I don't know of published results of studies comparing red and blue filters. Red light has been used for the visible part of the NDVI equation for as long as NDVI has been around. There are good reasons for this which Ned Horning explains here: https://publiclab.org/notes/nedhorning/11-01-2013/why-a-red-filter-should-work-well-for-ndvi. Using red light to compute NDVI requires a red or yellow filter on a single camera DIY NDVI system.
Claytonb did some careful tests and confirmed that a red filter gave more meaningful results: https://publiclab.org/notes/Claytonb/02-08-2016/plant-health-ndvi-red-vs-blue-filter.
I have done some casual tests and concluded that a red filter is generally better: https://publiclab.org/tag/red-filter/author/cfastie.
However, there are some other important variables that could overshadow the advantage of using a red filter. Using a modified consumer camera to take jpeg photos without a calibration routine in varying light conditions can produce inconsistent results with many types of artifacts. So unless you develop a well thought out protocol, it might not matter which filter you use.
Also, all red filters are not the same. The Rosco filters in the Public Lab Infragram products have been subjected to less systematic testing than other types of filters (MidOpt, Wratten, etc). The research notes I linked above do not use Rosco filters.
The only way to capture separate data about NIR and visible light without cross contamination is to use two cameras. Cameras are now small enough and drones are big enough that this is no longer much of an obstacle. Kites and balloons have long been capable of lifting two cameras.