experimental design (7) - university of glasgokerry/level2/lev2expdeslec7.pdf · experimental...

27
Experimental Design (7) Kerry Kilborn Department of Psychology

Upload: lamhanh

Post on 13-May-2018

215 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Experimental Design (7)

Kerry KilbornDepartment of Psychology

Page 2: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Overview

• Confounding variables• Experiment vs. Correlational Study• Between-Subjects Design• Equivalent Groups• Quasi-Experiments• Summary

Page 3: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Experimental Studies

Manipulation of IV Change in DV

causal link

Alcohol level Reaction Time

Memory load Recall Rate

Drug/Placebo Pain Score

Page 4: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Sample (N = 100)

Alcohol

No Alcohol

50

50

AlcoholNo Yes

Rea

ctio

n Ti

me

[ms]

0

300

325

350

375

Page 5: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Confounding

IV Reaction Time Testing Time

No Alcohol 325 ms 10 am

Alcohol 366 ms 10 pm

Confounding Variables

Page 6: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Possible Confounding Variables

Person-specific Situation-specific

– Age Experimenter

– Education Time point of testing

– Socio-economic status Testing environment

– Motivation Apparatus

– Memory Stimulus intensity

– Intelligence Duration

Confounding Variables

Page 7: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Experimental Studies

• what happens in an Experiment:

• Manipulation of independent variables (IVs)

• Control of confounding (extraneous) variables

• Measurement of dependent variables (DVs)

Page 8: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Experiments - Evaluated

Strength Weaknessisolates cause and effect participant biascontrol of extraneous variables artificial conditions and→ high internal validity measures

→ (low) external validityelimination of alternative participants contributionexplanations completely prescribedeasy to replicate kind of studied phenomena

is limited

Page 9: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Experimental Method

• Manipulates IV and observes effect on DV

• Comparable Conditions acrossall levels of IV

• application limited• cause-effect relationship

Correlational Method

• Observes IV and DV

• Further (extraneous) variables may covary with levels of DV

• widely applicable• ambiguous cause-effect

interpretations

Experiments - Compared

Page 10: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Between-Subjects Design• Experiments compare at least two conditions A and B→ at least 2 levels of independent variable (IV)

• Subjects who participate might be placed into condition A, B or both

→ 2 different types of experimental designs

• If subjects receive either level A or B but not both → between-subjects design

• If each subject receives both levels of IV (A, B), i.e.,both levels exist within the same subject

→ within-subjects design (repeated measures design)

Page 11: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Between-Subjects Design

• Sometimes a between-subjects design must be used. If the independent variable is

• a subject-variable (e.g., anxiety, gender,..)

• manipulated in a certain way that precludeswithin-subjects measures (e.g., social Ψexperiments), i.e., participating in onecondition makes it impossible for the sameperson to be in a second condition

Page 12: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Between-Subjects Design• Example (Sigall & Ostrove, 1975):

on the influence of physical attractiveness ofa defendant on recommended sentence

• written descriptions of a crime - asked torecommend a jail

• IV1 = Type of crime (2 levels: burglary inwhich woman stole 2,200 $ vs. swindle in whichwoman induced man to invest 2,200 $)

• IV2 = Attractiveness of woman (2 levels: veryattractive vs. unattractive (vs. no photo)

Page 13: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Result

Attractiveness of woman

Crime attractive unattractive control

burglary 5.2 yrs 5.1 yrs

swindle 4.4 yrs 4.4 yrs

Between-Subjects Design

Page 14: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Between-Subjects Design

Result

Attractiveness of woman

Crime attractive unattractive control

burglary 2.8 yrs 5.2 yrs 5.1 yrs

swindle 5.5 yrs 4.4 yrs 4.4 yrs

Page 15: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Advantage

• subjects enter the study fresh andnaive with respect to procedures

Disadvantage

• large number of individuals needed

• differences between conditions might be dueto differences between groups

Between-Subjects Design

Page 16: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Between-Subjects Design

Group 1 Short Group 2 Long1 N1 17 6 N6 252 N2 16 7 A1 143 N3 19 8 A2 164 N4 20 9 A3 175 N5 18 10 A4 15__________________________________________Mean 18.0 17.4SD 1.58 4.39

