Change the phrasing here to be more charitable and I would say instead:

"There is an obvious common factor that explains some of the variance: dementia would stop people from dancing; the beginning stages would likely do so as well. Thus we should expect to see a correlation between dancing and good health even in the absence of dancing being preventative. Because of this expectation, seeing a strong link may be enough to locate the hypothesis, but it is likely to be some combination of statistical outlier and reversed causality. Thus if an effect exists, it is likely to be much smaller than the first measurement."