Eval. Psych Research Final Exam
Terms
undefined, object
copy deck
- repeated measures designs
-
-RANDOMIZED GROUPS DESIGNS: diff. participants assigned each of conditions in and exper.
(aka: btwn group design- diff behavior diff groups)
-W/IN SUBJ DESIGN: diff accross contitions w/in single group participants
(aka: repeated measures design) - type IV
-
-ENVIRO. MANIPULATIONS: exper. conditions modified of participants physical or social envior.
-CONFEDERATES: accomplices of researcher
-INSTRUCTIONAL MANIP: vary IV through verbal instructions participants recieve
-INVASIVE MANIP: create physical changes in participants body through surgery or drugs - Experimental or Control groups
-
-EXPERIMENTAL: recieve IV dosage
-CONTROL: recieve 0 level IV
(NOTE: not always necessary use control) - Assess impact IV
-
-many exper. fail
-IV not manip. sucessfully
-IV not strong enough
-PIOLT TEST: try them out handful of people before actually start exper
-assure level IV enough be detected
-MANIP. CHECK: ? design determine whether IV was manip. sucesfully
-SUBJ. VARIABLES: personal char. of research participants (ex: age/gender/ self esteeme/extraversion - DV
-
-response measured in study
ex:how many words - Assign of participants conditions
-
- test IV assume groups participants roughly equivalent beginning of study
-want to be confident diff. produced by IV
-could be diff. at the start!!! - simple random sample
-
-place participants exper. conditons every participant has = chance being placed in condition
EX: flip coin - Matched random assignment
-
-up similarity among exper. groups
-match var. pretest measure of DV
-put in clusters/ blocks of size
K= #conditions in exper
-answer smae put into cluster w/ others answered the same
-each exper. conditon contains participants whos posess comparable level of memory ability - door in the face effect
- unreasonable offer followed by smaller offer = will accept it
- experimental allows us to test direct hypoth. about _______ of behavior
- causes
-
Experiment:
3 properties -
*research assigns participants to conditions and manipulates at least 1 IV
1. vary at least 1 IV assess effects participants behavior
2. power assign part. various exper. conditions a way assums initial equivalence
3. control extraneus var. may influence participants behavior - Manip. IV
- vary across effects diff. conditions behaviors
- IV
-
- 2 or more levels
*Conditions*
- diff levels IV in experiment
-differ quanity each participant obtains - advantages w/in subj designs
-
-more powerful(ability to detect effects IV)
-participants each condition identical
-eliminate indiv. diff.
-require few participants - disadvantages w/in subj designs
-
-ORDER EFFECTS-each participate recieves all levels IV and order which recieved affect behavior
-Use COUNTERBALANCING-present in diff order to diff participants
-all possible orders used
-may choose smaller subset possible orderings
-LATIN SQUARE- each condition appears once each ordinal position and each precedes and follows every other condition
-CARRYOVER EFFECTS- effects one level of IV still present when another level IV intoduced - experimental control
- eliminationg or holding constant extraneous factors might affect outcome of the study
- systematic variance
-
(btwn group variance)
-part of total variance reflects diff. among experimental groups
-if IV had effect observe systematic dif. scores various exper. conditions
-scores differ systematically btwn conditions systematic variance exixts in scores - treatment variance
- -proportion variance part. scores due to IV (aka:primary variance)
- error variance (w/in groups variance)
-
-result unsystematic diff. among participants
-treat each participant differently introduce randome variability into the data
-does NOT invalidate experiment - an analogy
-
Total = treatment+ confound + error
variance var. var. var.
|__________________| |______|
systematic + unsystem.
variance variance - ideal exper
- up treatment variance/ eliminate confound variance/ down error var.
- internal validity
-
-degree to which researcher draws accurate conclusions effect IV
-eliminate potential sources confound var.
-CONFOUNDING OCCURS: if something other than IV differs in systematic way
-fatal flaw in exper.
-achieved through exper. control - threats to Internal Validity
-
-introduce alternative rival explainations
-no one likely take them seriously -
Threats to Internal Validity:
6 threats -
4.HISTORY:effected by extraneous events occur outside research setting
*event measuring happened to them already that week
5.MATURATION:occurs long span of time create confounds
-dev. maturation = go through age related changes
6.MISC. DESIGN CONFOUNDS:every participant treated same way
-can be controlled for -
experiment expectancies:
deman char. and placebo effects - -affect by beliefs what SHOULD happen in experiment
-
experimenter expectancie effects:
Rosenthal Effect - -researcher expectations about study influence participants reactions
- demand char.
