Study Guides (390,000)
CA (150,000)
U of G (8,000)
PSYC (1,000)

PSYC 2360 Study Guide - Statistical Conclusion Validity, Null Hypothesis, Internal Validity


Department
Psychology
Course Code
PSYC 2360
Professor
Naseem Al- Aidroos

Page:
of 7
Things to Remember - Methods Final Exam
Two Factor Designs
when we have 2 independent variables each with 2+ levels
we calculate an F statistic like in one-way ANOVAs where
o 
 , if the null hypothesis is rejected
that means that F > 1 and there is a significant difference
between the groups larger than what we would expect if it
was due to chance alone. If the null hypothesis is not
rejected, F = 1 therefore there was not a big enough
difference between the groups to be considerably different
than what we would expect if due to chance
marginal means are when we calculate the mean of the entire row
or column
main effects are when we look to see if the marginal means are the
same or different for a given factor
o if there is a significant main effect (the marginal means for
the factor’s levels are different) we can say that that factor
has a significant effect on the dependent variable, ignoring
the effects of the other factor
an interaction is when the effects of one factor on the dependent
variable depends on the level of the other factor (when the lines
intersect)
o to see if we have an interaction we do decomposition and
see if we can get back the cell means by only knowing the
marginal means and the grand mean
a) get the grand mean (mean of all cells)
b) subtract grand mean from each cell
c) get the new marginal mean of the cells
d) add the marginal means and the grand mean and see if
you get the cell means back - if you do, there is no
interaction - if you do not, the leftovers are the
interaction
o when we have a significant interaction it just tells us there is
a difference between the means but it does not tell us the
pattern of the means, therefore we use simple main effects
simple main effects are when we look at the effect of a factor on
the dependent variable across a level of the other factor
o when we have a significant interaction, the main effects
should be interpreted with caution because the effect of one
factor DEPENDS on the level of the other
if we were to do a three way interaction (3 independent variables)
it gets a lot more complex
o 1 three way interaction
o 3 two way interactions
o 3 main effects
many repeated measures designs are factorial and have the need
for counterbalancing
o e.g. Latin square design (each condition appears in each
order and equally follows each other condition)
when you have more than 2 levels in the factor, a simple main
effect will not tell you where the differences are within the means
therefore we need to conduct mean comparisons
o pairwise comparisons - we compare any mean with any other
mean, causes type 1 error probability to increase
o planned or priori comparisons - we compare two means that
we had expected would be different
o posthoc comparisons - we only do comparisons if we have a
significant effect
o complex comparisons - we compare more than 2 means at a
time
Internal Validity
there are a few threats to validity
o construct validity - is the test measuring what it is supposed
to measure
o statistical conclusion validity - is our conclusion valid
o internal validity - the extent to which we can say that our
independent variable is causing the dependent variable
o external validity - the extent to which our findings generalize
experimental control is when we try to make sure and control as
many variables as we can - the greater experimental control, the
more internally valid
o we can have extraneous variables that noise the data and
cause random error however we can still have an experiment
that is internally valid with extraneous variables
o we can also have confounding variables in which we cannot
say that our test is internally valid because confounding
variables always produce an alternative pathway and we have
no way of knowing which pathway is causing the change in
the dependent variable
since people are different, there are different designs we can do in
order to control for as many people differences as we can
o limited populations design - similar to convenience sampling
where we sample from population like university in which
subjects will be similar to each other (homogeneous) in a
variety of factors, this increases our statistical power by
decreasing the effects of extraneous variables
o before and after designs - are similar to repeated measures
designs except the group is only in one condition and the
dependent variable is measured twice - once before the
manipulation is presented and once after the manipulation
and participants serve as their own controls, this still has the
problems of retesting effects such as fatigue, practice,
guessing the research hypothesis
o matched groups designs are when the sample are measured
on a variable and then based on these scores, are put into
groups, this works as long as the variable they are being
scored on is related to the dependent variable, often this is
not necessary and random assignment is enough, this
reduces the differences between the groups and increases
the power of the test