Recall
•
IBE is a common reasoning form, often used in science
•
What distinguishes good from bad IBE?
•
What distinguishes good from bad science?
•
What distinguishes good from bad theories?
•
What is science?
Science
•
Pragmatically, science is a trusted source of factual
knowledge
•
Scientists often enjoy prestige and authority
•
‘Scientific’ denotes reliability, honesty, accuracy, care in
production, generally positive associations.
•
The claim that a statement is ‘scientific’ isn’t especially
helpful (compare ‘reasonable’)
•
What makes a statement (argument, method) scientific?
Science is...
•
Just the facts
•
•
A method
•
•
Mostly fit retrospectively or partially
Naturalism
•
•
No science proceeds theory free
How to distinguish supernatural from natural
Verifiability & falsifiability
•
Both are a matter of degree which don’t demarcate
An attitude
•
No need to appeal to unwarranted objects (phlogiston)
•
Always allow the possibility of error (ID pledges)
•
Domain specific application of many critical thinking ideas
•
•
In short, a rejection of dubious statistics, biases
(psychological & otherwise), fallacious reasoning, sloppy
methods, irreproducibility, domatism
Does not provide a sharp science/pseudo-science divide!
Evaluating causal explanations
•
Usually, causal explanation is about constructing an 'abductive'
argument (aka Inference to the Best Explanation, IBE)
•
These are of the form (non-deductive/invalid):
•
•
One observes O
•
E is part of the best explanation for O
•
Therefore, E.
E.g.: I see a broken vase and small, muddy feline like foot prints
nearby. My cat breaking the vase is part of the best explanation for
it being broken. Therefore, my cat broke the vase.
Finding the best explanation
•
Determining which explanation is the best can be quite difficult.
•
Generally we should prefer:
•
•
explanations that rely on already widely established theories
•
explanations that are specific (hence more evaluable)
Often the best explanation will come from/be part of the best
established theory.
Evaluating theories
•
So how do we determine which theories are good to have
and which aren't?
•
Generally, a good theory has:
1. Content: The theory should be testable.
2. Scope: The theory should be truly general.
3. Unity: The theory should integrate with other theories.
4. Accuracy: The theory should have many confirming cases
(i.e. consistent with the data).
5. Uniqueness: The theory should rule out competing
theories.
Content
•
A theory has content just in case it has consequences that
could be false
•
The more ‘non-accidental’ the consequences, the more
content it has (e.g. ] precision = ] content).
•
The content of a theory is diminished if:
•
It lacks clarity
•
It is imprecise, meaning that although quantitative predictions are
made, they have large ranges
•
It relies on ad hoc (literally 'to this') elements.
•
It is vacuous.
Scope
•
It’s truly general
•
A theory that reduces the number of beliefs we need to
have while explaining the same phenomena will be a strong
candidate.
•
Ockham's razor - simpler theories are to be preferred over
more complex ones.
m
a
lli
Si
i
rW
Unity
•
A theory is generally better if it meshes with our previous
beliefs, especially if we can derive it from those beliefs.
•
We should generally be skeptical of theories that posit
bizarre, unproven forces.
Accuracy
•
A theory should have many confirming cases.
•
Or, in contrast, we could say that theories should not be
falsified (i.e., have disconfirming cases).
•
However, falsification in science (and elsewhere) is seldom
straight forward.
•
We must consider the theory as a whole when determining
if it is best or not.
Uniqueness
•
A theory that predicts an outcome that no other theory
predicts, or, better yet, contradicts previous theories will be
very convincing.
•
This essentially gives that theory a lot of content. The more
such predictions, the more content the theory will have.
ID vs. evolution
•
Question: How do evolution and creationism compare
given these criteria?
•
Content: The theory should be testable.
•
Scope: The theory should be truly general.
•
Unity: The theory should integrate with other theories.
•
Accuracy: The theory should have many confirming cases
(i.e. consistent with the data).
•
Uniqueness: The theory should rule out competing
theories.
Kansas
•
On October 18, 2004, the Defendant Dover Area School Board of Directors passed by a 6-3
vote the following resolution:
•
Students will be made aware of gaps/problems in Darwin’s theory and of other theories of
evolution including, but not limited to, intelligent design. Note: Origins of Life is not taught.
