An Intro to Causal Reasoning

This spring I was invited to give a lecture on causal reasoning at the Catholic University of America to undergraduates in its honors program. What follows is the outline of that lecture.

General Notes

  • we’ll be talking about how to reason about efficient causation
    • i.e. one thing making something to be the case or binging something about
    • billiard ball-causation
    • causal claims don’t all use the word “cause”
  • sufficient vs. necessary conditions
    • “c is necessary for e”: if e then c
      • being a US citizen is necessary for becoming President
    • “c is sufficient for e”: if c then e
      • being elected by the general population is sufficient for becoming President
    • causes are typically parts of sufficient conditions
      • severe lack of vitamin C is both necessary and sufficient for scurvy
      • striking a match is neither necessary nor sufficient for starting a fire
      • going through a car wash is sufficient but not necessary for having a clean car
    • sufficient conditions are always sufficient: if X really was all it took to bring Y about today, then X would bring Y about tomorrow too
  • there are two kinds of causal statements we’ll focus on
  • general/repeatable == there are many instances of them
    • Smoking causes lung cancer
    • HIV causes AIDS
    • Cold weather causes low tire pressure
    • Lack of vitamin C causes scurvy
    • Caffeine prevents sleep
    • Thanksgiving causes traffic jams
    • Turkey dinner causes drowsiness
  • particular/unrepeatable == there is only one instance of them
    • Terrorists caused 9-11
    • John Wilkes Booth killed Abraham Lincoln
    • German aggression started WWII
    • A meteor caused the extinction of the dinosaurs
    • Human CO2 emissions caused global warming
    • The pyramids were created by aliens

Reasoning about Particular Causal Claims

  • Exercise:
    • “JFK was killed by your teacher” – prove me wrong
  • arguments for particular causal claims usually involve two steps:
    1. showing the deductive consequences of a hypothesis (hypothetico deductivism)
    2. evaluating whether that hypothesis constitutes the best explanation (inference to the best explanation)
  • the argument that JFK was not killed by your teacher essentially has this form:
    • true causal statements are good explanations
    • “JFK was killed by your teacher” is not a good explanation
    • hence, the statement is not true
  • best explanation reveals normative principles, i.e. “scientific virtues”
    • Conservativism: the less rejection of prior beliefs, the better
    • Modesty: the more normal, unsurprising, ordinary the better
    • Simplicity: the simpler the better
    • Generality: the wider the application the better
    • Refutability: the more easily refuted the better
    • these all apply ceteris paribus
  • this form of evidence is relatively weak and inconclusive
    • underdetermination of theory by data: there are always theories that we haven’t thought of that fit the data as well as those we have
    • confirmation holism: any given prediction is the result of many beliefs, so “anything can be held true, come what may”
    • sometimes it’s all we have

Reasoning about General Causal Claims

  • Exercise:
    • you begin college in the fall
    • you’ve were a constant weight for most of high school
    • when you return from college, you’ve gained 25lbs
    • how would you figure out what caused you to gain the weight?
      • did you start eating more?
      • did you start eating differently: more salt, more bread, more dessert?
      • did you start eating late at night?
      • did you start snacking?
      • did you stop exercising?
      • did you start taking some new medication?
      • did you change how you were exercising?
      • did you start drinking?
      • did you start smoking?
      • did your stress levels increase?
    • Notice: you’re trying to find things that changed
  • Another observation
    • what is the first thing a car mechanic wants to know: how to reproduce?
    • why is this important to the mechanic?
  • There are two main steps in causal reasoning about general claims
    1. create a reproduction case
    2. find differences that make a difference to the reproduction case
      • sometimes you can only do (2)
  • controls in science
    • all experiments are comparisons to “controls”
    • example: genetic screen to determine function of SFN
  • why controls/differences matter: correlation vs. causation
    • buying diapers causes having a baby
    • ice cream consumption causes shark attacks
    • sleeping with light on causes short-sightedness
    • Washington redskins home games cause presidential election results
  • controls attempt to get around the correlation problem
    • sufficient conditions are always sufficient
  • this form of evidence is relatively strong