Statistical correlations measure meaningful relationships to identify opportunities through advanced data mining.
The Correlation Matrix uses Pearsons correlation coefficient to better identify the next steps in a marketing campaign, to improve site design, or to continue in-depth customer analysis for additional correlation dependencies.
The Correlation Matrix compares metrics over a countable or non-countable dimension. The matrix can then be modified to highlight correlations within the visualization through color picking or to render it as a text map, heat map, or both.
Right-click Visualization > Predictive Analytics > Correlation Matrix. The dimension table will open.
Select a dimension, such as Time > Day of the Week from this menu. The correlation table will open with the dimension identified in the corner of the matrix and its associated metric placed in the row and column. For the Day of the Week dimension, Visits is the associated metric.
The correlation is 1.000 because you are comparing a metric against itself (which reflects a perfect, but unusable, correlation.)
Right-click and select Change Metric to change a metric in either the row or column. This sets up a correlation between two metrics of value.
For this example, change the Visits metric in the column to Internal Searches. Right-click and select Metric > Custom Events > Custom Event 1-10 > Internal Searches.
Right-click in a metric column or row. For example, from the Metric menu, add
Metric > Custom Events > Custom Event 1-10 > Sign in Error.
The new metric will appear in a column with a correlation number. You can add other metrics, such as Email Signups, to build out the table.
Or add metrics to rows to compare against metrics in columns.
Right-click in the workspace and select Table. From the open dimension table, press Ctrl + Alt and drag the element over a metric in a column or row. The element will appear next to the metric in brackets.
For example, for the Visits metric, you can constrain it by selecting the Country as New Zealand.
Notice that when you select a dimension element, the correlation changes in all metrics based on the selected dimension element. Only the Visit metric will be constrained for "New Zealand" once the dimension window is closed.
The following are general goals for building a Correlation Matrix.
Identify the relationship between two metrics against a specified dimension. In the example, the matrix was built around the core dimension, Day of the Week, with the metrics Visit, Email Signups, and SignIn Errors compared against Internal Searches, Login, and Survey Displayed metric events.
Develop hypotheses to focus analysis. After running a correlation analysis, your next step is to look for dependencies and correlation of the metrics. For example, understanding that internal searches has an effect on email sign-ups provides a path to predict that relationship and to modify marketing campaigns or web site navigation design.
Identify metrics to include more advanced data mining algorithms. In most cases, the key metrics will be identified because they will be seen affecting multiple correlations. You can now take those key metrics and apply them to additional data mining analysis for deeper insight.
Filtering and selecting on dimension elements within a table compares like values. For example, using Day of the Week dimension and then clicking into an element of that core dimension, such as clicking on a specific day within the Day of Week dimension table, renders a one to one match at 100% that provides no usable correlation. Because the root dimension was Day of the Week, any selection within the Day of the Week dimension table will alter the matrix to be a one-to-one correlation.
However, the 1 to 1 correlation (when a single selection is made of all elements) is only on that specific day. If you make multiple selections then it does not necessarily remain a 1 to 1 correlation, and will not always yield a 100 percent match regardless of selecting 1 or 1+ days of the week.
Statistical correlations are not equal to the Correlated Data Model, the historical reference of Adobe Analytics products. The statistical correlation in data workbench is based on the Pearson Correlation model.