•
On November 19, 2004, the Defendant Dover Area School District announced by press
release that, commencing in January 2005, teachers would be required to read the following
statement to students in the ninth grade biology class at Dover High School:
•
The Pennsylvania Academic Standards require students to learn about Darwin’s Theory of
Evolution and eventually to take a standardized test of which evolution is a part. Because
Darwin’s Theory is a theory, it continues to be tested as new evidence is discovered. The
Theory is not a fact. Gaps in the Theory exist for which there is no evidence. A theory is
defined as a well-tested explanation that unifies a broad range of observations. Intelligent
Design is an explanation of the origin of life that differs from Darwin’s view. The reference
book, Of Pandas and People, is available for students who might be interested in gaining an
understanding of what Intelligent Design actually involves. With respect to any theory,
students are encouraged to keep an open mind. The school leaves the discussion of the
Origins of Life to individual students and their families. As a Standards-driven district, class
instruction focuses upon preparing students to achieve proficiency on Standards-based
assessments.
ID vs. evolution
•
Essentially these criteria were used to successfully challenge
this attempt to introduce intelligent design into Kansas
classrooms
•
http://files.findlaw.com/news.findlaw.com/hdocs/docs/
educate/ktzmllrdvr122005opn.pdf
Criticisms of IBE
•
We can criticize abduction by attacking either of the
premises or the conclusion.
•
In attacking the first premise, we can either
•
•
show that it is false (i.e. O wasn't observed); or
•
show that there is a further observation that doesn't fit
with E
In attacking the second premise we can
•
apply the previously discussed criteria for evaluating
explanations in the context of a theory
Criticizing abduction (cont.)
•
In attacking the conclusion we can
•
Give a counter example to E; or
•
Derive consequences of E and show that they don't obtain
(through experiment perhaps)
Example
•
E.g.: Last night when I was out walking my dog, I
looked up at the perfectly clear sky and saw a stationary
light that hovered for a moment and then zoomed off. It
must have been an alien space ship.
•
•
You didn't see a light in the sky, it was a firefly.
•
Many lights with those properties aren't alien space
ships.
•
If it was an alien space ship, we'd know about such
things by now. If 'they' didn't want us to know, why was
it lit?
The UW meteorological society published a notice
saying that they'd launch a weather balloon at exactly
that time, and it burst shortly after takeoff.
Summary
•
A general belief or theory is justified if it is part of the best
explanation of what is observed.
•
A particular belief is justified if it follows from some
justified general belief.
•
It should be clear that this incorporates a kind of circularity.
The reason the circle is not vicious is because of the
observations (and secondarily some of the criteria of what
determines a 'best' explanation).
Testing explanations
•
We want to determine which explanation is the best.
•
•
E.g., suppose people who take a certain medication are more
likely to get better than those who don't.
Possible explanations (the observation doesn’t choose between):
•
the medication caused the improvement in their condition.
•
most people get better one day after getting sick
•
something else they ate/did cured them
•
most people get better given any medicine
•
etc.
Testing explanations (cont.)
•
A first step is to do generate these kinds of possible explanations.
•
A second step is to set up a test (experiment) to determine which
explanation is the best one for the original observation.
•
That is, do a controlled experiment, so we can compare the
predictions of theories.
•
Even if you never want to do a controlled experiment, it’s useful
to know how to evaluate them, and what typical pitfalls are.
Comparing predictions
•
The predictions of theories can differ in three ways:
1. they conflict (mass is/is not constant);
2. one is more specific than the other (orbit of Mercury); or
3. one makes a prediction that the other is silent about (photons
have momentum)
•
These identify differences in content (recall: content is only one of
the criteria for evaluating theories).
Comparing predictions (cont.)
•
Conflict is the most useful kind of test
•
This occurs when two theories differ in their predicted out come
•
E.g., if there’s no difference in improvement between a placebo
group and the real group, then ‘mere intervention’ (e.g., a placebo)
explains the results otherwise (ideally) the medication is effective…
(or?)
Comparing predictions (cont.)
•
When testing hypotheses, we make a number of assumptions about
the test conditions (namely that we have controlled all relevant
variables).
•
•
So, reproduction of results across a slightly different test
conditions is important.
People sometimes speak of evidence confirming a theory. Strictly
speaking, this does not happen. Why?
