
Return to the compass to inspect evidence, uncertainty, and issue-level details for each demonstration institution.