When you buy a Kia and it breaks it's "unreliable".
When you buy a Ferrari and it breaks it's "temperamental" or "quirky".
A Toyota may cost a pretty penny and be more reliable than the Kia but the Ferrari looks much better in your garage and is much more fun to take for a spin.
It is tough, if not impossible to know for sure in cases like these. The most effective way is the use of controlled studies. The population is split such that the only difference between the groups is one variable. The very simplified version is that sometimes you find that there is only evidence for a correlation (between say relationships and good health), and other times you can find evidence for causation.
I can't find a link but there was an article here recently where researchers had concluded strong relationships contributed to better self control and thus had a knock on effect on health.
this isn't really a useful point. social sciences necessarily rely on this kind of research because you can't do a controlled trial on people's long-term behaviors really.
How do you actually prove that the relationship between the two attributes is causal versus correlated?
For example, one could conclude, instead, that being in good physical health is the cause of successful relationships.