I have a statistic that assign values to categories of products. This statistic shows strong bimodality (see graph). For analysis, I am trying to assign a value of that statistic to each product (edit: to perform a regression analysis in which products are observations). This is straightforward when product are in only one category. But it becomes difficult when products are assigned more than one category. Since the statistic is bimodal, taking the average of the values for all categories of a product is meaningless. I am curious if there is a way to get this sort of summary statistics?
My question has two related parts:
a) A quick search gave me the idea that there are a few ways of assessing multimodality (Ashman's D, Bimodality index, bimodality coefficient), but no straightforward way to summarize a number of values drawn from a bimodal distribution. But I am curious if I missed something? For the issue at hand, I think I will adopt the approach described in b, but for the future, I would be happy to know what is possible to do in such a case to summarize that type of data?
b) The approach I am considering adopting at the moment is to turn my statistic into three categorical ones: one for the values close to zero, one for the values around 10, and finally one for the values around 5. Then for each product, I would count the number of times the categories it belongs to are listed in each range. This makes sense to me theoretically, but I am wondering if there is some statistical pitfall I am missing? (This approach seems (very) loosely linked to the one adopted here, that looks into splitting the distribution in two populations).