That's in the same ballpark as what I'd expect from placebo, but nonetheless a little higher

Or stuff like publication bias (one reason I was interested in whether a funnel plot could be formed from the data).