I have a data set with very imbalanced groups (about 3% have an effect, while 97% don't). I experimented with logistic regression and was convinced that I can get to a very good accuracy of prediction without matching.
In which cases is matching necessary, then?