Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
90 Cards in this Set
- Front
- Back
Compte, August
|
- Positivism (1830s)
|
|
What is Positivism?
|
Every phenomena has objective reality, which can be studied through direct experience - the only way to reach a conclusion. The more direct experiences you have, the more likely the truth will reveal itself. Reveal that one's experience is consistent or inconsistent with reality.
|
|
How to find truth in positivism?
|
must observe numerous direct experiences. one is not good enough. EX - Avocado problem
|
|
What was critical change in positivism?
|
If you cannot directly experience something, you cannot study.
|
|
Who founded logical positivism?
|
Vienna Circle
|
|
Popper, Karl
|
Popped the positivism bubble.
Principle of Falsification direct experience opens door for interpretation rather than objectivity. Ex - pencil question also White Swan problem |
|
White Swan Problem
|
The “white swan problem”
Premise: All swans are white But: There is a black swan I don’t consider it to be black. It is a dirty white swan Therefore, “objective reality” is consistent with my subjective experience Popper says both statements are correct |
|
Principle of Falsification
|
a theory must be able to hold up to many attempts to knock it down
|
|
Hume, David
|
nothing can ever be show to be absolutely true; always a limiting conduction.
|
|
Duhem and Quine
|
Popper's notion to reject all theories if falsified too extreme.
"Conventionalism" - evidence against theories mean that those theories need to be modified. Theories evolve, not survive |
|
Khun, Thomas
|
scientists are "ego-involved" in theory, meaning that researchers will try over and over again to prove their theory, but not to falsify.
Popper is good idea in theory, but not in practice |
|
Ad Hoc Hypothesis
|
ego-involved result, as states by Khun, Thomas
|
|
Example of Ego-Involved
|
Plate Tectonics
|
|
The Structure of Scientific Revolutions
|
Kuhn's Book (1962)
|
|
Lakatos, Imre
|
Khun is consistant with Popper. Lakatos said that scientists falsify programs, not theories.
not necessarily rejecting theory, but rejecting new way of thinking. Ex - Plate tectonics |
|
Research Programme
|
Lakatos said that theories are succession of techniques developed over time
|
|
Lakatos beliefs about ad hoc hypotheses
|
important start of a paradigm
|
|
Feyerabend, Paul
|
society falsifies programs; the method is irrelevant
|
|
Eugenics, LSD on Schitzo, Medical Marijuana
|
Associated with bad thoughts about society. Methodology may be sound, but society rejects any notion that this can be studied
|
|
Meehel (1920 – 2003))
|
Inferential statistics are meant to test point predictions under conditions of exact measurement.
applying inferential tests to soft data is incorrect |
|
Meehel's view of null hypotheses
|
automatically false because there exists Measurement error, meaning there will be variation in scores
|
|
Meehel's view on Testing a Theory
|
Testing a theory under certain conditions. If the theory is wrong, the conditions are likely wrong
|
|
Meehel and the Mean
|
Generate likely level for the mean value and then get confidence interval
|
|
Meehel and ANOVA
|
soft data using ANOVA is incorrect
|
|
Meehel finds evidence to support theory
|
Evidence in support of theory does not mean theory is correct
|
|
IF P, then Q.
Q. Then P |
Affirming the consequent
|
|
If not Q, then not P
|
Contraposition
|
|
Does Meehel think theories are evaluated correctly?
|
No. Need to test what should NOT happen, as well as what SHOULD happen. consequent and contrapositive.
|
|
According to Meehel, how should evaluate theories?
|
Estimate population values, estimate confidence intervals, estimate effect size, collect data, and later meta-analyze
|
|
What does it mean to say one thing causes another?
|
Aristotle: "an agent producing a change."
Descartes: mechanistic View - World is a big machine; everything has a cause |
|
Mechanistic View
|
Descartes - World is a big machine; everything has a cause
|
|
Hume - Three Needs for Causality
|
Three needs for causality:
Contiguity: Cause and effect must connect Priority: The cause comes before the effect Constant conjunction: - Same cause always leads to same effect - Specific effect always results from same cause Example: 2x4 causes headache - does not have constant conjunction |
|
Three needs for causality example
|
Starting my car.
