Posts Tagged ‘statistics’

Lies in the service of policy

Tuesday, March 11th, 2008

Politics has always been infested with lies. As it becomes more important in our daily lives, those lies become more significant. Woodrow Wilson said he would keep us out of war. He lied although there is some possibility that he believed it when he said it. Roosevelt said something similar but there is no chance that he believed what he was saying. A few years ago, the issue of minimum wage was influenced by a published report which purported to prove that raising minimum wages, contrary to economic theory, would not increase unemployment for low income workers. The study was deeply flawed but it has remained a popular basis for those who wish to justify the policy of raising the minimum wage.

Now, the major domestic issue that influences public policy is immigration. Sure enough, a new study has appeared that purports to show illegal immigration raises average wages for the native-born poor. Once again, it has been shown that the study in question is bogus.

I’ve always been a little skeptical of the Ottaviano-Peri evidence. A couple of years ago, Jeff Grogger, Gordon Hanson, and I worked on a paper that examined the link between immigration and African-American economic status. As a by-product of that work, we explicitly attempted to replicate the Ottaviano-Peri finding–but couldn’t. Since then, we’ve been quite interested in trying to see what explains the discrepancy between our evidence and theirs.

Then they found why the discrepancy existed. The other authors had doctored the data.

The Ottaviano and Peri data includes currently enrolled high school juniors and seniors. They classify these high school juniors and seniors as part of the “high school dropout” workforce. Their finding of immigrant-native complementarity disappears if the analysis excludes these high school juniors and seniors.

Things that seem too good to be true usually aren’t.

This is not a new phenomenon. I saw something very similar in surgery 30 years ago. At one time, there was a flurry of interest in what was called “The no-touch technique” in colon cancer surgery. The principal author was George Crile Jr, often known as “Barney” Crile. His father had founded the Cleveland Clinic and was a famous pioneer surgeon. The son had ambitions to emulate his famous father and had become a senior surgeon in the clinic his father had founded. He published the “no-touch technique” study when I was a resident in surgery and we all immediately adopted the method as Crile’s study suggested a significant improvement in survival of the patients. Years after it was shown to be a fraud, it is still being studied. It is difficult to find the original paper anymore but it is still being referred to proudly in Cleveland Clinic literature. In that account, Rupert Turnbull is credited with the development of the technique, which involved isolating and ligating the veins from the colon before the tumor bearing area was touched or dissected. It made sense logically in that tumor cells were thought to flow in the venous blood to the liver where they lodged and became metastases. By ligating the veins first, tumor cells disturbed by manipulating the tumor would not escape and flow to the liver. Every surgeon who did colon cancer surgery adopted it.

A few years later, I attended the GI cancer postgraduate course at the American College of Surgeons annual meeting. One of the items on the program was a study of the effect of injecting 5-FU, a chemotherapy drug, into the colon before removing the tumor. The theory here was that the chemotherapy drug would flow, in the same distribution of portal vein blood as the cancer cells, toward the liver. It was a reasonable premise but the study produced one of the most dramatic scenes I have ever witnessed in a medical meeting.

The senior author was describing the 5FU study and pointed out that the control group for his study was the same as that for the “no touch” study. The veins were not ligated until the colon and tumor had been completely dissected. Any tumor cells that would tend to break off and flow to the liver should make the control group results worse than the no-touch treatment group and similar to the control group of the Crile study. In fact, that did not happen. The control group of the 5FU study did as well at five years as the treated group of the no-touch study and the control group of the no-touch study had a significantly lower survival than any of the three other groups. Why ?

The senior author of the 5FU study answered the question for all of us right then and there. He had contacted the Cleveland Clinic statistician to learn why the results were so different and he finally figured out what had happened.

All medical studies that involved time-survival statistics use what are called “time-life tables.” These are usually generated by actuaries for life insurance companies. Over five years, a certain percentage of people will die of various causes and the percentage who die is based on their age and sex and other factors that these tables consider. Any medical study that considers survival over five years or longer must use these tables to be valid statistically. Some people will die from causes unrelated to the treated condition and these must be allowed for.  You have to correct your results for the normal death rates or you will show more deaths in the treated group (and control group) than can be attributed to the disease you  are studying. The 5FU study author had learned that Crile, who had written the “no-touch” paper, had used time-life tables for the treated group in his study (thus improving the survival) but not for the control group. This is not poor statistical method; it is lying. He twisted the data to make his study look like progress in cancer treatment. In fact, there was no benefit to the early ligation of the veins. Cancer is not affected by those theoretical considerations, probably because host resistance is far more important.

Rupert Turnbull, a justifiably famous colon and rectal surgeon, was in the audience at that conference and the author of the 5FU paper invited him to comment. Turnbull declined, saying that they would have to “ask Dr. Crile about methods.” Crile was not there and nothing further was said but the tension was tremendous. Turnbull was, no doubt, humiliated but everybody knew about Barney Crile and his obsession to surpass his father. There were questions about his earlier work on breast cancer and the validity of his papers on that subject. Ironically, his son, a journalist and author of “Charlie Wilson’s War” would become more famous. Also ironic is the fact that CBS was successfully sued for libel by General Westmoreland because of a George Crile III report on Vietnam. Maybe that’s another family tradition; manipulating data.

Isn’t it interesting that the “no-touch” technique is still being promoted as a science breakthrough 30 years after the study was shown to be a fraud? I suspect that few people who were not at that American College of Surgeons meeting are aware of what happened. I suspect the other fraudulent studies will be influencing public policy years from now, as well.

I have been called a cynic.