What are good ways of visualizing many effects ordered in groups and subgroups? In epidemiology, we often deal with lots of factors associated with disease. This multitude of factors makes plots of effect sizes (often hazard ratios) of individual factors confusing. Accepting some oversimplification, factors can for the most part be reasonably assigned to some group like nutritional, lifestyle, social, psychological, demographic, and maybe even subgroups like e.g. macro-/micronutrients for nutritional factors.
I am searching for a plot that gives an overview of the effects within these groups while facilitating a visual comparison of size and distribution of effects within and between groups.
I have tried 


*

*small multiples of histograms/smoothed density estimates and parallel boxplots of HRs in each group

*forest-plot-like plots showing effects sizes and their CIs as points and horizontal segments and groups separated by horizontal lines


These ad-hoc approaches are not optimal. The first discards most of the information on the level of the individual factor, the second is too detailed on that level. Neither provides real subgrouping. Neither provides an informative visual comparison of distribution AND size of effects within groups.
Is there something better?
 A: So many comparisons is certainly a difficult task, and any way you can reduce the sets down into more meaningful subsets (especially the 137 variables within) will aid the task. Below I have added two examples of 8 groups and 137 variable forest plots. The images are shrunk to fit within the space here on CV, but if you open the images in a new tab they are larger and allow for better visualization.
You really want two comparisons, within group variability and between group variability. With fewer variables you could get away with assessing both in one plot, but with this many I would suggest two seperate ones. The first plot shows the between group variability paneled by variables, and the second plot shows the within group variability paneled by the groups.
Design choices for the graphs I would recommend are being as minimalistic as possible. Here I make the forest plot bars just a simple line, and the effect estimate a small cross. I add a reference line at zero, as these are really only good for assessing general trends/distribution, if you need to find an exact value of an effect a table is better suited (so I say forego gridlines with so many plots), or displaying a much smaller set of data. Perhaps one wants to try a graph without the bars so one can see the effect locations directly.
Also not shown is any ordering of the factors, but this will be very helpful for the 137 within groups. Either ordering by logical groupings, or by effect sizes will be benificial (this is pseudo random data, can you guess how I made it by looking at the second, within groups plot?) For example, for the second plot I may just pick one of the groups, and order the variables by effect size for all panels according to that group.
For references on displaying so many values in such a small space I would recommend Visualizing Data Patterns with MicroMaps by Carr & Pickle. The maps aren't pertinent, but they show many examples of packing alot of information into a small space in similar type dot plots (it also includes a comprehensive review of visualization research). 


A: One of my recent blog tried to address this kind of questions. 
http://jamesxli.blogspot.ca/2012/08/tracking-attributes-in-mds-map.html
