Skip to main content
All CollectionsFrequently Asked Questions
How to Use the Top Correlations Widget
How to Use the Top Correlations Widget

How to get proactive insights for your data

Ryan Stuart avatar
Written by Ryan Stuart
Updated over a week ago

The Top Correlations widget on the dashboard is a tool that you can use to quickly find the strongest relationships across all of your themes and segments.

How to use the Top Correlations Widget

In the widget, you will see a pair of any combination of segments, themes, and numerical fields along with the Odds Ratio value, and frequencies of each pair value.

Odds Ratio Value

To gauge correlation we use Odds Ratio, which is a statistical measure for determining the strength of the association between two categorical variables. Fundamentally, it looks at the ratio between two things occurring together versus apart. For example, if we take the first pair "Checkouts // Self Service" and "Banner: SUPER STORE", the correlation tells us that customers that shop at the SUPER STORE Banner are 7.14X more likely to mention Checkouts // Self Service compared to other customers. This means we can quickly look at the Odds Ratio value to determine how much more likely it is that a pair occurs with each other compared to normal. For more information, hovering over any pair will give you additional statistics about the pair.

Note: We filter all results so we only show those that are statistically significant (p-value < 0.05).

Sorting

The default order of pairs is sorted by absolute values for the correlation value, but that can be changed by clicking on the title for the column.

Configuration

You can configure the specific fields, the minimum pair frequency (i.e. the frequency that both components of the pair occur together), and the pair combinations to hone in on pairs that make sense to you.

The Misc Filters menu provides some advanced options for improving the relevance of correlations. It allows filtering the results with a minimum and maximum Odds Ratio and by default is set to exclude any negative correlations. Additionally, there is a filter "Coverage < 100%" that when selected will exclude pairs where either variable has 100% coverage within the pair. For example, for the pair Theme: Checkouts and Theme: Checkouts // Self Service, the Checkouts // Self Service theme will have a coverage of 100% because it is a subset of the Checkouts theme; as a result, it would be excluded from the displayed correlations.

Drilldown

If you click on any pair combination that contains a theme or structured field (i.e. categorical or numerical), you will be given the option to explore the association further via the Pivot Table widget.

To understand the theme component further you can click into the Drilldown view for that theme. You can also dynamically reduce the noise in your correlations widget as you scroll through the themes. If you come across a field that you do not wish to compare you can select to exclude it from the widget.

Did this answer your question?