Contiguity. I continue to turn the key until the car starts Priority. The ignition will not turn over unless I turn the key Constant conjunction. Turning the key always makes ignition spin If the ignition is spinning, it’s because the key has been turned Therefore, turning the key causes the engine to start |
|
Possible 4th Requirement for Causality
|
Necessary Connection -
What does B follow A? Can't directly experience connection. Compelled by instinct to believe connection exists if connection happens frequently. |
|
Positivists view on Necessary Connection
|
It is not relevant
“When I experience A, I experience B immediately after” Can’t experience connection, so irrelevant EX - can't see electricity in wires |
|
Realists view on Necessary Connection
|
Necessary connection relevant
It is objectively there Opens up possibility to study unobservables |
|
Problem of Coincidental Events
|
John Stuart Mill - Wednesday
Contiguity - Wednesday is connected to Thursday Priority - Wednesday comes before Thursday Constant Conjunction - Wednesday always comes before Thursday If it is Thursday, it was Wednesday Therefore, Wednesday causes Thursday. NO!! It is a coincidence that the days are ordered. |
|
John Stuart Mill
|
coincidental events and eliminative inference
|
|
Eliminative Inference
|
John Stuart Mill
Identify, test, and rule out possible causes of an event Necessary and sufficient conditions for causality Four methods: Agreement, Difference, Residues, of Concomitant Variations |
|
Mill's Four Methods to establish conditions of Causality
|
Agreement: If B is present, A is present
Difference: If B is Absent, A is Absent Residues: If A and X are paird, and B and Y are paired, and X is known to cause Y, then A causes B Concomitant Variations: As A changes, B changes Then, A causes B if ALL FOUR ARE MET Each of these is necessary to establish causality But it is only sufficient when all four are supported |
|
If 3 out of 4 of Mill's conditions for causality are met, does A cause B?
|
NO. Each of these is necessary to establish causality
But it is only sufficient when all four are supported |
|
Example of Mill's Necessary Conditions
|
If i turn the key, the engine starts (Agreement)
If the engine is not on, the key is not turned (Difference) If I turn the key (A) and pull the windshield wiper lever (X) down, the engine starts (B) and the wipers (Y) move, and I know the wiper lever causes the wipers to move, then A causes B. The point here is to eliminate possibilities. (Residues) If I turn the key forward and turn the key back, the engine starts and stops (Concomitant Variation) Then A causes B |
|
Ronald Fisher (1890 - 1962) History
|
Astronomy student at Cambridge
As undergrad, became interested in question of how to establish “true” difference between groups, given error variation First job, working with fertilizer data, led him to take up question of how to execute careful experiments started ANOVA ANCOVA degrees of freedom |
|
1935: The Design of Experiments
|
Fisher
|
|
Method of Inductive Inference
|
Can only test likelihood of null being true
The parameter of interest is equal to population value The parameter of interest is equivalent across groups What is the probability null is true, given data? Probability you would observe statistic value by chance p-value If low, null is poor descriptor of data So p should be treated as continuous value |
|
Fisher and Degrees of Freedom
|
Statistic has to be evaluated in light of DF
|
|
Degrees of Freedom
|
The number of data points free to vary, given a fixed mean
Some data points do not contribute to variation - must evaluate points by considering truly variable points |
|
Lower P Value
|
Reject Null
More confident that Null is False |
|
High P Value
|
Retain Null
Less confident Null is false |
|
Region of uncertainty
|
Fisher
Level of probability where you’re not sure what to think Eliminates the need for absolute cutoff Go get more data! |
|
What’s going on if null is retained?
|
Fisher says, Who Knows?
- Null could be correct - Null could assess wrong things |
|
How do I know what I should have measured?
(Fisher's Answer) |
You can't ever know.
Who cares? Goal of science is to show null is false. |
|
Fisher Summarized
|
- P-value is continuous
- Lower P-value, reject null - Higher P value, accept null - Region of Uncertainty - go get more data - Alternate Hypothesis is irrelevant because goal of science is to reject null |
|
Egon Pearson (1895 – 1980)
Jerzy Neyman (1894 – 1981) |
They care about alt. hypothesis
Inductive behavior approach to hypothesis testing Either null or alternate must be true Need a decision criterion. dichotomous question - if null is wrong, another explanation is right |
|
Type I and Type II errors
|
Type I: Support alternate when not true (SANT)
Type II: Support null when not true (SNNT) |
|
Decision Criterion
|
sufficiently tough so as to minimize likelihood of Type I error
State desired p(Type I error) Identify exact value of statistic that corresponds to this value Compare calculated statistic to this value If calculated value is more, ALT is true. If calc value is less, Null is true |
|
Pearson and Neyman's views on P-value. Why?
|
p-value is just a tool. It has no explanatory value
Does not matter what p-value is associated with your calculated statistic b/c You are making a dichotomous decision |
|
Why is convention of P-Value set to 0.05?
|
Fisher (kind of) suggested it!
“Convenient to work with twice the sd” as the point at which null is rejected In normal distribution, ±2 sd leaves approx 5% of points in the tails No one knows where he got this Thus p = .05 is arbitrary value |
|
Neyman and Pearson's view on Power
|
need to work in an environment that minimizes likelihood of decision error
|
|
Fisher's view on power
|
not an issue with Fisher approach. No absolute decision
|
|
What if you fall just short of criterion?
