KHA350 Lecture Notes - Lecture 5: Lincoln Near-Earth Asteroid Research, Latin Square, Sphericity
25 Jun 2018
School
Department
Course
Professor
Research Methods week 5: Repeated measures and experimental designs
- Interaction: tells you about the consistency of one IV on the DV across the
second IV
Repeated Measures:
- Between subjects: all the IVs have different people in each of the levels
osome IVs that are always between subjects: eg. Sex
- Within subjects designs:
oWhere the IV involved have the same people providing data for each
of the levels
oOften referred to as repeated measures designs
oExtracting the IVs: need to identify in addition, whether design is
between or repeated measures
Substantial implications for setting up data sheet
And how you run the analysis
oFactorial design: can have combination of repeated measures and
between subjects design
All of one design, or a mix of them
oThe same subjects receive different treatments but the same DV is
measured for each
oA subject receives a number of different tests normed to have the same
scale
Eg. The WAIS subtests are all on the same scale (intelligence
scale)
Same participant providing data from multiple different tasks,
that have the same scale
oThe same variable is measured over time:
Eg. Depression pre-treatment, depression post-treatment
oMultiple DVs are measured on the same people:
But they are on different scales
Anxiety assessed using paper and psychophysiological tests
after treatment
Eg. Heart rate, self reported scales, cortisol secretion
Put on the same metric via MANOVA
This is a special case of within subjects design. Which we will
discuss in future weeks
oAdvantages of repeated measures:
Economy of subject numbers
Within subjects design requires the same people come
back
Each subject acts as their own control:
The power of the technique
This reduces error variance: making the test more
sensitive
A smaller difference between means will be sufficient
to produce a significant F ratio
oDisadvantages of repeated measures:
find more resources at oneclass.com
find more resources at oneclass.com
Order effects
Due to such factors as learning or fatigue, habituation,
sensitization to conditions, adaptation
oEg. Adaption to the noise in the room, lesser
effect over time of the white noise in the
background
oEg. Sensitization: eliciting fear through images
or readings; after seeing a number of them you
get use to it; reach the point where
oEg. Habituation; getting use to noises etc.
This example was exaggerated but the impact of these
issues need to be considered
Can compensate for these though counterbalancing:
smearing the effects of learning and fatigue equally
across conditions
Learning and fatigue are most influential and need
greatest consideration
oEg. Reaction time as DV over IV of trials
oInitial quick improvement in performance:
learning
oA later decline in performance: fatigue
oMakes it look like the IV is impacting the DV,
when actually it is likely to be these effects
oNeed to account for this: each part of you IV is
presented equally in each block: eg. Smearing
through initial stages of learning, middle phases,
and fatigue later phases
Differential carryover effects: when the difference of going to
level one to level two, is not the same as going from level two
to level one
Eg. Memory experiment
Baseline condition: learn list of words however you
want, other condition is new technique
oWill see much improvement between trials
If counterbalanced so that level two is done first, they
are not going to go back to other older technique
Introduced noise and contaminated effect of treatment
You cannot get around this: go to between subjects
design
Need to be sensitive to even subtle effects of differential
carryover
oRationale for within subjects design:
Big difference between participants: the higher the score the
better you are at the test
Big differences between participant scores, with small
differences of the treatments
Differences between means are swamped by the difference
between subjects
find more resources at oneclass.com
find more resources at oneclass.com
We can separate out the subject differences using repeated
measures ANOVA
We have information about each subjects:
Total variation:
oVariation due to treatments (treatment main
effect)
oVariation due to noise (among participants in the
groups, measurement error etc.)
TOTAL VARIATION:
oVariation between participants
oVariation within participants
Variation due to treatments: treatment
main effects
Variation due to noise
We can reduce unexplained variance by pulling out
variation between participants (one was hungover and
other were nerds)
oEasier to identify an effect as being statistically
significant
oThe thing we are comparing with is smaller
oRepeated measures ANOVA becomes more
powerful at identifying effects that are
significant
Another assumption: Sphericity:
Only relevant for repeated measures ANOVA (not
between subjects)
Because when the treatments are independent, then this
is always met: but now the treatments are no longer
independent
Mathematically, the assumption is that there is
compound symmetry of the covariance matrix:
oData of groups needs to meet: homogeneity of
variances of variance
oHomogeneity of covariances among treatments:
variances of the differences between treatment
oThis is always met when there are only two
levels of your within subjects independent
variable
oBut is usually violated when there are more than
two levels
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
Research methods week 5: repeated measures and experimental designs. Interaction: tells you about the consistency of one iv on the dv across the second iv. Between subjects: all the ivs have different people in each of the levels: some ivs that are always between subjects: eg. sex. Substantial implications for setting up data sheet. And how you run the analysis: factorial design: can have combination of repeated measures and between subjects design. All of one design, or a mix of them: the same subjects receive different treatments but the same dv is measured for each, a subject receives a number of different tests normed to have the same scale. The wais subtests are all on the same scale (intelligence scale) Same participant providing data from multiple different tasks, that have the same scale: the same variable is measured over time: Depression pre-treatment, depression post-treatment: multiple dvs are measured on the same people:
