Making Comparisons

Correlation Requires Variation \& Selecting Dependent Variable

• The Problem of Selecting on Dependent Variable
  • Don't even have correlation
  • No value for evaluating causal claim
  • Even though the "evidence" is very intuitive
  • Feel confused and will have a hard time thinking that claims are not supported by evidence
    • Implicitly making a comparison that is not presented
• Correlations require variation
  • Unless there is variation in both $X$ and $Y$, we don't have a correlation
  • The missing data could make any correlation possible.

• Selecting on the Dependent Variable

  • Mistake "selecting on the dependent variable."

    The dependent variable (DV) is an outcome variable of interest that we want to explain.

    Independent Variables are the variables we try and correlate with the DV, use to predict the DV, or claim have a causal effect on the DV.

    • If collected data only includes observation with particular outcomes (success, old age, violence), we will
      • learn nothing about correlations in the population
      • derive wrong conclusions over and over again

  • Selecting on DV = Looking only at the sick or only at successful companies
• Correlation is necessary but not sufficient condition for causation.
• Step one on the route to causation is compare.