Alcohol Driving Studies, Multicollinearity, And Missing Variables

A comment I posted on John Goodman's Health Policy Blog, "Alcohol: Is the Medium the Message?" by John Goodman:

Actually auto accidents and alcohol drinking is not as clear-cut as common wisdom believes.

Night driving, driving while tired (later at night after being out), younger, inexperienced drivers, and risk takers all have higher accident rates than the general public, even without alcohol consumption.

Studies about alcohol and accidents do not include these other accident causative factors. If they did, the increase in accident rates from effects of alcohol consumption would be greatly diminished and maybe even found to be not statistically significant.

By excluding the additional causative factors, the studies would show a continuing need to lower alcohol consumption to lower auto accident rates. The vested interest groups which fund the studies and which lobby for lower alcohol consumption do not have any interest in seeing studies showing a lower effect of alcohol consumption on driving accidents.

When variables are correlated among themselves, multicollinearity, or when variables are excluded, the strength of the effect on the outcome of any of the included variables, like alcohol consumption, cannot be determined accurately. Alcohol's correlation with the missing variables may make it appear that it is a cause of accidents when in effect it may not be.