Controlled studies, lack thereof

[QuarkChild]

Reading the news is so depressing. "Health news" is inevitably summaries of "studies" by "researchers" who compare two groups of people, sort them by some characteristic, and declare that whatever characteristic they were sorted by causes some effect. Today's example is from Reuter's (I just picked the first article that I saw):

"Exposure to tobacco smoke nearly doubles a child's risk of having cavities," said study author and pediatrician Andrew Aligne....
[Logic OK so far]

"That gives us one more piece of information that passive smoking is bad for children and that all children deserve to grow up in a tobacco-free environment," he said.
Later in the article it said that the data was based on surveys. original article (http://asia.reuters.com/newsArticle.jhtml?type=scienceNews&storyID=2363339)

The scientists may very well be right in their conclusions, and the mechanism they proposed seemed plausible, although I'm not an expert in the field so I'm not qualified to judge. However, I would dispute their method. Am I correct in thinking that this type of research cannot prove causation, only correlation? I can probably think of 5 other explanations for the data, besides direct causation; for example, if smoking correlates with lower quality health care, the dental health of the children might suffer. Etc. Etc.

I'm not suggesting that my stupid explanation is correct--only that this study provides no means of deciding in favor of causation, because it wasn't controlled.

I'm not blaming the researchers for poor study design in this particular instance, since it would probaby be unethical to randomly assign children to be exposed to 2nd-hand smoke. But can't they at least acknowledge the limitations of their methods?
Am I correct in asserting that "Passive smoking is bad for children" is not a valid conclusion that can be drawn from this study?

I would have said, "this survey suggests that exposure to 2nd-hand smoke is correlated with poor dental health in children."

Maybe I just like to get worked up over molehills.

[to.by]

Epidemiological studies of this kind (actually any statistical study) can never do anything else than find (or fail to find) statistcal correlations. They are important, but you have to look elsewhere for a causal mechanism. Furthermore you have to take into account the possibility a third factor causing the other two, other factors influencing the other two in other non-caustive ways or the possibility that this is only a statistical fluke.

[garys_2k]

Of course these types of studies only show correlation, but the media jumps to the "causation" conclusion. If the study was well written it would disclaim a causation link and describe how more refined studies could be designed to further the correlative relationship. Eventually, if other causes could be ruled out, then a causative relationship may be the best explanation.

But, off the top of my head, I can think of a few other things to check before jumping to conclusions, too. Like:

- Relationship between smoking, family income and trips to the dentist

- Relationships between smoking and the quality of the food that people buy (higher sugared food)

- Relationship between smoking and the quality of overall child care (do the parents insist that kids brush their teeth)

[DrMatt]

Originally posted by to.by
Epidemiological studies of this kind (actually any statistical study) can never do anything else than find (or fail to find) statistcal correlations.

Indeed.
Consider the control needed. Exposure to second-hand smoke may well corrollate strongly with growing up in a culture with a lackadaisical attitude towards health in general, and tooth-brushing in particular.

I'm not convinced second-hand smoke is harmless, by any means. In fact, I tend to suspect tobacco smoke is the main cause of "cancer clusters under power lines". But it's just a suspicion.

[QuarkChild]

In the other thread about this topic, it was pointed out that the researchers tried to control for possible links between smoking and family income/healthcare quality:

Quoted by Marvel Frozen in the second-hand smoke thread:
This relationship persisted even when the researchers accounted for the influence of family income, the region where the children lived and how often they visited the dentist.
That's probably about the best one could do in a situation where it isn't possible to do a truly controlled study. The problem with this type of accounting, though, is that there could be other factors that introduce a correlation which the researchers don't know about or don't think of. (For example, quality of diet, which garys_2k brought up.) :

[Agammamon]

Correlation is not causuality.

Originally posted by Agammamon
Correlation is not causuality.


If x and y are correlated, x can also cause y.

[garys_2k]

Originally posted by Whodini



If x and y are correlated, x can also cause y.
Yes, or x may also not cause y. I'd bet a lot of diabetics have insulin in their refrigerators, but I'd also bet that putting insulin in your refrigerator will not cause diabetes.

[DrChinese]

Yes, it's a shame that most science coverage by the popular media is primarily designed to sell advertising rather than inform the public. You can always count on a kooky study to get coverage if it has the right amount of controversy - or head-turning content - attached.

Remember the "blondes going extinct" story a few months ago?

Yes, of course the studies showing X- and Y- correlated issue means that neither may be caused by the other. The story is usually placed into the public's conscienciousness in such a way as to leave the impression that scientists accept the correlation as evidence of causality. A disclaimer in the study itself - which usually sinks to footnote status in the article summarizing the study - states that "more study needs to be done."

Right. Let's get real here. The scientist is taking full advantage of this little "out" as a way to get something published. That is the truth, and we all know it. It is fairly easy to perform studies of spurious correlations. After all, the knowledgeable scientist should know what to look for.

So is it really a shock when a scientist performs a study that says X is correlated to Y? Is that scientist REALLY surprised at the outcome? That scientist had NO idea that a correlation would be found that might be of interest on casual examination?

Biggest culprit: studies of human behavior.