Philosophy 105: practical reasoning exam 2

evidence test

• determines whether a piece of information supports, undermines or is irrelevant to the claim

• is the claim more likely to be true if H or ~H

• should increase confidence in claim or in counter claim

strength test

• how much more likely is this if H is true than if not

• suppose H is true, how likely is this, suppose H is not, how likely is this

strength factor

• a measure of the strength of evidence for hypothesis

• SF= P(E|H) / P(E|~H)

2 major pitfalls

• one sided strength test

• heads i win, tails we’re even

one sided strength test

only looking for conforming evidence, ignoring anything disconfirming evidence

heads i win, tails we’re even

treating supporting evidence as decisive while dismissing counterevidence as irrelevant

selection effects

• survival bias

• selective recall

• selective noticing

• media/publishing bias

survival bias

only observing the survivors of a process (plane example)

selective recall

remembering some information more easily, can be tied to emotion

(serial position effect)

selective noticing

noticing what fits expectations (reading in a noisy room)

ignoring other sensory inputs

media/publishing bias

echo chambers, publication bias toward exciting or positive (headlines)

statistical generalizations

inferences that move outward, from small sample to represent larger group

ex: 70% of X, inferring based on small sample

statistical instantiation

inferences that move inward, from a known fact to a specific

ex: 60% of people support X, so 6 out of every 10 support X

sample size

larger samples are more reliable, smaller samples fluctuate

law of large numbers

as sample size increases, the mean gets closer to average of whole population

convivence samples

samples chosen because they’re easier to access, however often unrepresentative of population

selection effect in sampling

biases in who ends up in the sample

ex: only motivated people

stratified random sample

a sample that preserves proportions of key groups within populations

participation/response bias

people who choose to respond differ systematically from those who don’t

ex: certain groups refusing to participate

measures of certainty

mean, median, and mode captures the “center” of data

cognitive pitfalls

• loose generalizations & stereotypes

• representative heuristic

loose generalizations & stereotypes

over generalizing from limited cases

representativeness heuristic

assuming something belongs to a category because it “looks like” it

post noc ergo propter hoc

assuming that since B follows A, A caused B

causes v correlations

• correlation alone never proves causation

• causation usually produces correlation, but stronger evidence is needed

correlation

a statistical relationship between variables, they move together

causes

one event directly produces another, changing the cause changes the effect

three steps of a casual argument

• show correlation (establish association)

• rule out alternative explanations

• provide plausible mechanism

show correlation (establish association)

• demonstrate that two variables regularly occur together

• if A really causes B, then A & B should correlate

rule out alternative explanations

• most important & most difficult step

• even if A & B might correlate, might be caused of misleading correlations

• reliable casual reasoning = eliminating competing explanations

provide plausible mechanism

• after showing correlation, must explain how A could cause B

• ex: smoking → lung cancer

• establishing casual link requires identifying variables

base rate

background frequency of an event, ignoring it leads to faulty casual reasoning

statistical significance

assesses whether a correlation is likely due to chance

misleading correlations

• reverse correlation

• common cause

• side effect/placebo effect

• regression to the mean

• mere chance

reverse causation

B causes A, not the other way around

common cause

third factor causes both A & B

side effect/ placebo effect

effect caused by expectations or unrelation factors

regression to the mean

extreme cases naturally move toward average

mere chance

random fluctuation mistaken for a pattern

components of a proper experiment

• random assignment

• control group

• manipulation of independent variable

• blinding

• control confounding variables

• sufficient sample size

• clear out come

• ethical considerations

random assignment

participants must be randomly selected to groups to ensure similarity, reduce selection bias & confounding factors

control group

• provides baseline for comparison

• receives placebo, standard treatment, no interventions

manipulation of independent variable

intentionally change on factor to see if it produces change in outcome, establishes causality

blinding

• prevents expectations from influencing results

• single blind: participants don’t know their groups

• double blind: participants and researchers don’t know groups

controlling confounding variables

everything besides independent variable must be constant to prevent alternative explanations for the results

sufficient sample size

large sample ensures proper randomization, meaningful statistics, less chance of distortion

clear out come

• define in advance what you’re measuring & how

• prevents “moving goal post”

ethical considerations

experiments must follow ethical guidlines

prior vs new confidence

• updated beliefs depend on prior confidence and strength of new evidence

• new odds = prior odds x SF

cognitive pitfalls

• neglect of the priors

• neglect of total evidence

• ad hoc modification

neglect of priors

• ignoring how plausible claim before new evidence

• ex: over estimating rare disease risk after single pos test

neglect of total evidence

• focusing on one part of evidence while ignoring the rest

• ex: ignoring stats for specific detail

ad hoc modification

• patching theory to protect from evidence

• ex: slightly altering original conclusion

assessing priors

• hypothesis more plausible when coherence and simplicity

coherence

fits well with what we already know

simplicity

avoids unnecessary complications