Very quick summary, so apologies if my previous sentence or this one isn't sufficient or well put together: most studies tend to be statistical/high-level correlation searches.

With something like breast- feeding, an effect-size for breast-feeding must be taken implicitly in the context of what happens when you 'don't breast feed', given that basically everyone is still going to feed the baby, and there can't be a third state of neither breast feeding and not breast feeding.

I don't mean just statistical attempts to control for demographics, education, etc, although these also have a complex effect/interaction.

I mean here things like 'if you're not breast feeding, you're doing something else', and the mechanics of that something else are not constant throughout the world.

So for instance, do you have to earn a wage (be away from baby) to buy formula to feed the baby, does the formula at a shop nearby and within its used-by date, what is the quality of it, do you have maternity leave and is there an inherent opportunity cost between breast feeding and work and providing the baby other things, do you have running and sanitary water, are they in day care/familial care, what foods are you giving baby, etc.

All those things effect whether breast feeding will show a statistical positive/negative effect.

Even something like "increased/decreased infections" isn't easily straight-forward: for instance, if the number of infections is related to the number of people baby is exposed to, and baby is in day care, then maybe its day care or N family members looking after baby that's the full story. And then even once you've got THAT, there's a real question as to whether the total number of infections a person gets is statistically static over their lifetime. Early day care could substitute infections earlier into the babies life. This too would appear in a basic study as a positive for breast feeding at home, because they don't measure the increase in infections after the breast feeding period ends once the baby starts to socialise.

Most studies are not vaguely done at this level of investigating or answering these questions/mechanics around why or how the effects are mechanically operating, so many of the findings are likely reporting on these.

Good word choice.

I think the GP is trying to say that the studies look at the benefits of avoiding the alternatives to breastfeeding. For example, breastfeeding is a good way to avoid feeding your infant the tainted water required to make formula.