They're not correlating the use of a flash with photographs being "better" in some general way; they're correlating the use of a flash with specific ratings of

They're not correlating the use of a flash with photographs being "better" in some general way; they're correlating the use of a flash with specific ratings of the attractiveness of the people in the resulting photos. If the data is accurate, then the correlation is real. It might be that most people don't know how to use a flash; it might also be that natural light makes people appear more attractive, or that the position of the light source (i.e. not in the same place as the camera) does the same. Building a reliable causality model out of correlations can be extremely difficult, but that doesn't mean that anything short of a complete causality model is worthless. In this example, dispensing with the flash would still likely improve the ratings of most users' photos (other things equal, of course), even if they might squeeze a bit more out of the result by mastering the nuances of flash photography.