Forest Plot
The forest plot visualization allows the analyst to compare the subgroup-specific treatment effects (SSTEs) and identify any general patterns that would suggest treatment effect heterogeneity. An example of the graph that is generated is below, for simple simulated data structured with effect modification and no confounding. As described within the user interface, these data have eight subgroups, with treatment effects of (1, 2, 5, 5, 6, 6, 9, 10). Our partition procedure partitioned the data into ten subgroups.
On the x-axis is an integer denoting subgroup membership - a proxy for the magnitude of the SSTE, since hetviz requires that subgroup membership be indicated by consecutive integers where "1" corresponds to the subgroup with the smallest SSTE [#27 in future update]. The y-axis is the scale for the SSTE.
In the plot above, there are 10 red dots, denoting the median treatment effect for each of the ten estimated subgroups. For each SSTE there is a thick error bar ranging from the 25th to the 75th quantile of estimated individual treatment effects (ITEs) for that subgroup (much like the central box of a box plot) and a thin error bar with a length < 1.5 times the interquartile range (much like the whiskers of a box plot). This presentation was chosen to be more accommodating to large numbers of subgroups than the traditional box plot presentation.
If the dataset is small (i,e., <= 5,000 observations and <= 30 subgroups) then the user is able to select an alternative forest plot presentation, displayed below, that includes a traditional box plot presentation in addition to the specific estimated ITE values.
The user is able to further customize their plot, as indicated directly on the images.