Causality vs. Statistics
I absolutely, positively guarantee you that if you beat a drum during a solar eclipse, it will go away. The correlation is 100%, but no one, except primitive tribes, believes there is any causal relationship.
You can’t prove causality with statistics. Consequently, in medicine and other scientific fields where controlled experiments are possible, double blind tests are the rule. But even then, drug administrators can inadvertently tip off the patients if they know the real drug from the placebo. And the recipient of the dummy pill can feel better anyway.
Furthermore, dedicated researchers can still cheat. In a far from unique case, a friend who later died of melanoma of the eye was kicked out of an experimental drug test because he wasn’t responding well and the researchers wanted the medicine to be proved effective. Some years ago, medical researchers found they could become instant media celebrities by showing the relationships between various food intakes and cancer. So they pumped rats full of almost anything, even water, to get the desired results. Maybe the rats decided that death was preferable to another syringe full of coffee.
Outside strict controlled tests, it’s almost impossible to use statistics to determine causality. You can run rats through different mazes to see if they behave differently, but you can’t rerun history with different assumptions. As a history professor of min said, “There are no ‘ifs’ in history.”
Sure, there are instances when statistics are strong indicators of causality. When the economy is weak, voters vent their anger by throwing incumbents out. That was true in 1976 when Carter replaced Ford, in 1980 when Reagan trounced Carter, and in 1992 when Clinton ousted Bush. But in 1984, the 7.2% unemployment rate didn’t stop Reagan’s re-election. If an investor has high absolute gains in both bull markets and bear, he probably has superior ability and not just one-shot and random good luck.
Statistics supports causality when leads or lags in the expected direction are observed. High unemployment is followed by declines in labor force participation because people drop out of the labor force after they can’t find work. But unexpected leads or lags make causality suspect, regardless of strong correlations. You’d think that rising consumer sentiment would lead to higher spending, but statistically, sentiment follows spending.
Led by the media and other pundits, many leap from high statistical correlations to causality. Americans have always seen education as the route to financial success, and the statistics bear this out. But does four years of college make people smarter and better equipped to make money? Or are we in a society where everyone with any brains and ability goes to college even though they might well have succeeded without baccalaureate or higher degrees? Elite colleges take credit for their graduates’ accomplishments, but are they simply able to attract the cream of the crop that would have done well regardless? The vast majority of CEOs graduated from Podunk U.
The belief that a college degree guarantees success has resulted in a tremendous dumbing down of American education. As long as someone pays the tuition, often government, colleges are created for any intellectual level. A Pew Charitable Trust study several years ago found that half of four-year college grads couldn’t understand credit card offers, tables relating exercise to blood pressure, or newspaper editorial arguments. Many businesses no longer assume that a college degree guarantees much of anything in brains and skill, and conduct their own tests.
The assumption that an MGA guarantees even more financial success has proliferated those programs. But even in the top schools, do MBAs learn much or simply meet classmates with whom they’ll do business later?
High correlation may be the result of seemingly unrelated forces, and if those forces change, the correlations collapse. Up until 2006, many thought housing demand would never drop because of rising populations, so huge mortgages were safe. Then they learned rising prices drove demand, and when prices collapsed, so did demand.
Furthermore, causality can run opposite from the expected direction, or maybe both ways. In countries where government’s share of the economy rises, stock markets do poorly. But is that because rising government activity impedes private sector activity and productivity? Or are governments responding to weak economies with inherently underperforming stocks? Or is it some of both?
The next time some wizard tries to prove causality with statistics, think twice.
Gary Shilling is President of A. Gary Shilling & Co., Inc. and publisher of INSIGHT.
www.agaryshilling.com
Copyright 2011, author retains ownership. All Rights Reserved.