-
-assumptions about nature of a study can also affect outcome (try to figure it out)
-how participants should behave and how researcher expected you to repond
-eliminate this conceal purpose of exper. from participants
-use double blind procedure - placebo effects
-
-psychological or physiological chage occurs as a result of mere suggestion chage will occur
-PLACEBO CONTROL GROUP- administered ineffective treatment - error variance
-
-less fatal problem than confound variance
-seldom eliminated from experiemtnal designs -
sources error variance:
"static" in an exper. -
1.INDIV DIFF:prexisting diff
-use homogeneous sample=>same people
ex: use littermates
2.TRANSIENT STATES:healthy vs. ill
-current moods, attitudes, and physical conditions affect behavior
3.ENVIRO EFFECTS:where study conducted
-external noise distraction
-collect data diff. parts day
4.DIFFERENTIAL TREATMENT:treat exactly same all exper. conditions
-plesant/unpleasant/attractive/M vs. F
5.MEASUREMENT ERROR:cause scores vary unsystematic ways -
exper. control and generalizability:
E's dilemma -
-up exper. controls = artifical situations and highly specific
-more controlled = more difficult is to generalize findings
-EXTERNAL VALIDITY:degree which results obtained one study be replicated or generalized to other samples, research settings, procedures
-generalizability research results to other settings
-E'S DILEMMA:conflict btwn internal and external validity
-choose btwn interna and external validity
-usually chose INTERNAL over external - Brennan et al. (1990)
-
-50 Q about celebs --> 15 undergrads
x = 10.5 elicited tip of tounge
(range = 4-22)
T of T =>1.repeat question 2.picture indiv 3. initials -
Piliguin et al (1995)
"bystander" -
-on NY subway =>elderly gentleman (white) had cane
=>fell/ clapses
1. would ugly birthmark face impact willingness offer assistance =>
2.medical intern same subway car - The Sleeper Effect
-
-passage time up persuasiveness of an idea
ex:low fat diet pizza
Rate: 1-10: they rate 5
-get SE commercial has to be placed in a source mistrust than if released by scientist that company
=>SE artifact memory =>encode article and ads
=>dont give crediability over time
=>article and ad fades away - one-way designs
- -exper. designs which only 1 IV is manipulated
- 2 group exper. design
-
only 2 levels of IV (2 conditions)
-min 2 conditions needed so can compare to other - Assign Participants to Conditions
-
One way designs: randomized groups design/ match-subj/ repeated measures/ w/in subj. designs
-RANDOMIZED GROUPS DESIGN:btwn subj. participants randomly assigned to one of two or more conditions
-MATCHED SUBJ. DESIGNS: participants matched into blocks on basis of a var. researcher believes relevant to experiment then assign to exper or control
-REPEAT MEASURES (W/IN SUBJ) DESIGNS: each partic. serves in all exper. conditions - Posttest and Pretest -Posttest Designs
-
-only way exper. designs above called posttest designs
-POSTTEST DESIGNS:DV is measured ONLY after the exper. manip has occured - Pretest-posttest designs
-
measure DV twice: onece before IV manip and after
ADVANTAGES:
1.Partic. did not diff w/ respet to DV at beginning of exper (efectiveness documented)
2.see exactly how much IV changed partic. behavior (baseline data)
3.more powerful - drawback posttest and pretest designs
-
may sensitize participant to respond to IV differently than would if not pretested
-posttest diff. btwn conditions indicate IV has an effect
-each 3 designs can be used as posttest and pretest-posttest design - factorial nomenclature
-
-use factorial designs to study indiv. and combined effects of 2 or more factors w/ in single exper.
