> Covariates. All models adjusted for ... age, sex, race/ethnicity, educational attainment, and log-transformed current household income.
Let's look at something else which wasn't controlled for at all: doing physical activities.
People who, for example, run every week or bike every week often do so because they have a group of friends who also does that, and doing such a physical activity also builds friendships better than not doing such activities.
Perhaps exercising is both correlated with health and with building groups of friends.
Or perhaps exercising is correlated with being attractive, and being attractive is correlated with building social connections.
... I guess what I'm trying to say here is that this study shows correlation only, and there are so many confounding factors I consider it pretty tenuous.
That could be a factor, though 1) the study found no correlation with cortisol levels which is affected by exercise, and 2) income and education are highly correlated with physical activity levels already, so wouldn’t put too much hope on that.
Let's look at something else which wasn't controlled for at all: doing physical activities.
People who, for example, run every week or bike every week often do so because they have a group of friends who also does that, and doing such a physical activity also builds friendships better than not doing such activities.
Perhaps exercising is both correlated with health and with building groups of friends.
Or perhaps exercising is correlated with being attractive, and being attractive is correlated with building social connections.
... I guess what I'm trying to say here is that this study shows correlation only, and there are so many confounding factors I consider it pretty tenuous.