Causal Arguments

1
Causal Arguments
2

• A causal argument draws the conclusion that one
  group of events lead to another group of events.

3
The Nature of Causal Relations
4
Effects follow upon their causes

• Effects can occur either after their causes, or
  simultaneously with them, but they cannot precede
  their causes.

5
Causes and effects are correlated

• If we think A causes B, then events like A must
  be correlated with events like B.

6

• These are necessary conditions for a causal
  relation. If A is the cause of B, then A and B
  must be correlated, and B must follow upon A.

7

• The fact that B regularly follows upon A is not
  itself sufficient to conclude that A and B are
  causally connected. To suggest that it is commits
  the following fallacy

8
Post Hoc Fallacy

• The name is taken from the Latin post hoc ergo
  propter hoc after this because of this. Simply
  because one event follows another in temporal
  sequence doesnt necessarily mean they are
  causally connected.

9
Example

• Just because three babies were born with three
  eyes after the plastics factory burned down is
  not itself sufficient reason to conclude that the
  burning of the factory caused the birth defects.
  There might be other factors responsible for the
  defects that happened to coincide with the fire.

10
Fallacy of Jumping from Correlation to Cause

• We cannot infer that wherever there is a
  correlation between two events, one causes the
  other.

11
Correlation vs. Cause
12
What is a correlation?

• Two properties or events are correlated if and
  only if occurrences of or changes in one are
  accompanied by occurrences of or changes in the
  other.

13
There are two kinds of correlations
14
1. Positive Correlations

• There are more occurrences of A among members of
  B than among non-Bs.

15
Example

• If there are more occurrences of breast cancer
  (A) among Canadian women (B) than among American
  women (non-Bs), then A and B are positively
  correlated.

16
2. Negative Correlations

• There are fewer occurrences of A among members of
  B than among non-Bs.

17
Example

• If prostate cancer (A) occurs rarely in male
  children (Bs) but frequently in men over 50
  (non-Bs), then being a child is negatively
  correlated with having prostate cancer.

18
Why cant we conclude that there is a causal
relation where there is a correlation?
19

• 1.  In the case of many positive correlations,
  there might very well be a causal relationship
  between A and B, but it could be that A caused B,
  or that B caused A. Without further evidence
  there is no way to decide between these two
  possibilities.

20
Example

• Drug use and poverty are positively correlated,
  and there may be a causal relation between them,
  but poverty might cause drug use or drug use
  might cause poverty. Either one is possible.

21

• 2. In any correlation between A and B there might
  be a third factor C that causes A and B, in which
  case A neither causes B nor is caused by B.

22
Example

• Manual dexterity is correlated with intelligence,
  but it is unlikely that one causes the other.
  They might both owe their occurrences to a third
  factor (a common cause) such as brain development.

23

• 3. In the case of any correlation, the
  correlation might be a coincidence. For instance,
  the myth that the stork delivers babies is based
  on such a correlation.

24
Control Groups

• Identifying a genuine correlation is itself a
  difficult task. It is not enough to notice that
  two things happen together frequently to have a
  correlation that will support a causal connection.

25

• To establish a positive correlation between X and
  Y we need to see that Y occurs more frequently
  among members of X than among non-members of X.

26
Example

• Is there a correlation between women with
  silicone breast implants and women who develop
  connective tissue disorder? Even if many women
  with breast implants develop connective tissue
  disorder, that might not establish a genuine
  correlation.

27

• We need to determine whether or not connective
  tissue disorder occurs in high numbers of women
  without breast implants.

28
This requires a comparison between two groups
29
A Test Group

• The members of the test group possess the
  property whose causes or effects we want to
  study. In this case it would be women with
  silicone breast implants.

30
A Control Group

• A group that does not possess the property whose
  causes or effects are under investigation.

31

• The control group must be as similar as possible
  to the test group, with the exception of lacking
  the property being studied.

32

• For a study on breast implants, the control group
  will not consist of men, or five year old girls,
  but women in the same age-group as those with
  breast implants, of the same general level of
  health as the women before they had their
  implants, and so on.

33

• If we discover a significantly higher occurrence
  of connective tissue disorder among the test
  group than the control group, then we have a
  correlation between having breast implants and
  developing connective tissue disorder.

34
The strength of the conclusion

• In any causal argument, just like an argument
  from experience, or an analogical argument, the
  conclusion can only be shown to be probably true,
  not certain.

35
Types of Studies

• Depending on the kind of study that has been
  done, there are different questions we should ask
  to evaluate causal arguments.

36
Correlational Research

• The researcher does not control any of the
  conditions. This kind of research simply involves
  collecting data.

37

• This is most commonly used when it would be
  unethical to expose research subjects to the
  causal factor under consideration, since it might
  be harmful.

38
Controlled Laboratory Experiment

• The researcher controls all of the conditions.

39

• These kinds of experiments are only performed on
  non-human animals, such as rats. This involves a
  tremendous level of control over every aspect of
  the lives of the test and control groups. It is
  not feasible or ethical to give experimenters
  that much control over human beings.

40
Control Group/Test Group Experiment

• The researcher controls the causal factor (the
  substance the causal effects of which are the
  object of the study).

