Could a data study compare predictive accuracy?

Yes, in principle—but this calculator does not run that study. Its results are broad symbolic descriptions, not predictions scored against observed outcomes. To compare methods fairly, each tradition would need to make the same clearly defined prediction before the outcome is known.

1. Choose measurable outcomes. Define the target and time window in advance—for example, a validated personality scale or a specific event outcome. Avoid vague descriptions that can be reinterpreted after the fact.
2. Predefine each method. Record the exact chart inputs, rules, and prediction for every system before outcome data are revealed. Use the same participants and scoring rules across methods.
3. Compare with baselines. Test each astrological method against a simple chance or base-rate prediction and a regression model using relevant non-astrological variables. Regression can test whether birth-chart features add predictive information after those variables are considered.
4. Test on unseen data. Keep a test sample untouched while building or tuning models. Report out-of-sample accuracy, uncertainty intervals, and model fit diagnostics; an in-sample fit or high R-squared alone is not enough.
5. Separate prediction from causation. A reproducible association or improved forecast would not by itself show that celestial positions caused an outcome. Causal claims need a separate design and evidence supporting the causal assumptions.

A credible study would also preregister its hypotheses, account for multiple comparisons, disclose missing data and exclusions, and protect participants’ birth and outcome data. No such participant dataset is collected by this page.

NIST: model validation and why R-squared is not enough · Scikit-learn: cross-validation and held-out testing · American Statistical Association: interpreting p-values