LECTURE 8 non para stats Flashcards
what are contasts
when want to further investigate a significant main effect found in an ANOVA
what can you do to further investigate a main effect
multiple t test (type 1 error)
planned ( a priori) comparison contrast
unplanned (post hoc) comparison contrast
simple main effects
what is a planned (a priori) contasts
do not require significant anova
carry out small no of specific comparisons that are driven by a hypothesis
variance caused by experiement due to subjects assigned to diff groups - can break down to test hyp about which groups differ
what is an unplanned (post hoc) contrast
no planned hypothesis that requires significant anova to carry out several comparison simultaneously
designed to prevent type 1 error by keeping significant level at 0.5
what is simple main effects comparison
in 2 ay anova test for effect of one factor at each level of the other factor
rules when chosing contrasts
contrasts must be independent of eachother (test unique hyp)
only do two contrasts of variance at once
k-1 - 1 less contrast than no of goups
(THEREFORE if compare first two cond and not last, 1 = 1,, 2 = -1 and 3=0) - add to 0
formula for planned comparison
C = C1M1 + C2M2…etc
C = weighting of coefficient - sum to 0
therfore test whether m1 sig diff from m2
spss contrast : Deviation
each condition compared to overall mean except 1st and last category which is reference mean
spss contrast: simple
mean of each condition contrasted with first or last cat
spss contrast: helmert
mean of each cond except last compared with mean of subsequent cond
spss contrast: repeated
mean of each cond compared to mean of next
spss contrast: polynomial
means contrasted in terms of fitting to a trend ie linear, quadratic etc
Post hoc test for repeated measures
bonferonni
divide significance level by number of tests
run t test
compare sig level in spss to new sig level
#’bonferonni correction’
independent measures post hoc
bonferonni
tukey
dunnett
REGWQ
describe tukey post hoc test
pairwise comparisons between condition means when no prior expectation of when differences equivalent to t tst - differences lie in SE
general error term from diff between mean (MSerror)
mean 1 sample - mean 2 sample / SQRT of MSerror/n