Bad Stats + Bad Sense = Bad Science
In any given day, I much prefer to read the straightforward result of an experiment (tissue culture, animal studies, clinical trials) than the latest bewildering association discovered through statistical acrobatics with epidemiological and registry data.
University of Graz Study Finds Vegetarians Are Unhealthier, More Mentally Disturbed Than Meat-Lovers
Bad Sense: it seems even common cow sense is not very commonplace. Take this study that got carnivores gloating and got the goat of vegetarians. Instead of postulating that vegetarianism somehow causes ill health and anxiety, one can simply understand how ill health and vexing over it may cause someone to consider adopting healthier eating habits, hence the apparent association.
Bad Stats: choosing to correlate only all cause mortality with cholesterol level (which demonstrated an inverse relationship) while dismissing the direct correlation with cardiovascular mortality is simply playing to the wanton living lobby.
Obviously, the low cholesterol cohort is dying from other causes other than cardiovascular causes (everyone must die from something), ranging from cancer, infections, accidents, etc. The cachexia in cancer is linked (causally) to low cholesterol.
What is repeatedly and extensively proven by far better (experimental) science is that statins reduce cardiovascular events and deaths in the high cardiovascular risk population (hypertensive, diabetic, dyslipidaemic, previous heart attack, post-stenting). Period. No multiple regression analysis. No irrational or mystical mechanism of action.
Telling people to stop their necessary medicines based on such bad stats is highly irresponsible, if not reprehensible.
This is a tough read, but if one seeks to discern the sensationalistic headlines that are periodically churned out from epidemiological studies, one should at least understand the limitations in the complex statistics that seek to adjust for selection bias and confounding variables inherent in these studies.
Bias, Confounding and Effect Modification | STAT 507
