He is absolutely right: confounding is explained in the first chapter of any book on statistics. Alexa's stats do not fall into "randomized controlled" category of studies, therefore they are pretty much guaranteed to be wrong.