Please enable JavaScript.
Coggle requires JavaScript to display documents.
Module 1 (Module 1 (Contrast analyses (Orthogonal (statistically…
Module 1
Module 1
Multiple testing
Adjustments
LSD
No adjustment?
Bonferroni
Devide alpha by nr of tests
Tukey's HSD
best correction
Reduces power
Contrast analyses
alternative of multiple testing
contrast coefficients
1,1,-1,-1
Orthogonal
statistically independent
sum of product of coefficients=0
Design
Experiment
Variables
Independent
Factors
categorical
leads to groups
Often multiple
Dependent
single
Outcome
Can't change independent
Test
ANOVA
Groups
Within subjects
Between subjects
Observational
Test
Regression
Variables
Independent
Many
quantitative
General linear model
ANOVA
Assumptions
quantative dependent variable
Independent groups
Random samples
normal population
robust if N>15 and no skewness
Equal population variance
Robust
if equal sample sizes and small difference
not robus
if largest SD > 2*smalles ST
Adjustments
Data transformation
Adjusting alpha
Welch procedure
Non-parametric test (Kruskal-Wallis)
Distribution
Skewness
confidence interval
+-2*SE=0
-=left
+=right
Kurtosis
multi/unimodal?
Peakedness
Confidence interval
+-2*SE=0
Outliers
Range of values
Central tendency
Mean
median
most common value
Standard deviation
Two way
Design
Balanced
same relative fequency distribution in groups
statistically independent
Effects (main) can be estimated independently
Unbalanced
Statistically associated
Main/interaction effects cannot be associated independently
non-orthogonal
correction
Type III sum of squares
only non overlapping parts
Hierarchical SS
How important is an effect
Most important gets all the variance
Steps
delete nonsignificant interaction
Look at main effects