When wrong or missing data ruins your day!

[leemc-data-ed]

Share a story of missing or wrong data. It could be something minor, or a major problem like the California man who got the license plate NULL for his car.

[maltenform]

I have two, both from my days at Jefferson Neurosurgery

Item 1: I was doing an analysis of stroke outcomes. One of the predictors was smoking status. The regression indicated that in our data, smoking was protective -- that is, associated with better outcomes. I tweaked my code several ways, same result, then bit the bullet and asked one of the MD's on the project. He explained that smoking status was assessed at discharge -- they interviewed the patient, which meant that only the patients who could be interviewed at discharge could possibly be labeled as smokers.

Item 2: some lovely post-doctoral fellows from abroad -- I shan't name the country, but as I'd hate to seem like I was picking on any nationality, but they had clearly been traveling together -- were assembling tables on various aspects of stroke. They had one table for demographics. One table for labs. One table for surgeries. One table for outcomes. What these bright, friendly, charming people did NOT have was any key that could be used to link the tabes...if they did, it would have been a killer study, as they had painstakingly recorded thousands of patients!

In both cases, I was tempted to quote eugenicist and general jerk Sir Ronald Fisher, who observed that calling in a statistician after the study is completed is like asking them to perform a post-mortem -- they can tell you what the study died of!