Critical value = 2.69; you calculate 2.72? |
Major criticism of N & P approach
Fisher, no problem |
|
Neyman and Pearson Summary
|
dichotomous - either null or another reason is true
P-value is a means to an end |
|
Do researchers today use N & P or Fisher?
|
Combination of both - "greatest hits"
Wrongly |
|
Power
|
How much power is behind your study
Ability to detect an effect, given that the effect exists 1 - B (probability of Type II Error) Type II error is when you select no effect when there is an effect |
|
Effect
|
Why research is important
|
|
Four Factors of Power
|
Type 1 Error Rate - there is an effect when there is not
Sample Size - Stronger case for power is a small sample size Difference between treatment and null means (d) Standard deviation (s) |
|
Maximum Power
|
Large type one error rate (easy to find)
Sample Size (large - many cases) Large Difference between treatment and null means Small standard deviation |
|
Type I Error Example
|
Over magnification is a bad thing. We want the type I error to be large, but in reality, want it to be small
over magnification is like seeing too many Jupiters |
|
Sample Size Example
|
Baseball bats cause concussions
small case is better than large sample size. can find anything with large enough sample size. influences effect size |
|
Difference between treatment and null means
|
an Exact value.
Where do we get it from? Estimated guess based on literature |
|
Standard Deviation
|
Sources of error not dealt with
to get small standard deviation, thoroughly plan design |
|
Interpreting "Power"
|
0 - 1.00 Scale
0.99 is MAX - how many times would find effect if repeated study 100 times over 100 samples 0.8 and up is considered HIGH POWER 0.6 - 0.79 - study is salvageable, but need modifications > 0.6 - study is flawed |
|
Pearson's Goal of Scientific Research
|
To reproduce population values as accurately as possible.
|
|
. Three alternatives to significance testing have been proposed by followers of Meehl. Identify and explain them. (3)
|
Meehl himself advocated estimation of population-level distribution values.
Cohen advocated the reporting of effect sizes that tell us the magnitude of the impact of the variable in question. Hunter and Schmidt argued for simple aggregation of data, followed by meta-analysis to determine the relationships within the aggregated data. |
|
How is today’s typical approach to theory testing an example of the logical fallacy of affirming the consequent? What else needs to be done in order to truly evaluate a theory? (2)
|
Researchers only attempt to find evidence in support of the theory’s predictions. A true test of a theory requires also testing for things that the theory says should not happen. This is the principle of contraposition.
|
|
A positivist would argue that it is impossible to study emotions. Explain why not. (2)
|
An emotion will not be experienced in the same way by different people—there is no objective emotion. For example, different people will describe “being happy” in different ways. If there are no objective emotions, there is no objective truth to discover.
|
|
A researcher sets her Type I error rate to .05. She executes an analysis of variance and obtains a result that is associated with a p-value of .02. She concludes that her result is “highly significant.” Explain why her statement is consistent with neither Fisher nor Neyman and Pearson. (2)
|
Fisher said that the whole point of scientific research was to reject the null hypothesis. In his work, there is no Type I error.
Neyman and Pearson said that the p-value is irrelevant |
|
Numerous studies have shown that the DARE program, which is designed to teach elementary-school children how to avoid drugs, does not work. Despite this, DARE remains a heavily-used program in American elementary schools. Which philosopher of science would predict this continued popularity of DARE? Explain your reasoning. (2)
|
Feyerabend would point to this as an example of society deciding that research program is worthy, even though the scientific evidence goes against this notion.
|
|
. Explain why, according to Meehl, it is appropriate to use significance testing in the hard sciences but not soft sciences. (2)
|
Significance tests are designed to test point predictions. Models in the soft sciences do not generate point predictions, but rather general predictions (“Treatment will produce an improvement relative to control”). It is inappropriate to analyze such data as if a point prediction had been made.
|
|
Explain what Popper meant by “falsification” of a theory. According to him, what does one do should a theory be falsified? How did Duhem and Quine modify this idea? (3)
|
Falsification means that it should be possible to demonstrate that a theory is wrong.
If this happens, Popper said the theory should be discarded. Duhem and Quine said that, rather than discarding the theory, its weak points should be identified and modified. The modified theory then needs to be tested under the conditions in which the original theory worked, to make sure the modified theory also works. |
|
Effect Size
|
Measure of the magnitude of an effect in the population
Important because extreme power will identify many trivial effects as significant Many different measures of magnitude of effect keeps Power in check |
|
η2
|
Proportion of variance attributable to IV
influence in a sample |
|
ω2
|
Population-level estimate of variance proportion
influence that could exist within a population |
|
Problem with squared measures
|
Tend to underestimate effect size because squared rations get smaller
|
|
popular measures for experimental designs
|
variance ratios
|
|
Recommended Measures
|
Correlation (r)
|
|
Correlation(r)
|
Extent to which IV and DV are related
(-1, +1) - Normed +- indicates relationship (direct or inverse |
|
Standard Difference between groups (d)
|
Difference divided by pooled variance
usually bounded by (-4, +4) If d<= 1.6, d = 2.5r |
|
Is it challenging to Interpret Effect Size?
|
Yes.
|
|
Cohen's standards
|
Small effect: d = 0 - .20 (r = 0 - .10)
Medium effect: d = .20 - .80 (r = .10 - .40) Large effect: d > .80 (r > .40) Not empirically based Base interpretation on goals of study |