Correlation is not causation...
People who general ignore healthy eating habits have many other unhealthy habits. General enough at least to skew the study by 2% between groups.
Like some trailer trash smoking and drinking pop and beer constantly, with bad genetics already and obesity issues will die before a rich kid with a long line of rich kid genetics even if they drinks 2 sodas a day. On average though, the rich kid will probably avoid soda because they were heavily institutionalized from a young age to live and act a certain way. Generally speaking, enough to skew a study by 2% between groups.
Have you ever seen a study like this that says "We took 100 people age 60 who all were semi-pro or pro athletes of non-contact sports for at least 10 years. We gave them all a physical to pass them into the test. We found 50 who love soda and 50 who really don't care for soda and told them to keep it up for 20 years and we will see who dies first"
No, too hard to find a group that specific, so they move their line back until they can find enough people to fit a category. By the time the line stops moving, the group is already out of control in the sense of factors in why they are already different. Study has no control at that point.