41

• Other conditions are not controlled. The group of
  subjects is divided into the test group and the
  control group. The test group is given the causal
  factor and the control group is not. All
  participants must be made aware of any potential
  risks created by exposure to the causal factor.

42

• The participants in the study (both in the test
  and control groups) must be representative of the
  population. If the drug is one like Viagra, then
  there isnt much point in having women
  participate in the study.

43

• Also, the members of the control group must be as
  similar as possible to the members of the test
  group to ensure that it is the causal factor, and
  not something else, that gives rise to
  differences between the two groups later on.

44
Replication

• For the results of an experiment to be reliable,
  the results must replicable. The experiment is
  one that can be repeated with the same results
  under the same conditions.

45

• If the results cannot be replicated, then it is
  likely that some of the conditions were not
  properly controlled, affecting the results.

46
Means of controlling experimental conditions
47
Blindness

• In a blind experiment, participants do not know
  whether they belong to the control group or the
  test group.

48

• The purpose of blind experiments is to prevent
  subjects from affecting the results because of
  their expectations.

49
Placebo effect

• The main aim is to avoid the placebo effect. This
  occurs when someone who receives a placebo (an
  inert substance given to the control group)
  reports feeling certain effects because he or she
  believes they are receiving the causal agent.

50
Double-blindness

• In a double-blind experiment neither the
  participants nor the experimenters know who is in
  the control group and who is in the test group.

51

• This is to prevent the placebo effect among the
  test subjects, and to prevent experimenters from
  tainting results either by cuing test subjects or
  by looking at data with a set of expectations.

52
Evaluating Causal Arguments
8 questions to ask

• 1.    What is the causal claim being tested?
• 2.    What is the sample?
• 3.    What is the population?
• 4.    What kind of study is involved?

53

• 5.    What is the test group?
• 6.    What is the control group?
• 7.    Are the test and control groups
  similar?
• 8. How are the results measured?

54

• Pop music may help schoolchildren pass exams. In
  a nationwide British study, 11,000 students in
  250 schools were randomly split into three
  groups. They listened to either Mozart, the pop
  group Blur, or a radio chat show, while taking a
  test on spatial reasoning.

55

• The students who listened to the pop group scored
  56 the other two groups, 52. The difference
  approached significance. The author of the study
  cited a California study in which adults
  performed better on a similar test while
  listening to Mozart, and said that this may show
  that adults process music differently.

56
The Causal Claim Being Tested

• That pop music may help students pass exams.

57
The Sample

• 11,000 Students.

58
The Population

• Students.

This should be more specific. Grade school
students? High school, university?
59
Kind of Study

• Control group/test group experiment.

60
Test Group

• Students who write the test while listening to
  Blur (though the other two groups could each be
  the test group as well).

61
Control Group

• Any two groups serve as a control group for the
  third. In this case, given the causal claim
  tested, the groups that listen to Mozart and
  talk-radio are the control groups.

62
Are the Test Group and Control Groups Similar?

• Difficult to say. We have not been given any
  information about this. They are similar to the
  extent that they are all students of the same
  nationality, but we know nothing of their ages,
  gender, etc.

63
Measuring Instrument

• The scores on tests on spatial reasoning.

64
Overall Evaluation

• The size of the sample is good. Quite large, and
  the number of schools that participated in the
  study is high enough to be representative of
  schoolchildren in general.

65

• One problem we have already seen is that neither
  the sample nor the population is well defined in
  terms of age. So we should wonder about how
  representative the sample is of the population.

66

• We should wonder about the control groups used.
  The results might be more plausible if a silence
  condition were used in one. Perhaps the students
  scored more poorly than they would have without
  any background noise at all, but since they
  probably like Blur, were less distracted by it
  than by Mozart or talk-radio.

67

• Only one kind of test was administered. It is
  unlikely we can generalize to all kinds of tests
  (exams) on the basis of how the students score on
  spatial reasoning.

68
Does the study give the conclusion strong
support?

• It seems not, in light of the above problems.
  Furthermore, it was claimed that the differences
  between the test and control groups merely
  approached significance.

69

• Without a more definite result there is little
  reason to think that there is a causal
  relationship between music and passing exams.

70

• What about the claim about the differences
  between adults and children?
• Do the test results suggest that adults and
  children process music differently?

71

• Probably not. It is more likely that adults are
  more familiar with Mozart than with Blur, in
  which case what the study more plausibly shows is
  that adults are less distracted by Mozart than
  students are.

72

• Day care is dangerous for infants. Studies
  conducted on children in war refugee camps and
  wartime orphanages during and after the Second
  World War show that these children were likely to
  suffer permanent damage. Experiments on baby
  monkeys, who were deprived of their birth mothers
  and given substitute mothers constructed of
  wire-mesh, showed that the monkeys suffered
  severe emotional distress.

73

• Dazzle laundry detergent is the best laundry
  detergent money can buy. We washed the soccer
  uniforms worn by Janie and her two friends after
  their championship game. We washed Janies
  uniform in Dazzle and her friends using two
  leading competitors. Janies uniform came out
  white and bright. Her friends were dull and grey
  by comparison. Dazzle gets clothes their
  cleanest.