Transcript Lecture15 - University of Idaho

PSYC512: Research Methods
Lecture 15
Brian P. Dyre
University of Idaho
PSYC512: Research Methods
Lecture 15 Outline


Review of Lecture 14
 Multi-factor research designs
Today: other specialized research designs
 Statistical “nuts and bolts” in conducting
hypothesis tests of means
 Combining experimental and correlational
designs (Analysis of covariance or ANCOVA)
 Quasi-Experimental Designs
 Developmental Designs
 Small-n designs and Psychophysical Methods
PSYC512: Research Methods
Statistical “nuts and bolts” in
testing hypotheses about means

The central limit theorem
 The mean of the sampling distribution of means is equal to the
mean of the population from which the samples were drawn.
 The variance of the sampling distribution of means is equal to
the variance of the population from which the samples were
drawn divided by the size of the samples. Expressed as the
standard deviation:
X  / N

If the original population is distributed normally (i.e. it is bell
shaped), the sampling distribution of means will also be
normal. If the original population is not normally distributed,
the sampling distribution of means will increasingly
approximate a normal distribution as sample size increases.
(i.e. when increasingly large samples are drawn)
PSYC512: Research Methods
Does a sample mean significantly
differ from a population mean?

 of population is known
 z test of sample mean
z
X 
zX 

 of population is estimated as
s




t-test of sample mean
As N approaches infinity, t
approaches z
For small N distribution of
sample variance is skewed
Df = N-1

X 
X

X 
/ N
X 
s
X  X 
tX 

sX
s/ N
t
PSYC512: Research Methods
The t-test

Differences in means of matched samples

Differences in means of Independent samples
PSYC512: Research Methods
Specialized Research Designs





Combining between-subjects and within-subjects factors in
research design – mixed designs
Combining experimental and correlational designs
(Analysis of covariance or ANCOVA)
Quasi-Experimental Designs
Pretest-posttest designs
Developmental designs (Longitudinal or cross-sectional)
PSYC512: Research Methods
Combining Experimental and
Correlational Designs

Covariates in experimental designs
 Measure your subjects on a covariate—a variable that you believe
may be correlated with your dependent variable
 If left unmeasured these covariates add error variance and might
obscure significant effects
 Measuring the covariate allows you to use correlational statistical
techniques in your analysis (e.g., Analysis of Covariance or
ANCOVA) to “subtract out” the error variance associated with the
covariate, thereby increasing the statistical power of your
experiment
 Example: measuring IQ in a learning experiment
PSYC512: Research Methods
Combining Experimental and
Correlational Designs

Quasi-independent variable in experimental designs
 “Quasi” means “kind of, but not really”
 Similar to including a covariate, except
 measurement of covariate is used to assign Ss to
groups
 Covariate is thus treated as an quasi-independent
variable
 Quasi-independent variables are referred to as “quasi”
because they cannot be manipulated, they are
essentially dependent variables (measures) that are
treated as independent variables in the experimental
design and analysis
PSYC512: Research Methods
Quasi-experimental Designs


Quasi-experimental designs are those in which only quasiindependent variables are used
Time series vs. pretest-posttest designs
 Time series: Measure behavior several times prior to
and following a treatment (time series design) or
change in your quasi-independent variable (interrupted
time series design)
 Pretest-posttest: Measure behavior once prior to and
once following the change in your independent variable
PSYC512: Research Methods
Quasi-experimental Designs


Equivalent time samples design
 Time-series design especially useful for treatments with
transient effects
 Repeatedly measure behavior following multiple
applications and withdrawals of the treatment
Non-equivalent control group designs – helps control for
history confounds which should affect both groups equally
PSYC512: Research Methods
Developmental Designs




Used to assess changes in behavior related to a person’s chronological
age, which serves as a quasi-independent variable
Cross-sectional designs
 Simultaneously test subjects assigned to two or more age groups
 Generational effects can confound the age variable
Longitudinal designs
 Repeatedly test a single group of subjects over time
 Controls for generational effects—but, may still limit external validity
 May be confounded by history, mortality, and/or multiple observation
effects
Cohort-sequential design
 Combines longitudinal and cross-sectional designs by measuring
multiple age groups over time which allows evaluation of
generational or historical confounds
PSYC512: Research Methods
Single Subject Research Designs


Research that focuses on identifying functional
relationships between variables and performance of
a single subject (e.g., behavioral analysis and
psychophysics)
Typically involve
 Large number of observations
 Rigid experimental control
 Investigations of powerful variables whose effects
are easily detected
PSYC512: Research Methods
Baseline Designs

Same as time-series design


Time series: Measure behavior
several times prior to and
following a treatment
Two phases (A and B)


A: baseline phase to establish
behavioral baseline
performance on DV prior to
treatment, requires that a
stability criterion be reached
B: intervention phase that
measures performance on DV
after treatment
PSYC512: Research Methods
Baseline Reversal Designs




Problem: Time confound
Solution: ABA Design—
reverse the treatment by
removing it and see if
performance returns to
baseline (unlikely to occur
by coincidence)
Problem with reversal:
now behavior is at
baseline again
Solution: ABAB design
A
PSYC512: Research Methods
B
A
B
Multiple Baseline Designs



Used to assess irreversible
changes in behavior
Assess multiple independent
behaviors and introduce
treatment to only one
behavior at a time
Controls for time effects
(history, maturation)
PSYC512: Research Methods
Discrete Trials Designs
(Psychophysical Techniques)



Used to determine thresholds and difference
thresholds (just-noticeable differences or JNDs)
Ss receive dozens or hundreds of trials under tightly
controlled conditions
Methods
 Method of Constant Stimuli
 Method of Adjustment
 Method of Limits
PSYC512: Research Methods
Classical Psychophysical Methods
Fechner’s Elements

(1860)
Absolute threshold
(limen): how much
energy must exist in
a stimulus for it to
be detectable?
1.01
Probability of
Detection
of Psychophysics
Predicted by
absolute threshold
Probability of Detection

P abs( 50 )
Distribution
of “absolute”
threshold
“Real World”
P real( 50 , 10 )
10. P T ( 50 , 10 ) 0.5
0.5
0
20
40
60
80
E
Amount of Stimulus Energy
Amount of Stimulus Energy
Subliminal: below
threshold
Superliminal:
above threshold
PSYC512: Research Methods
100
Methods for Determining Thresholds

Method of adjustment
 Intensity or feature of stimulus is
changed until it matches a standard
 Hysteresis requires both ascending and
descending trials
 Average match across ascending and
descending determines threshold
 Fast, but least accurate
PSYC512: Research Methods
Methods for Determining Thresholds

Method of Limits (discrete method of
adjustment)
 Like method of adjustment except
adjustment is done in discrete steps whose
size is controlled by the experimenter
 Hysteresis requires both ascending and
descending trials
 Average match across ascending and
descending determines threshold
 Variant: Staircase method
PSYC512: Research Methods
Methods for Determining Thresholds
1.01
Probability of Detection
Method of Constant Stimuli
 Choose 5-9 stimuli,
some above, some
below threshold
 Present in random
order
 threshold equals
amount of stimulus
energy that detected
50% of the time
 Slowest but most
accurate
Probability of
Detection

P abs( 50 )
P real( 50 , 10 )
10. P T ( 50 , 10 ) 0.5
0.5
Threshold
0
20
40
60
80
E
Amount of Stimulus Energy
Amount of Stimulus Energy
For all Methods:
sensitivity = 1/threshold
PSYC512: Research Methods
100
Next Time…

On-campus (“Live”) students: please e-mail me
article presentation time slot preferences
PSYC512: Research Methods