Those kind of statistics are useless. They lack any kind of control whatsoever. For example, it's very easy to imagine that a company that knows its game is not telling a compelling story might be tempted to make a demo to try to get people into it, whereas a company with confidence in its brand might not. Who knows, the demo might be doubling sales in those otherwise no-hoper games.

You could just as easily rewrite the conclusion to be "companies who knows their game is crap often choose to make a demo". Without a more sophisticated statistical analysis, it's every bit as valid.