SCI 105'020 Scientific Inquiry

1
SCI 105.020 Scientific Inquiry

• Evaluating Causal Hypotheses

2
Causation vs. Correlation-similarities

• Relationships between two variables
• Associations in sample statistics can be observed
  and analyzed in the same way
• Direction
• Strength
• Margin of error (ME)

Red Green
x-group k-group
Small Large
Not E E
A random experiment
A jar of marbles
3
Causation vs. Correlation-differences

• Correlation is a symmetric relationship, whereas
  causation is not
• In a causal relationship
• There is a clear temporal order between the cause
  and effect variables you cannot cause something
  happen in the past
• Causes produce their effects in the causal
  production

Red Green
Large Small
Use ashtray
correlation
45/60 (or 75)Large
45/55 (or 82)Red
causation
Lung cancer
Smoking
10/40 (or 25) Large
15/45 (or 33) Red
causation
Being Large is positively correlated to being
Red, and vise versa.
4
Differences In Hypothesis Evaluation

• Evaluating correlation
• The real-world population
• Population sampled vs. population of interest
• The sample data
• Sample sizes and observed frequencies
• The statistical model
• Random sampling
• Representativeness and possible biases
• Evaluating the hypothesis
• Strength of correlation
• Summary

• Evaluating causation
• The real-world population causal hypothesis
• Identify C E variables and state the hypothesis
• The sample data
• Design of the experiment
• Random/prospective/ retrospective
• Random sampling
• Evaluating the hypothesis
• Effectiveness of the causal factor
• Summary

5
The Ideal Case

• Ideally, wed like to divide the real-world
  population into two disjoin group
• Subjects exposed to the cause variable
• Subject not exposed to the cause variable

x-group k-group
exposed
Not E E
x-group
not exposed
k-group
6
Random Experiment

• In reality, one can only mimic the ideal
  condition by random sampling

exposed
x-group
x-group k-group
Not E E
population of interest
population sampled
Random sample
k-group
not exposed
Random sampling
7
Principles of Experimental Design

• The fundamental principle is control
• The key is to control for possible effects of
  extraneous variables
• Principles of control include
• Comparison observe any differences in effect
  variable
• Randomization impacts of extraneous variables
  can be balanced out
• These can then be attributed directly to the
  cause variable
• Blindness further improve the effectiveness of
  the experiment by resolving the placebo effect

8
Prospective Studies

• Not controlled experiments, inherently not as
  strong
• Widely used in various medical contexts where a
  real controlled experiment is not applicable
• Efforts can still be made to minimize the impacts
  from other variables (O)
• Select only subjects not exposed to O
• Select only subjects exposed to O and make sure
  there is no interaction between o and the
  suspected cause
• Matching the subjects according to their values
  in O

9
Prospective Studies-an illustration

• Starting from a real population in which its
  instances have selected their groups already
• Random sampling is still achievable in selecting
  subjects from the divided population

C Not C
x-group
population of sampled
k-group
E
E
Random sampling
10
Retrospective Studies

• Retrospective studies mean to survey the past
• Just opposite to prospective studies, which look
  into the future
• Starting from a x-group with subjects that are
  known to have the effect and a k-group which is
  free from the effect
• Ideally the subjects in the k-group will match up
  with the x-group subjects in all other aspects
• Efforts are needed to verify it as far as
  possible
• The result of the study was presented as
  percentages of subjects who had the cause, not
  percentages of those who developed the effect

11
Retrospective Studies vs. Survey Sampling

• Both employ techniques such as interviewing
• Both can come up with various frequency values
• A survey can also select subjects randomly
• Why should one bother to do a retrospective
  study?
• The difference is numbers!