You can certainly be a non-idiotic PHP programmer, and there may certainly be just as many good PHP programmers as there are good programmers for any other language. But you, and they, are outnumbered by the bad "this is my first language and first project" programmers, who tend to adopt PHP for that purpose. and because of that, if you know nothing else about a programmer, knowing that they primarily program in PHP should tend to be Bayesian evidence against their programming skill/experience, just like knowing someone lives in India should tend to be Bayesian evidence against the likelihood of them being wealthy, or knowing someone likes {pop, classic rock, modern country} music should be Bayesian evidence against the diversity of their taste in music. It's not that there aren't tons of counterexamples; it's that there's this one huge overlapping subgroup that happens to have that property, and dominates any statistical effects. (Respectively, "people learning their first language"; "people born in poverty with no way out"; and "people who only listen to music provided by their single local radio station.")

Of course, if some other language were easier to learn than PHP, the dominating subgroup of "people who know 0.5 programming languages" would become attached to that language instead of PHP. So the fact that this group is attached to PHP is actually a point in favour of PHP as a platform. It's still a point against its statistical median user.