Research

Front 1

Inus condition

Back 1

an insufficient but non-redundant part of anunnecessary but sufficient condition (Ex. A lighted match in starting a forestfire)

Front 2

Counterfactual model-

Back 2

something contrary to fact(ex. the difference between a control group and the experiment group. Whataffect the action had on the control group) This is an ideal but it isimpossible to get a perfect representation.

Front 3

Rubins casual model

Back 3

a modelused for cause inference in case-control studies

Front 4

John Stuart Mill-

Back 4

cause and effect are relatedif (1) the cause preceded the effect,(2) the cause was related to the effect (3) we can find no plausiblealternative explanations for the effect other than the cause. Therefore experimentsare the best scientific method for this end.

Front 5

Confound

Back 5

a third variable that explains a correlation between two other variables

Front 6

Popper-

Back 6

it’s easier to find falsifying evidence than toprove something

Front 7

Kuhn Falsification dependson two assumptions

Back 7

1. The causal claimis perfectly specified.2. Falsification requires measures that areperfectly valid reflection of the theory being test

Front 8

Cronbach – two problems withgeneralization

Back 8

1. Generalizing tothe “domain about which the questions is asked”, which he called UTOS 2. Generalizing to“Units, Treatments, Variables, and settings not directly observed which hecalled *UTOS

Front 9

Correspondence Theory

Back 9

knowledge claim is true if it corresponds to theworld – example we can tell it is raining if we see rain

Front 10

Coherence theory-

Back 10

says that a claim is true if it belongs to acoherent set of claims – If A is true then B is true then C is true so A causesB

Front 11

Pragmatism-

Back 11

a claim is true if it is useful to believe thatclaim – “electrons exist” because going off that brings meaning andpredictability to the observations

Front 12

Three Principles ofQuasi-experimentation

Back 12

1.Identification and studyof plausible threats to internal validity2. Primacy of Control byDesign (To eliminate threats of internal validity (by design or by statistics)3. Coherent Pattern Matching

Front 13

Predictions from

Back 13

In this order:

1. Abstract Casual Theories

2. Authorities Prescription/assertions

3. Assertions of interest groups

4. No theories

Front 14

Five factors

Back 14

n -semi fixed

delta - can control by choosing a context where it is expected to be large

Correlation - Can control by incorporating/ ruling out other causes (e.g., Vsand Zs)

alpha - fixed

beta - risk of failing

Front 15

with R XO

and R O

How do you improve

Back 15

a. ANCOVAto estimate and remove difference due to V

Front 16

Non random asssignment

How do you improve

Back 16

Make groups more equivalent by deleting non overlappingobservations before trt

Front 17

Main contributions to research

Back 17

1. New Data

2. New estimation

3. New Theory (or Problem)

Front 18

Different Types of Validity

Back 18

1. Statistical

2. Internal

3. Contruct

4. External/

Front 19

Ways to help with V's and Z's

Back 19

use equal sample size

measure better

increase strength of treatment

Increase variability

Switch Subject

Front 20

Mautration

Back 20

some event (naturally occuring that affects the experiment

Front 21

Instrumentation

Back 21

the masure device wears

Front 22

Conflation

Back 22

mixed effect of one factor with another

Front 23

Three different spheres of research

Back 23

Accounting

Auditing

Professional Structure

Front 24

Improving One group pretestand posttest design

Back 24

Double Pretest - add anotherpretest before the first one to understand any fluctuations before treatment isconsideredNonequivalent DependentVariable – Having a second variable that shouldn’t be affected by the treatmentbut measures similarly the experiment’s dependent variable (ex. toys advertisedaround Christmas time vs. Toys that are not)

Front 25

Ancova

Back 25

Analysis of covariance look at the projection of the points parrelell on the line on the y axis ( use when their is no overlapping

Front 26

Anova

Back 26

analysis of variance (just project to the y axis and look at the distribution based on that (can use when their is alot of overlap with V values)

Front 27

Improving Quasi- Experimental Design w/o pretest (3 ways)

Back 27

Independent Pretest Sample- draw from a random group from both the control and the testgroup (cross section) this is used when they think the prestest may have anaffect.Proxy Pretest –use a proxy to measure what the V’s and Z’s would have been (ex. algebra finalgrades if measuring effects of a calculus class)Matching or Stratifying – Example – using twins in a study

Front 28

Types of Matching

Back 28

Index

Cluster

Benchmark

Optimal

Front 29

Index

Back 29

Index Matching-selects multiple control units above and below a treatment unit

Front 30

Cluster Group

Back 30

- used to embed the treatment group in a cluster of similar control units

Front 31

Benchmark Group matching

Back 31

– selects control units that fall close to the treatment unit on a multivariate distance measure. Studies indicate that cluster and benchmark are better than index

Front 32

Optimal matching

Back 32

- each treatment unit may have multiple matched controls and vice versa

Front 33

Problems with Matching

Back 33

Selection Bias– you are picking the control groupsUndermatching –not finding the predictors of the outcomeIn general matching -may create an answer that doesn’t actually exists.

Front 34

Principles to bettermatching

Back 34

(1) Select groups that are assimilar as possible before matching and look at overlapping sections(2) Use matching variables that are stable and reliable (if theycorrelate with the outcome) [Stable Matched Bracketing]

Front 35

Four threats too Quasi- Experimental Design w both control group/pretests

Back 35

Selection Maturation

Selection-instrument threat

Selection-regression-(described in last reading)Selection-history –

Front 36

Selection Maturation

Back 36

– may arise if respondents in one group are growing more experienced, tired, or bored than respondents in another group

Front 37

Selection-instrument threat

Back 37

– nonequivalent groups begin at different points on the pretest. They are more accurate (1) the greater the initial nonequvalence between groups, (2) the greater the pretest posttest change, and (3) the closer any group means are to one end of the scale, so that a ceiling or floor exits.

Front 38

Selection-history

Back 38

– the possibility that an event occurred between pretest and posttest that affected one group more than another

Front 39

Five Outcome Patterns

Back 39

1. Both groups grow apart in the same direction – ( maturation)

2. No Change in the Control Group – Could dissect the data based on other Z’s or V’s but in generalyou can’t rely on the trend

3. Initial Pretest Difference Favoring the Treatment Group thatDiminish Over Time- Less likely due tomaturation

4. Initial Pretest Differences Favoring the Control Group thatDiminish Over Time – Subject to scaling(selection-instrumentation) and local threats (selection-history) probably notselection maturation

5. Outcomes that Cross Over in the Direction of Relationshipsregression risk,selection-instrument threat and selection maturation is less likely (harder toconclude; should get stronger design controls)

Front 40

Elements of Design

Back 40

1. Assignment

2. Measurement

3. Comparison Groups

4. Treatments