Sunday, October 7, 2012


 Freakonomics - Steven Levitt/Stephen Dubner, 2005

This work illustrates, through case studies, how conclusions that appear to be intuitive/obvious are actually logical fallacies of the type cum hoc ergo propter hoc (correlation does not imply causation). Disproving such arguments is fairly common in some problems in the science/engineering fields, where the conventional methods of doing it are thorough theoretical methods such as regression analysis or via Monte Carlo methods, empirical measurements using:
  •  An identification of all factors that impact the results of an experiment
  •  Repeated experiments under controlled conditions,varying each parameter separately while keep all others constant.
As one may expect, these techniques are either not applicable or are inadequate to apply to problems in sociology/economics, especially the kind of questions that Levitt seeks to answer which involve incomplete/inaccurate quantitative data. This makes his techniques and arguments very interesting, mostly involving deep studies/interviews, arguments of logic  (where data is unavailable) and cross domain collaborations with experts from other fields.

An outline of the work:
  •  Study of incentives and cheating (Chapter 1)
  •  Study of information and effect of information asymmetry (Chapter 2)
  •  Study of correlation vs. causation or how conventional wisdom is often wrong (Chapter 3/4/5/6)
Chapter 1: Incentives and Cheating
  • Incentive: Mechanism to induce one behavior (favored) over another (unfavored) by providing a reward
  • Cheating: Mechanism to defeat an incentive: acquire reward while performing (unfavored) behavior
  • Three types of incentives: Moral, Social, Economic
  • Any clever incentive scheme will result in the creation (or attempts to create an) equally clever cheating scheme
  • Some conclusions from case studies:
    • Approx 90% of humans do not attempt to cheat particular systems despite ability to do so (bagel experiment)
    •  Cross correlation analysis can be used very successfully to detect cheating (Chicago public education system, Sumo wrestling)

Chapter 2: Information
  • Information asymmetry; When two parties to a transaction have vastly different degrees of expertise
  • Internet has reduced information asymmetry
  • Often exploited (real estate agents, car salesmen)
    • Can be extremely subtle: Terms in real estate ads, selling cars
    • Some revealed through correlation analysis
  • Information crime
    • Crimes committed by exploiting information asymmetry (Enron)
    • Difficult to discover, something drastic must happen

Chapter 3/4: Correlation vs. causation (Conventional wisdom)
  • Convention wisdom (CW) can be incorrect in several cases.
  • Dramatic effects can be caused by subtle, overlooked, non obvious events
  • Events can be explained by careful study of the correct causes
  • Case studies demonstrating this include:
  • CW: Drug dealers make  a lot of money
    • Reality: The structure of a drug dealing organization is similar to a corporation
    • Workers at the bottom have low wages, bad working conditions
    • Upper management keeps disproportionate share of profits
  • CW: Crime decreased in the 90s for several reasons,
    • Correlations and logic used to check a number of likely causes
    • Closest correlation is with the Roe vs. Wade case and the legalization of abortion

Chapter 5/6: Case studies on parenting
  • Correlation between parenting approaches and the future success of children
    • A number of parenting factors are examined including those which show correlation (positive/negative) with child's success and those which have no correlation
    • All factors  appear to be correlated in some way to status of parents (education level of parents/affluence etc.)
    • What a parent does is irrelevant compared to who the parent is (education level/affluence etc.)
  • Correlation between children names and future success of children
    • Choice of name and success are uncorrelated, even though name choice has very strong correlation with race
    • There is a strong correlation between name choice and parents' characteristics (not childs' future)

No comments:

Post a Comment