• with a small number of participants it could happenthat random assignment places all A-subjects intoone group → non-equivalent groups

Page 17: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

• Creating Equivalent Groups• Random Assignment

method for placing randomly selectedsubjects into the different groups

• → equal probability for each subject to beassigned to a specific condition

• → spread possible individual differencefactors evenly across conditions

Between-Subjects Design

Page 18: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Equal probability of assignment PLUS

Allow for relevant individual differences

Between-Subjects Design

Group 1 Short Group 2 Long1 N 17 6 N 272 N 16 7 N 263 N 19 8 N 264 A1 10 9 A3 175 A2 11 10 A4 15__________________________________________Mean 14.6 22.2SD 3.91 5.72_________________________________________

Page 19: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Between-Subjects Design• MatchingPair subjects together for a specific characteristicand then assign randomly to groups. You need tomeasure the matching variable in a reasonable manner.

• Example: obtain scores for test anxiety and then sortsubjects into pairs and assign subjects from each pairrandomly to the two groups (flip a coin)

P1 N1 - N4 P2 N6 - N5 P3 N3 - N2P4 A2 - A4 P5 A5 - A1 P6 A3 - A6

G1={N1,N5,N2,A2,A1,A6} G2={N4,N6,N3,A4,A5,A3}

→ Matched Pair Design (e.g. identical monozygotic twins)

Page 20: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Between-Subjects Design

ControlGroup

ExperimentalGroup

IVLevel 1

IVLevel 2

DV

DV

Sample

1. Random Sample2. Matched

Equivalent Groups

Identical conditions except manipulation of IV

Com

paris

on

Page 21: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Between-Subjects Design

• Manipulated vs. Subject Variables

• Comparisons may be made also between groups of people who differ from each other in ways not manipulated by experimenter

• → comparison between factors which are non-manipulated variables or ex-post-facto variables

→ subject variables

• Refer to already existing characteristics of the participants in the study (e.g., gender, intelligence,age, RT)

Page 22: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Example

Group study of relationship between anger level and cardiovascular responsiveness (CR) to film scenes

a) induce different levels of anger and measure CR

b) select two groups differing in pretest-level of anger

• here subjects cannot be randomly assigned to groups

• Pre-test: measure of participants before anexperiment in order to balance or compare groups, orto assess change by comparison with scores after theexperiment

• → No "true" experiment !

Page 23: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Between-Subjects Design

• Problems with subject variablesexperimenter can not hold all other variables constantextraneous variables can not be controlled

• e.g., person with higher scores in anger may also differ in the way they cope with everyday lifesituations; they might be prone to have cardiovascular problems, ...

• → no cause-effect conclusions can be drawn in contrast to a confound free experiment

Studies using subject variables are also called ex post facto studies or quasi-experiment

Page 24: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

Between-Subjects Design

• Ex post facto researchstudy where pre-existing and non-manipulated variables among people are measured

• Quasi-experimentstudy in which experimenter does not have controlover the allocation of participants to conditions and/or the independent variable

• Group difference studystudy, which compares the measurement of an existingvariable in two contrasting groups (male vs. female,intro- vs. extrovert)

Page 25: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

University A

University B

Control GroupTraditional

Teaching Method DV

DV

Nonequivalent Groups

Com

paris

on

Experimental GroupNew InteractionalTeaching Method

Quasi-experiment

Page 26: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

ControlGroup

ExperimentalGroup

NoTreatment DV

DV

Dyslexics

Voluntary participation in dyslexia treatment program(i.e., self-selection)

Nonequivalent Groups

Com

paris

on a

fter 3

yea

rs

Treatment

Quasi-experiment

Page 27: Experimental Design (7) - University of Glasgokerry/Level2/Lev2ExpDesLec7.pdf · Experimental Design (7) Kerry Kilborn Department of Psychology. Overview • Confounding variables

• True Experiment– Manipulation of IV– Control of confounding variables

• Quasi-Experiment– Manipulation of IV– No control of confounding variables

Summary