Unfortunately, Nature reporter Richard Van Noorden (the collector of some of the data used in the graph) notes that Professor Hicks did not use the most relevant data for his retractions (i.e. US retractions, funded by the NIH) for his graph. When you do, the correlation is not nearly so clear.
I agree with Richard that the hypothesis is sound (i.e. as it gets more difficult to get funded, the incentive to cheat goes up), but it remains unproven.