Cause and correlation (cont.)
•
We can find correlations by examining two groups, one of which has
A and the other of which doesn't.
•
In this case, many alternative explanations for the presence of Z
are ruled out (since they are likely to be equally present in both
groups)
•
It is seldom, if ever, true that we can have two groups that differ
only with respect to one factor.
•
So, we are never absolutely certain about our claim that A causes
Z
•
But, we can try to maximize the plausibility…
Controlled experiments
•
To perform such an experiment, ideally, we do the following six
things:
1. identify an hypothesis
2. find a (random) sample of subjects in which neither the cause nor
the effect is currently present
3. divide the subjects into two groups on the basis of some irrelevant
feature
4. introduce the hypothesized cause into one group
5. see if the groups differ in the hypothesized effect
6. ensure that our hypothesis is the best explanation of any observed
difference
Controlled experiments (cont.)
•
In step 5. we need a measure to determine if the hypothesized effect
is more or less common in the control group.
•
Refer to our statistics discussion for how to measure such sizes
appropriately
•
Ideally, only the experimental group exhibits the effect (e.g., gets
better). This probably won't happen.
•
We want:
•
fairly large groups, and a large effect
Problems with controlled
experiments
•
Accidentally introducing other differences
•
It is extremely difficult to ensure that the cause of interest is the
only difference:
•
just introducing the cause of interest can introduce other causes
(e.g. group therapy)
•
placebo effect, i.e., being treated can make people ‘better’ (e.g.
prostate cancer)
•
more generally, being in the experimental group can introduce
changes that alter the results of the experiment
Problems (cont.)
•
Accidental biasing
•
Here are four ways that bias can be introduced into the
experiment:
•
•
preconceptions influence the recording of results;
•
patients try to ‘help’ the experimenter;
•
subtle signals to patients for the "correct" response;
•
intentional fraud.
The best way to guard against either intentional or accidental bias
is to make an experiment “double-blind” (not always possible).
Problems (cont.)
•
Ethical barriers
•
It can be unethical to introduce a causal factor is that cause is
harmful to the subjects or if that cause can be extremely beneficial
to the subject.
•
E.g. of harm, bloodletting, placebo surgery,
•
E.g. of help, "Tuskegee experiment" with syphilis, AIDS research
•
Must ensure that the benefits of the experiment are sufficient to
justify it.
Problems (cont.)
•
Economic barriers
•
Controlled experiments are often very expensive. Must pay (large
groups) for:
•
recruiting
•
tracking
•
administering the cause
•
record analysis
•
participation
Problems (cont.)
•
Other problems
•
Not all causally relevant factors can be introduced (e.g., gender,
ideology, etc.).
•
Experiments might not be worth the personal investment for an
investigator to perform such an experiment since professional
recognition is important.
Avoiding problems
•
To avoid some of the problems with controlled experiments,
another kind of experiment is often performed: An observational
study
•
Unlike controlled studies, observational studies do not explicitly
introduce the cause they are designed to understand
Observational studies
•
To perform such an experiment, we perform the following four
steps:
1. identify an hypothesis
2. identify cases of the cause in the population and the cases of the
effect in the population
3. determine if a significantly larger proportion of those with the
cause have the effect compared to those without the cause
4. ensure that the cause causing the effect is the best explanation of
observed difference
Observational studies (cont.)
•
Prospective: look at the possible causal factor and see if the effect
occurs with more frequency among those with the cause
•
Retrospective: identify cases of the effect and see how common the
cause is in those cases.
•
More generally, those without the cause are somewhat like the
control group in the controlled experiments.
•
But, we have not randomly assigned individuals to the control
group.
Problems with observational
studies
•
Observational studies share some problems with controlled
experiments. (e.g. ethical)
•
Not in general, however. For observational studies we:
•
must find adequate records
•
deal with whatever inadequacies the data contains
•
can’t eliminate placebo effects
•
have no way of randomly assigning subjects
Advantages of observational
studies
•
•
Advantages:
•
avoid certain ethical problems (e.g. smoking);
•
getting large groups;
•
paying high costs; and
•
long-term continuation of the experiment.
There are some steps we can take to avoiding the problems:
•
attempt to take a population that does not have the causal factor
but that matches those with the causal factor in every other way
Question
•
Question: