Correlation
See how closely a client's metrics move together, whether that is holding, and how long one takes to show up in another.
Overview
The Correlation screen compares a client's metrics over time, such as ad spend and new-customer revenue. It shows how closely they move together, whether that is holding lately, and how long one takes to show up in the other. Open a client and choose Correlation.
Correlation reads the client's BigQuery dataset. If the client does not have one yet, see Exports.
Ask a question
Start from one of the suggested questions, such as Does spend bring visits?, or set it up yourself:
- Metrics: pick 2 to 5. The first is the one that leads in the lead time scan.
- Window: the last 30, 90, 180 or 365 days.
- Grain: compare by Day, Week or Month.
- Rolling window: how many points each "lately" reading uses. Leave it empty for the default. 0 turns it off.
- Max lag: how far to shift one metric when looking for a lead time. Leave it empty for the default.
- First-time lookback (days): for new-customer metrics, the history checked before the window so returning customers do not count as new.
Read the result
With more than two metrics, choose the Pair to look at. For each pair:
- Overall: the correlation (r) from -1 to 1, with its strength and direction. Near 1 means the metrics rise together. Near -1 means one rises as the other falls. Near 0 means no clear link.
- Lately: the correlation over the most recent rolling window, and whether it is Holding, Weakening or Strengthening.
- Lead time: how long the first metric takes to show up in the second, or None.
- Trend: both metrics over the window.
- Correlation over time: the rolling correlation.
- A table of r at each shift, with the Best one marked.
Ask agent asks the assistant in the chat panel to explain the result.
Warnings
| Warning | Meaning |
|---|---|
| Too few points | There is not enough data for a reliable answer. Try a longer window or a finer grain. |
| A series never changes | One metric is flat, so it cannot correlate. |
| A metric starts or stops mid-window | One metric has data for only part of the window. |
| No real lead time | The best shift barely beats no shift. The metrics move together in the same period, with no delay. |
| The peak may be further out | The best shift is at the edge of the scan. Try a larger max lag. |
| A few outliers drive the fit | A few unusual points account for most of the correlation. |
Correlation shows that metrics move together. It does not prove that one causes the other. To measure what ads really add, see Experiments.
For developers
See Correlation Analysis.