Forum Discussion
Using a clustered chart to compare distributions
I have an issue where I am trying to use a clustered bar chart to provide A/B distribution comparisions. The problem I am encoutering is when calculating the percentage of grand total, it is using the global grand total as a denominator, rather than to total for each individual dataset. This is causing the larger dataset to swamp-out the smaller one, and make the comparison not useful. Is there a setting or option to segregate the category totals for the clustered data sets?
To illustrate, the graph below shows the primary distribution:
The next graph shows the secondary latency distribution:
The clustered chart would ideally be a composite of these two distribution graphs. However, when combined the %GT ioCount appears to cover the ioCount for both latency categories, giving a graph as follows:
This is causing the green category to be dominant, and the black category to be statistically insignificant. Is there an option that would normalize these data sets better?
2 Replies
gaglenn , In such cases you should prefer combo visuals and use line for the second measure. Like Line clustered column chart. But that will show vertical bar
Or try Log axis
Thank you for the suggestion. In this case, I just opted to go for two side-by-size bar graphs, as it seemed the clustered graph was not going to work the way I wanted it to.
