As others have commented, it's hard to draw conclusions from such a biased sample set. To truly determine causation we would need to run a double blind experiment.