In this article
Key Drivers Widget in CDL allows users to adjust quadrant axes, set thresholds for scores and measures, and perform advanced configurations.
1: Initial Setup
Begin by selecting your dependent and independent variables. Afterward, the widgets will show your data and the properties will become visible in the editor.
Note
When utilizing the regression model for the Key Drivers widget, it's advisable to limit the number of drivers to 10 or fewer to prevent intercorrelation. The designer will receive a warning if they select more than 10 drivers.
2: Quadrant Chart Axis Definitions
2.1: X Axis: Satisfaction
The key driver widget converts all scores to percentages to facilitate comparison between scores with different scales.
For instance, an average score of 4/5 will be expressed as 80%, as will an average score of 8/10
2.2: Y Axis: Importance
Depending on algorithm, will either show the correlation value, or the standardized beta coefficient
3: Dynamic Control
The Dynamic Limit toggle provides the ability to adjust the quadrant axes dynamically. This allows the chart to zoom in on the relevant data range, spreading the data points over the entire chart area, avoiding clustering around a small area of the overall scale.
This property is optional and applies only to the XY chart. If activated, the quadrant table view and any quadrant labeling will be disabled to prevent providing misleading information to the user.
For example, if all data points have a minimal spread around a low importance value, automatically centering the importance axis and using explicit labels on the quadrants can give a false impression that a value, even with a low importance score, is in the top half and therefore important.
Example of a key drivers widget and the settings
4: Satisfaction Limit %
The Key Driver widgets convert all scores to percentages for easy comparison of different scales. The Satisfaction Limit % setting is the intercept point for the vertical line on the quadrant chart, determining which drivers are displayed in the left or right quadrants of the table view. The default limit value is set to 75%, but users can change it to reflect their desired definition of a "good" score.
5: Importance Limit
The Importance Limit sets the threshold that separates the top and bottom halves in the quadrant table by intercepting the horizontal line in the quadrant chart. It has a value ranging from 0 to 1, with 0.5 being the default limit value. Whether using the correlation algorithm or the regression algorithm, this value reflects the correlation value or the standardized beta coefficient, respectively.
6: Quadrants Properties
The Chart and Quadrant Properties section governs the aspects of the chart and quadrant, such as labeling and color settings. You have the option to create custom warning text that will only appear if one or more of your drivers have insignificant values.
Example two of a key drivers widget and the settings
7: Model Validity Limit (Regression Model Only)
7.1: Minimum Sample Size
The regression model in Key Drivers Widget calculates based on a minimum sample size that you can set. The default recommendation is to have 10 cases per predictor variable, but you can adjust it manually if needed.
Minimum sample size field
Note
The model uses pairwise deletion algorithm, so all independent variables must have a value for a record to be considered in the analysis. This could result in differences in average scores for dependent variables compared to when calculating average scores in other widgets, such as KPI widget, if there is missing data in the variables included in the analysis.
8: R-Squared Limit
As the dashboard designer, you can set a limit for the R-squared value in the regression key driver algorithm, which serves as a measure of the accuracy of the regression model.
R-squared represents the percentage of the variance that the model explains. As a guide, the following R-squared values indicate the level of accuracy in the model:
| R-Squared Value | Model Accuracy |
|---|---|
>80% |
Excellent |
>70% - 79% |
Very Good |
>60% - 69% |
Acceptable |
>50%-59% |
Weak |
<50% |
Very Weak |
When using the regression key driver algorithm, the dashboard designer can set a limit for the model R-squared value as a measure of accuracy. A low R-squared value, less than 0.5, indicates that the results of the key driver analysis should be considered directional rather than prescriptive based on the fit of the model.
The accuracy of the model may change with any dynamic filtering applied by report viewers. If the model falls below the R-squared threshold set by the designer or the default value of 0.5, a warning message will be displayed on the Key Drivers widget.
9: Individual driver contribution
A driver is marked as not contributing strongly to the model if the significance value for the driver exceeds 0.1. While this threshold value is not configurable, the message delivered to the user about this can be customized.
10: Viewing Model Details
In the "Model Validity Limits" section, toggling "View Model Details" to on adds a pancake icon to your widget.
When clicked, you are presented with a pop-up showing the details for that model. This is useful if you want to quickly view the model details from within the widget.
Toggling "View Model Details" to off will remove the icon from your widget.
Enabling the View Model Details toggle
The model details pop-up
11: Example Setup of a Key Drivers Widget
Example setup of a key drivers widget
Example of a key drivers widget with above setup
12: Advanced Configuration
Advanced configurations can be performed in CDL in addition to the properties available in the visual editor.
13: Additional Considerations/Common Questions
Key Drivers does not support using weighted data.
When using regression, differences can appear between metric scores used on the Key driver and metric scores elsewhere in the dashboard due to the last 10,000 records limitation, and only using the records where all values exist (the “listwise deletion” rule).
Key drivers could in theory be used on custom table data, but the challenge is that numeric variables in custom tables do not currently support bounds (upper and lower limit values) so a change would be required to support this case.
For details of the properties and settings, refer to widget keyDrivers in the CDL Reference Guide.