-2 way factorial designs = 2 IV
-3 way (F) designs = 3 IV and so on - factorial designs
-
-written as 2x2/ 3x3 tell how many IV there are and how many levels there are
*2x4 =2 IV = 1 w/ 2 and 1 w/ 4 levels
-2x2x2 = 3IV each var. has 2 levels
a)2x2 b)3x3
c)4x2
a)2x2x2 b)2x2x4
-tell # conditions factorial design has mult. #'s in design (2x2 = 4) - randomized groups factorial designs
- participants assign randomly to one of possible combinations of IV
- matched factorial design
- -involves 1st matching partipants to blocks on basis some var. that corr w/ DV as many partic. in each matched block as there are exper. conditons
- repeated measures factorial design
-
-requires all participants to participate in every exper. condition
-larger designs order and carry over effects problem! - mixed factorial design
-
-combines one or more btwn-subj. var. w/ one or more w/ in subj variable
-ex:visual clif exper. - main effects and interactions
-
factorial designs used examine var scores due to
1.Indiv effects each IV
2.combined or interactive effects IV
3.to error var. - main effect
-
effect single IV in a factorial fesign
-ignore effects other IV - interactions
-
- SIMPLE MAIN EFFECT: effect 1 IV at a particular level of another IV
-show which conditoin means w/ in interatction differ from each other - factorial design w/ 2 IV
-
1.main effect A (ignore B)
2.main effect B (ignore A)
3.interaction A and B - factorial design w/ 3 IV
-
1.examine effects each 3 IV seperately (a,b,c)look indiv efects interactions
2.look at 2 way interactions
-A by B (ignore C), A by C (ignore B)
-tells if IV diff at diff levels of another IV
3.info combined effects all 3 IV
-3 way interaction A by B by C
-only 3/4 IV max used exper. - subj variable
-
age, sex, intelligence, ability, personality and attitudes moderate or qualify the effects of situational forces on behavior
-react diff in diff. situations (everyone) - expericorr (experimental-correlational)
- combine features of an experimental design which IV are manip and features of corr designs in which subj. var. are measured
- uses of mixed designs
-
1.investigate generality of an IV effect
ex:only partipants w/o certian char
2.researchers use expercorr in attempt understnad how certian personal char. relate to behavior under varying conditions
3.split into groups participant var. researchers make participants in exper. condition homogeneous
-down error variance - classify participants into groups
- mixed designs classify into groups on basis of measured participant variable (gender etc.) then randomly assign that group to levels to IV
- median split procedure
-
researcher identifies median participants scores var. of interest then classifies participants w/ scores below median low on var and sores about median as up on variable
If these used:
-lead to bias results
-can be problematic - cautions in interp results of mixed designs
-
-can not include "causes"
-subj. var. measured rather than manip. - moderator variable
- a var. that qualifies or moderates the effects of another var. on behavior
- quasi-experimental designs
-
(not true exper)
-lacks random assignment participants to conditions/compares people in groups already exist or w/in single group participants before and after some event has occured - quasi independent variable
-
used to indicate the var is not a true IV manip. by the researcher rather is an event occured for other reasons
-threats internal validity present
-quality quasi exper. depends on how many threats there are to internal validity it sucessfully eliminates - one group pretest-posttest design
-
a preexpermental design which a group of participants is tested before and after quasi indep. var. has occured fails to control for all threats internal validity
DO NOT USE!
-effects not looked at:maturation effects, history effects, testing effects -
effect of one group pretest
postest design:
regression to the mean - tendency for extreme scores in distribution move, or regress toward the mean of the distribution w/ repeat testing
- preexperimental design
- lacks necessary controls to minimize threats to internal validity/ not involve adequate control or comparison groups
-
nonequivalent control
group design -
-looks for one or more groups participants appear to be reasonably similar to the group that recieved quasi IV
-2 var = posttest and 1 pretest and posttest -
nonequivalent groups posttest
only design -
-measure both groups after one of them recieved quasi-exper. treatment
-before = baseline
-not eliminate all threats to internal validity -
local history affects and
selection - by -history interaction -
1. something happen to one group doenst happen to another
2. "history" effect occurs in one group but not in the other - time series designs
- -measure DV on several occasions before and one several occasions afer quasi- IV occurs
-
simple interrupted
time series design -
taking several pretest measures before introducing IV (or quasi IV) then take several posttest measures aferwards
(01,02,03,04 * 05,06,07,08)
-interuption quasi IV - contemporary history
- not rule out that observed effects where due to another event occured at same time as quasi IV
- interrupted time seires w/ a reversal
-
-observe participants behaivor when quasi-IV or treatment introduced then removed
(01,02,03,04 x 05,06,07,08
-x 09,010,011,012) - interuruped time series design w/ mult. replications
-
-reintroduce IV observe effects and remove second time
(01,02,03 x 04,05,06 -x 07,08,09 x
010,011,012 -x 013,014,015) - interruped time series design w/ mult replications: LIMITATIONS
-
1. research has not power remove IV (seat belt laws)
2. effects some quasi IV remain even after var. itself is removed
3. removal quasi IV may produce chages not due to effects of var. per se -
Control Group Interrupted
Time Series Design -
-perform analysis on group recieved the quasi IV and on nonequivalent control group did not recieve the quasi IV
-helps rule out certian history effects - longitudinal desings
-
the quasi IV is time itself
01,02,03,04,05
GOAL: uncover dev. changes occures function of age but something other dev. has produced observed changes -
longitudinal designs:
drawbacks -
1.difficult obtain samples participants affree be studies against again over long period of time
2. trouble keeping track of participants many move and may die
3. requires great deal time, effort, and money - longitudinal designs
-
-examine how indiv. participants change w/ age
-important effects of time and aging on dev. - cross-sectional designs
-
compare groups diff. ages at single pt in time
-age related changes
DRAWBACK:gernerational effect- people diff ages differ in agge perse but also conditions under which their generation grew up
-track same age diff. yrs. in future - eval. quasi exper. designs
- -see if vary covary -> corr. and ANOVA
-
up confidence in quasi-exper.
