Why two good studies can give opposite answers

It's not sloppiness. It's what happens when you stack people with opposite reactions on top of each other and pull out an average.

200 patients with knee osteoarthritis were followed, and the researchers found a clear association: when air pressure rose, pain rose (coefficient 1.14, p = 0.02). Another study, with a proper case-crossover design, looked at essentially the same question and found nothing at all — not for pressure, temperature, humidity or rainfall.

Both studies were properly conducted. Neither of them was wrong because of sloppiness.

Four studies, four different methods

A systematic review from 2025 went through 979 studies of air pressure and migraine and found 14 that met the criteria — spanning from 1973 to 2023. The authors themselves point to why they're hard to combine: inconsistent measurement methods for air pressure, reliance on self-reporting, small samples skewed towards women, a lack of standardised tools for measuring severity, and limited geographic spread.

Only 6 of the 14 studies looked at severity, and the results were inconsistent — most used non-standardised scales, and only one, which used a standardised VAS scale, found a significant association (p = 0.027). Only 3 studies looked at duration. All three found no significant association — which is itself a result, not just an absence of one.

Another review, from 2019, concludes briefly and precisely: many studies have produced inconsistent results regarding the very direction of the association between pressure changes and headache. Not just whether there's an association — but whether it goes one way or the other.

The main reason: opposite reactions get stacked together

The systematic review itself points to the explanation that carries the most weight: a lack of allowance for individual variation in weather sensitivity. Another review puts it directly — some patients were sensitive to low air pressure, while others were sensitive to high air pressure.

A thought experiment, made up for the occasion: imagine ten patients. Five of them feel worse when pressure falls. Five feel worse when pressure rises. Measure them separately, and you find two clear, opposite associations. Combine them in one statistical model and calculate a single average, and both vanish — the result is a weak or non-existent association, even though ten out of ten actually reacted to the pressure.

That's roughly the trap a Dutch study fell into, when 222 patients with hip osteoarthritis were followed for two years with measurements every three months. Humidity was associated with the patients' pain scores, air pressure with their function scores — but the two variables together explained under 1% of the variation, both within and between patients. The authors' own conclusion was that the results support patients' experience that weather plays a role, but that the contribution is too small to be clinically relevant at group level.

What you can conclude when the answer is "it depends"

You can't use two contradictory studies to settle whether weather affects pain in general. But you can use them for something else: to understand that a group average is the wrong place to look for your own association, if you're one of the people who react the opposite way from the majority.

Contradictory studies aren't proof that weather doesn't matter. They're proof that averages are a poor place to look for your own association.