More like unintuitive rather than counterintuitive, because the prediction doesn't fail because wider lanes are more dangerous per se, but because of the behavioral response of the drivers. Which can actually be considered "intuitive" to someone who is used to trying to model such things.