Insights in visualizations for summaries by one or more explanatory fields
Insights for summaries are available when the summarization level is average, sum, minimum, or maximum for a continuous response field. Insights are computed and displayed at each category of a single categorical explanatory field, or each combination of categories of a pair of categorical explanatory fields in the visualization.
Overview
Use such visualizations when you are interested in comparing values of a response field across different categories, or across combinations of categories of explanatory fields.
Algorithms
If the summarization level is average, IBM® Cognos Analytics with Watson detects any relationship between the response field and explanatory fields and computes predictive strength of the corresponding model. If differences of average values across explanatory categories are statistically significant, Cognos Analytics identifies the most different explanatory categories or combination of categories under the meaningful differences.
When the response summarization level is sum, Cognos Analytics computes the average sum across explanatory categories or combinations of categories. If the differences of sums across categories are statistically significant, Cognos Analytics identifies the most different explanatory categories or combinations of categories under the meaningful differences.
For all applicable charts, the average insight displays the mean summarized response value across all explanatory categories. When the summarization level for the response is average, the weighted mean is computed using the displayed value and the count for each explanatory category.
Details
- Average by single explanatory field
-
When the summarization level for the response field is average and a single categorical explanatory field is available, Cognos Analytics applies one-way ANOVA analysis. Cognos Analytics uses F statistic to test whether average values across explanatory categories are equal. If any differences are significant, Cognos Analytics computes the adjusted R-squared as predictive strength of the relationship between the response field and the explanatory field. Reliable relationship and its predictive strength are reported to the user if the predictive strength exceeds 10%.
If the difference among averages is significant, Cognos Analytics conducts an influence t test to detect the categories that are the most different from the overall mean. This involves computing standard error for each category average and comparing the average with the overall mean by using the t test statistic. For categories with significant differences, Cognos Analytics also computes the corresponding effect size and reports the categories with the largest effect size under meaningful differences.
- Restrictions
-
The following list describes the conditions that determine whether insights are suggested for this algorithm.
- Average by two explanatory fields
-
For charts where the summarization level for the response field is average and two categorical explanatory fields are available, Cognos Analytics applies two-way ANOVA analysis. Cognos Analytics uses F statistic to test whether average values across explanatory category combinations are equal. If differences are significant, Cognos Analytics computes adjusted R-squared as predictive strength of the relationship between the response field and the two explanatory fields. Cognos Analytics also computes the adjusted R-squared for one-way models that include a single explanatory field each. If the predictive strength of a two-way model is larger than 10% and its relative predictive strength improvement over corresponding one-way models is more than 10%, Cognos Analytics displays the predictive strength of the two-way model and reports reliable relationship between the response field and the two explanatory fields. Otherwise, if the maximum predictive strength of one-way models exceeds 10%, Cognos Analytics reports reliable relationship between the response and the corresponding single explanatory field together with its predictive strength. If maximum predictive strength of one-way models does not exceed 10%, Cognos Analytics reports no relationship between the response and explanatory fields.
When the difference among averages across all category combinations is significant, Cognos Analytics also conducts an influence t test to detect the category combinations that are the most different from the overall mean. This test is similar to the test that is used for a single explanatory field. The main difference is that instead of considering categories of a single explanatory field, Cognos Analytics considers category combinations from the two explanatory fields. Category combinations with the largest effect size are reported under meaningful differences.
- Restrictions
-
The following list describes the conditions that determine whether insights are suggested for this algorithm.
- Sum by one or two explanatory fields
-
For charts where the summarization level for the response field is sum and one or two categorical explanatory fields are available, Cognos Analytics applies the sum comparison test. This test detects if any of the sums are different from the average sum value across all explanatory categories or combinations of categories. If this test is significant, Cognos Analytics proceeds by conducting the sum influence test that compares sum for each category or combinations of categories with the average sum. For every significant test, Cognos Analytics also computes corresponding effect size. Categories or combinations of categories with the largest effect sizes are reported under meaningful differences.
- Restrictions
-
The following list describes the conditions that determine whether insights are suggested for this algorithm.
- Minimum or maximum by one or two explanatory fields
-
For summarization levels of minimum or maximum, only the average insight is available. It is computed as the average value of the response minimum or maximum across all explanatory categories or combinations of categories.
- Restrictions
-
The following list describes the conditions that determine whether insights are suggested for this algorithm.