look @ the photofeeler page. They raise some valid points because OKcupid did not do any tests for statistical significance nor did they provide enough transparency in their data (or better yet, the actual derived data) for this to be inferred much less calculated
Looking at OKcupid's page and graph again, it looks like when it comes to looking AT the camera, BOTH platforms are consistent: It's better to SMILE. Where the platforms differ is when you're looking AWAY from the camera -- Okcupid says it's WAY better to NOT SMILE IF you're looking AWAY, while photofeeler is saying there's no statistically significant difference
Basically, Okcupid is saying the best type of pic looks like this: