I can't find a link to the original paper but it would be good to check the details of the analysis. Firstly, they'd have to take care doing comparisons between

I can't find a link to the original paper but it would be good to check the details of the analysis. Firstly, they'd have to take care doing comparisons between the 4 different groups - it's not enough just to do multiple-comparisons as the probability of a false-positive result escalates with each comparison (I would expect them to have done this). Also, the sample sizes seem low and it would be good to see how they estimated the variances (say by using boot-strapping) and shown that this is indeed not just a result by chance.