Finally found some time to go through all the individual fly data. All groups lost some flies that had to many or too long flight pauses, strong preferences already before training or some other screw-up during the experiment that I either failed to notice or only showed themselves in this quality control phase. And it was the red group, where I had thought everything just worked so fine, that lost most flies. More than a third of all the flies I had tested in this group failed the very basic quality control criteria – double the number of any other group. This is very annoying as it means there now are a lot more flies for me to test. Nothing to do about that, than to carry on. Here is the result after the massacre, where red lost 10 flies, green lost 5 and purple lost 2:

But even with so many flies removed, the red group still has the flies with the strongest optomotor response: