Reasoning About and Graphing Causes Review Distinguish necessary and sufficient causes Most causes are neither necessary nor sufficient Rather contributory or partial Increase or decrease the likelihood of an effect Attending class increases the likelihood of doing well on the exam Distinguish proximate and ultimate causes Review 2 Mill s methods designed to identify the likely cause from amongst possible causes Method of agreement Start with cases that agree in the effect and find what possible cause they have in common Method of difference Start with cases that differ in the effect and find if there is one possible cause on which they differ Method of concomitant variation Find a possible causal variable that varies directly or inversely with the effect Method of residues Find possible causal variable that is left over once all other effects have been traced to causes 1 Mill s methods and correlation Mill s methods only identify factors that are correlated with the effect But correlation does not establish causation What gives Mill s methods work to sort among possible causes Experiments operate like Mill s methods finding real causes amongst possible causes Must be able to independently identify possible causes before correlation can help establish causation The Importance of Hypotheses Understanding the world is not just a matter of observing it There is no simple procedure for figuring out what is causing something Need to start with a good hypothesis In order to figure out what caused TB Pasteur and Koch had to advance a hypothesis there was something that was passed from a ill person to another a germ Once a cause is proposed one can test whether it is responsible Diagramming causal relations To use correlational evidence in assessing causation it helps to portray clearly what causal relations are being hypothesized Using causal diagrams we can evaluate Whether correlational evidence does support causation What manipulations we need to perform when conducting an experiment What factors must be controlled for when experiments are not possible Use nodes boxes and arrows to represent actual and possible causal relations Nodes represent variables Arrows represent causal relations between variables 2 Developing causal graphs Representing relations between a battery a switch and a fan Three variables each in a box with its possible values Battery uncharged charged Switch open closed Fan off on Use arrow to represent hypothesized relation between variables If the value of the switch causally affects the fan put an arrow between them Switch open closed Fan off on Developing causal graphs 2 Does the state of the battery causally affect the fan Battery uncharged charged Fan off on If there are two independent causes use an arrow for each Switch open closed Fan off on Battery uncharged charged No arrow from Switch to Battery if the value of switch does not affect the value of battery Developing causal graphs 3 These are NOT circuit diagrams power flows from the battery through the switch but there is no causal affect of the battery on the switch Switch open closed Fan off on Battery uncharged charged Note with the above circuit diagram there will conditions under which the switch will not affect the fan but as long as there are conditions under which it will a causal arrow is used 3 Negative causation Sometimes a cause reduces rather than increases the value of the effect variable Flu shots and flu Still use arrow between nodes Flu shot yes no Flu no yes But add minus sign to indicate direction of effect Indeterministic causes When causes suffice to produce their effects we speak of them as determining their effects Causal determinism Causation does not require determinism Some causes are only contributory Such causes raise the probability of the effect without insuring its occurrence Example smoking and lung cancer Diagramming indeterministic causes In diagramming we do not distinguish between deterministic and indeterministic causes Driving intoxicated yes no Accident yes no Dying yes no The arrows in this diagram are justified if the probability of having an accident is raised by driving intoxicated And there is no other cause that is intermediate or common that screens off the effect 4 Causal intermediates Consider lighting a match What is directly produced by the striking action Match struck yes no Match lit yes no Tip temperature 350 350 In this case if the match tip does not get above 350 the match will not light no matter how much it is struck Therefore no direct arrow from Match struck to Match lit How do we detect causal intermediates Match struck yes no X Tip temperature 350 350 350 Match lit no yes no What if we prevent the temperature of the tip from exceeding 350 The correlation between match striking and match lighting is lost Preventing the temperature of the tip from exceeding 350 screens off the match lighting from the match striking now no change in the value of Match struck affect the value of Match lit Mediated cause vs direct cause Consider the light in your refrigerator What happens when you close the door Case Door Light 1 Open On 2 Closed Off It looks like the causal graph should be Door open closed Light on off 5 Indirect vs direct causation But then you discover the light switch Case Door Switch Light 1 Open Up On 2 Open Down Off 3 Closed Down Off No situation in which changing the value of the door variable alone will change the value of the light Door open closed X Switch down up down Light off on off Direct cause or common cause A thunderstorm wakes Joe up in the middle of the night He goes downstairs to get some milk to help him get back to sleep On the way to the refrigerator he notices that the barometer has fallen a great deal Joe concludes that the storm caused the barometer to fall and draws the following causal diagram Storm yes no Barometer low high Common causes In the morning Joe tells his wife about his conclusion and shows her his diagram She is not very impressed and tells him that it was a drop in atmospheric pressure that caused both the barometer to drop and the storm She shows him her diagram Atmospheric pressure low high Barometer low high Storm yes no 6 Common causes 2 What is the difference between direct causation and common cause Altering the value of Barometer alone will not affect the value of Storm Altering the value of Storm alone will not affect the value of Barometer Storm is screened off from Barometer Atmospheric pressure