Confidence Levels: Suggestive, Moderate, and Strong

Automatic Confidence Classification

After every Correlation Sweep, SentiLab assigns one of three confidence badges to the result. These badges give you an instant quality assessment so you can decide how much weight to give the finding.

Strong (Green Badge)

A result earns the Strong confidence badge when all three conditions are met:

• n ≥ 200 — Large sample size providing statistical power
• reliableBest = true — Survived Benjamini-Hochberg FDR correction
• |r| > 0.2 — Meaningful effect size, not just statistically significant

The interpretation: "Large sample, survives multiple testing correction. High confidence." You can rely on this result for informing trading decisions, though it should still be combined with other analysis (see Risk Management and Sentiment Analysis).

Suggestive (Yellow Badge)

A result receives the Suggestive badge when:

• n ≥ 50 — Minimum viable sample size
• reliableBest = true OR |r| > 0.15 — Either survived FDR correction or shows a non-trivial effect size

The interpretation: "Moderate evidence — interpret with caution. Consider alongside other factors." This result is worth noting but should not be the sole basis for a decision. Treat it as a hypothesis to validate with more data.

Likely Noise (Red Badge)

If a result fails to meet even the Suggestive criteria, it receives the Likely Noise badge:

• Too few data points (n < 50)
• No statistical significance (failed FDR correction)
• Negligible effect size (|r| < 0.15)

The interpretation: "Too few data points or no statistical significance. Do not rely on this result." This does not necessarily mean no relationship exists — it means the current data is insufficient to detect one reliably.

Upgrading from Suggestive to Strong

If your result lands on Suggestive and you want to push it toward Strong, here are three strategies:

1. Increase data points: Extend the time range from 30 days to 90+ days. More data narrows confidence intervals and increases statistical power.
2. Improve correlation strength: Apply quality filters (minQuality ≥ 0.8, Predictive Only) to reduce noise. Cleaner data often yields stronger correlations.
3. Achieve FDR confirmation: If your best lag barely missed FDR correction, more data points may push it past the threshold.

Common Pitfalls

Confidence badges help, but you should still watch for these traps:

Pitfall 1: Statistically Significant but Practically Meaningless

A result with p < 0.05 and |r| = 0.05 is technically statistically significant — the relationship is unlikely due to chance. But with a correlation of just 0.05, sentiment explains only 0.25% of price variance. This is practically useless for trading decisions.

Pitfall 2: Looks Good but Too Few Data Points

A result showing |r| = 0.4 with n = 25 looks promising on the surface — moderate correlation. But with only 25 data points, the confidence interval is extremely wide. The true correlation could easily be 0.0 or 0.7. You simply cannot tell with so little data.

Pitfall 3: Edge Lag Warning

If the best lag falls at the edge of the tested range (e.g., exactly -24h or +24h when using the Standard ±24h range), the true optimum may be beyond the tested range. In this case, re-run the analysis with the Extended (±72h) or Weekly (±168h) lag range to see if a stronger correlation exists at a more extreme offset.

Why This Matters

Confidence badges save you from the most common analytical mistake: treating all results equally. A Strong result and a Likely Noise result require fundamentally different responses. By understanding the thresholds, you can quickly triage your findings and focus your attention where the evidence is most compelling. Next, learn how to dig deeper into results with the Top Deviations Table.