Thursday, March 31, 2011

Correlation, causation and data

Here's a cool short post from the guys at Freakonomics about using data to advocate a viewpoint. You may recognize the name of the purported offender.

Many of you have had econometrics or are now enrolled in 422 (others will have me for 422 next spring), so hopefully if you can't relate to this, you soon will. People want answers. Analysts included. During tumultuous economic times this is especially so. Our job as economists is to seek those answers, find the truth, whatever it may be, and report it as objectively as possible. But many times the truth is "we don't know". People (especially those paying for analysis) don't seem to like that answer, which can create unfortunate pressures that can lead to tenuous conclusions.

Hopefully this class has increased your awareness that economics is messy, and "one size fits all" solutions are very rare.

For your work, be open about what you know, what you don't know, and what your assumptions are. When reading the work of others, question assumptions, check the data and always know your sources.

Thanks to RH for the link.


  1. John Taylor was a member of the President's Council of Economic Advisers during the George H. W. Bush administration and Senior Economist at the Council of Economic Advisers during the Ford Administration. I am sure in these positions he dealt with a lot of pressure and situations where "we don't know" was not a popular answer. It is interesting and a little scary to see someone like John Taylor, who has held such influential positions, present data to back up his claim that is not as strong as he presents. It is a good example of how careful you must be to check data when economics are used to strengthen an argument.

  2. No doubt that Taylor has been in situations where he had to take a position. As most of my work is in the area of applied micro, I'm certainly not in a position to question his modeling or conclusions. But, this seems to be a case where the best answer might be a combination of "it depends" and "we don't know".

    Taylor has posted responses to the recent critiques at his blog:

  3. this is from Nick:

    I am convinced that the absence of a strong correlation throughout the entire sample is based on the changing of government policies and varying levels of regulation since 1948. Because of opposing views among political parties along with the various obstacles and hardships our country has faced, national investment levels do not seem to hold a strong relationship with the unemployment rate.

    Perhaps this is more visible when examining the 1990-present sample that Taylor uses to back his claims. Although Taylor's sample using the most recent 19 years is confirmed to possess correlation by Mankiw's assessment, his claim loses value when analyzing the immediate 19 years prior to 1990. The financial ideals among the Nixon, Ford, Carter and Reagan were all very different and each term yielded different unemployment levels. Although Presidents George H.W. Bush, Bill Clinton and George W. Bush continued to "flip-flop" policies, the data shows a strong correlation during their collective tenure. I think it would be interesting to examine the data during the "Reaganomics" era to compare alongside other presidential terms.

    I like Taylor's claim because I feel it embraces the technology wave that has swallowed the globe over the last twenty years. The world is "flatter" than ever and the global economy continues to grow, functioning at its fastest rate in history. Not only does our economy now expand by larger amounts more quickly as innovations assist us, the August-08 recession proved that it now takes longer to catch par levels. As more domestic jobs are being replaced by workers from foreign markets and technology continues to hasten global news traffic, it will be interesting to see if these trends of high correlation continue to exist.

    [Research from Author: William A. Niskanen]

    --- Nick T.

  4. It seems the higher up the totem people get the more incentive they have to sway results in their party's favor. It is an unfortunate truth but true none-the-less. This is a prime example of how you should not read into everything you read or hear even if it is coming from a seemingly reliable source.
    However, as times change so may the relevance of certain information and studies. The world is drastically different than it was just 1 decade ago so it may be that this study has come to be more significant in our time.
    For this reason double and triple checking statistics is necessary to be sure information is as close to true as possible.

  5. From Kevin:

    Without the background in government policy or econometrics I'm going to take a stab at it...

    What if Taylor and the guys from Freakonomics are right? Taylor may have picked the last 19 years because it is the strongest data that supports his claim. The guys from Freakonomics may have overstated the "why not use all the data." Under certain laws and regulations, maybe Taylor's assumptions hold more weight. Maybe he is assuming that a certain kind of governmental policy will be uniformly applied. maybe the best data to use was the last 25 years, which could have possibly still supported Taylor's claim to a statistically significant level, but not quite as strongly. I am still going to go with the age old economics answer of "it probably depends."