KHA350 Lecture Notes - Lecture 6: Time Point, Repeated Measures Design, Analysis Of Variance
25 Jun 2018
School
Department
Course
Professor
Research Methods week 6: Factorial repeated measures designs
Why there are changes in the way ANOVA tables are structured when you have
different kinds of designs
-Mixture of designs
Different experimental designs:
-between subjects IV’s
owhen different participants
-Within subjects IV’s:
oSame participants
-When there are two or more IVs then there are three main experimental design
possibilities
oDepending on the nature of the IVs in the mix
-Between:
oIf all IVs are between
-Completely within subjects design:
oRepeated emasures
oBy definition, if fully within, then the same participants will provide
data for all levels of all IVs
-Mixed design: mixed factorial or split plot design
oCombination of different types of IVs
oSome are between some are within
oReason this is pointed out: each type requires a different SPSS set up
and analysis running, output looks different and error terms will also
be different
-SPSS:
oFixed factors: told previously to be IVs
IVs when the levels of the factor are chosen by experimenter
If replicated, these levels will be exactly the same
Specifically chosen levels
find more resources at oneclass.com
find more resources at oneclass.com
Fixed and unchangeable
Cannot generalise beyond these levels of the study: we don’t
know what will happen
oDependent variables
oRandom factors: now we may nee to use these
Levels are randomly selected from a larger population of
potential levels
Differ from one replication to the next
Situations where you don’t have control over the nature of the
variation that is going in to the IV
If you manipulated experiment, the levels would change
Eg. Subjects: when replicated in repeated measures designs
Changes how you calculate the error term in the ANOVA: for
each effect and interaction
Remember that the error term is the denominator when
we calculate the F ratio
-Repeated measures ANOVA generally has larger error term?
oFor one way, within subjects design, what we are doing is that we are
partitioning the individual variability out:
Where PI is a measure of variability due to a particular person,
representing how much that person differs from the average
person
Two sources of nice:
From the interaction of unexplained stuff and the
condition they are in
Error term: combination of interaction of the individual
person and the level they are in, and the unexplained
stuff
Interaction between participants and the IV
FIXED FACTOR: VARIATION DUE TO TREATMENTS
(TREATMENT MAIN EFFECT)
RANDOM FACTOR: variation between participants
find more resources at oneclass.com
find more resources at oneclass.com
oIf you want to get a sense of what is going on, watch the degrees of
freedom
Between subjects
And within subjects
Error term: multiplication of subjects and treatment DFs
Calculate error term for within as if it is an interaction between
the participants and the IV
oLast weeks example:
Overall DF is just number of levels x number of subjects
Error term is: (5-1)(9-1)
oCarryover effect when you get to factorial design
Now have situation where each main effect and interaction is
tested against an error term that is the interaction between
participants and IV
Practical upshot: when completely within subjects, have
a different error term for each of the main effects and
interaction
Between subjects: same error term for main effects and
interactions
-Other issues of factorial repeated measures designs:
oThe assumption of spherity still applies
More than two levels of any of your within subjects factors
then it is likely that you’ve violated the assumption of spherity
Corrections for violations are the same as one way
EPSILON
oFollow up significant interaction with tests of simple effects:
Much the same as for between subjects designs
Examining the main effect of one factors at each separate level
of the second factor
Just do these with one way repeated measures rather than trying
to use a pooled error term
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
Research methods week 6: factorial repeated measures designs. Why there are changes in the way anova tables are structured when you have different kinds of designs. Different experimental designs: between subjects iv"s: when different participants. When there are two or more ivs then there are three main experimental design possibilities: depending on the nature of the ivs in the mix. Completely within subjects design: repeated emasures, by definition, if fully within, then the same participants will provide data for all levels of all ivs. Spss: fixed factors: told previously to be ivs. Ivs when the levels of the factor are chosen by experimenter. If replicated, these levels will be exactly the same. Cannot generalise beyond these levels of the study: we don"t know what will happen: dependent variables, random factors: now we may nee to use these. Levels are randomly selected from a larger population of potential levels. Differ from one replication to the next.
