What is Correlation?
Correlation measures the strength and direction of a relationship between two variables. The most common statistic is the Pearson correlation coefficient r, which ranges from −1 (perfect negative) through 0 (no linear relationship) to +1 (perfect positive). In web analytics, you might check whether Engagement Time moves in tandem with Conversion Rate, or whether CPC tends to rise when ROI falls.
How Correlation is used in web analytics
Analysts use correlation to prioritize hypotheses before running causal tests. For example, a strong correlation between Campaign CPC and CPL (CPL) across Source or Device Category tells you where to dig deeper. Segment by Session attributes, UTM parameters, or cohorts (see Cohort Analysis) to see whether relationships hold in different contexts. Then validate with an experiment (A/B Testing) or with your Attribution Model.
Works the same whether your data comes from GA4, Plausible, Matomo, Simple Analytics, or your warehouse via BigQuery and a BI tool like Power BI.
How is correlation calculated?
- Pearson r (default): linear association of two continuous variables.
- Spearman ρ: rank-based; safer for non-normal data or monotonic but non-linear trends.
Always inspect distributions, consider transformations (e.g., log), and handle outliers.
Pitfalls & good practice
- Correlation ≠ causation. A correlation between Pageview/Screen View volume and Conversion Rate might be driven by seasonality or simultaneous changes. Use A/B Testing to establish causality.
- Confounding & Simpson’s paradox. Recalculate within segments: by Source, Campaign, device, geography, or cohort.
- Aggregation bias. Session-level relationships can vanish at user-level (see Client ID).
- Small samples & outliers. One viral spike can distort r. Check robustness and confidence intervals.
- Non-linearity. Zero correlation doesn’t mean “no relationship”—it may simply be curved.
Quick examples
| Pair (by segment) | Analyst intent |
|---|---|
| Engagement Time ↔ Conversion Rate | Does deeper engagement predict conversions? |
| CPC ↔ ROI by Campaign | Are paid clicks efficient across campaigns? |
| Pageview per Session ↔ Micro-Conversion | Do users who explore more complete key steps? |