results -
-"patch" up basic designs to provide most meaningful and convincing data possible
-more IV better -
Problem:
Multi tests inflate
Type 1 error -
-type 1 error up when perform greater # of tests
-more likely draw invalid conclusions about effects IV
-problem make type 1 error
-use (1 - (1-alpha)c
-c = # of tests - Bonferroni adjustment
-
divide desidred alpha level by # of tests plan to conduct
DRAWBACK: problem type 1 error down problem type 2 error up - rationale behind ANOVA
-
-F-TEST: ratio variance among conditions (btwn groups) to variance w/in conditions (w/in groups)
-larger the better
-test to see if estm diff. btwn condition means due to error variance - one way design w/ single IV
-
ANOVA breaks total var into 2 components
1. systematic variance
2. error variance - total sum of squares
-
-SUM OF SQUARE: reflets total amount of variance in set of data
-TOTAL SUM OF SQUARE: caluculate
1. subtract mean each score
2. square diff
3. add them up
= total amount of variability in set of data - sum of squares w/in groups
-
-sum of variances of the scores w/ in particular experimental conditions
-express variability in data NOT due to IV
-or = error variance - means square w/in groups
-
-divide w/ in groups variance by w/ in groups df obtain quantity
-provides estm. ave. w/in groups or error variance - sum of sq. btwn. groups
-
-var set scores assoc. w/ IV; sum sq. diff btwn each condition mean and grand mean
-estm. systematic var
1. subtrack grand mean each group means (small means dont differ much or IV NO EFFECT)
2. sq. diff
3. mult. each sq diff by size group
4. sum across groups
5. divide K -1 (k = # of groups)
6. divide by (k-1) = mean sq. btwn groups - our estm systemaic or btwn groups variance - F-test
-
-calculate ratio btwn groups variability to w/in groups variablitiy each effect testing
-IV no effect = numerator and denominator are estm. same
-IV effect- numerator larger than denominator
-value F exceeds critical value F on table at least 1 condition means differ orthers = IV has effect - follow up tests
- 1. calculate means signif. effects (a leave b out and vice versa)
- main effects (ME)
-
-2 means differ systematically, inspect means find out direction and magnitude
-signif. ME indicates diff. exixts btwn 2 of 3 conditions means does not indicate which means differ from which -
follow up tests
(aka: post hoc test or
mult comparisons) -
-identify which means differ signif
aka:LSD test, Tukey's test, Scheffe's test
-done only if F test is significant -
btwn subj. and w/in subj
ANOVA's:
Multi analysis of variance - MANOVA: test diff btwn means of 2 or more conditions 2 ore more DV simultaneously
- conceptually related DV
-
-several DV all tap some construct
-combines info 10 DV new composit var. then analyse whether participants scores new composit var diff among exper. groups - Inflations of Type 1 error
- -MANOVA contols this
- How MANOVA WORKS
-
1. reate new composit var. weighted sum of orginal DV
-CANONICAL VAR: composite var. is calculated by summing 2 or more DV have been weighted accoring to ability to differentiate among groups of participants
-produce single index var. of interest
2. multivariage version F-test is performed to determine if partic. scores canonical var. differ among exper. condtitions
3. now can perform ANOVA w/ out talk of Type 1 Error if significant