Prediction markets: For realreal or for playplay?

When running a prediction market is it more accurate to use play money (karma) or real money? This paper here and the easier to read presentation description here provide evidence it does not effect the markets accuracy. Both result in accuracies much greater then the average opinion of the people in the market.

This means that people with more confidence betting larger amounts tend to be more accurate. People without strong opinions being less involved in predictions seems to increase accuracy.

It is interesting to wonder why money does not seem to benefit accuracy. Could it be that using money attracts some of the wrong kinds of bettors? Would some combination of real and play money provide more accurate predictions? In this case there would have to be different types of bets in the real and play money that both happen to be of similar accuracy.

If it is the case that people bet the same with real and play money could it be that money is just a karma substitute?

One important factor with the use of real money is that it rewards those who are right in a way that they can eat. Karma does not butter any parsnips. For using prediction markets in science I think you should reward those who make accurate predictions in a way that aids their making predictions in future.