Wk 5- One Way ANOVA and Beyond to Factorial Design & Interactions

what are two ways to follow up when we know means are different but not sure which groups?

Post hoc and planned comparisons

which follow up test you know what to compare what to

planned comparisons (priori?) (pairwise?)

•: Specific contrasts of interests decided on apriori to test specific predictions

which follow test you dont know to compare

Post Hoc

we have no specific hypothesis about where differences might lie so we look at the comparisons between all possible means  

Type 1 error rate increases when?

increase in planned comparisons

what is Type 1 error? and how much percent chance to we having of making it for every tests we do?

rejecting the null hypothesis when it is in fact true

5% chance

when/where do you control for type 1 error and how much of an adjustment made for each, in terms of alpha?

in planned (priori) comparisons (small adjustment because you have limited and known comparions)

and/or post hoc comparisons (more substantial adjustments because doing heaps of comparisons)

what is bonferroni correction (aka dunns procedure)

dividing alpha value by number of comparisons done

•If the number of planned comparisons exceeds df_effect, use a Bonferroni adjustment.

•The Bonferroni adjustment maintains  at .05 for the set of planned comparisons by “dividing up” the .05 error rate among the comparisons. 

•For 3 comparisons,  = .05 ÷ 3 = .0167;   For 4 comparisons,  = .05 ÷ 4 = .0125;  etc.

when is it suggested by some people (and who) that no adjustment is needed?

•Keppel & Wickens (2004) suggest that if the number of planned comparisons doesn’t exceed df_effect (i.e., no more than  a-1 planned comparisons among a group means), there’s no need to adjust the a-level. Use  = .05.

 

when can planned comparisons be carried out in terms of the results of the overall F significance

•Planned comparisons can be carried out with or without a significant overall F.

what is the LSD option in spss

is fishers least nifty test and basically equates to making no adjustment, for the type 1 error

what are complex comparisons

comparing one GROUP to another GROUP instead of normal sleep placebo to sleep deprived caffeine it would be CAFFIENE vs PLACEBO

what do you look at when you run bootstrapping?

not looking at F value, looking at 95% confidence intervals and making sure the span from the lower to upper one doesnt include zero, where it does not span zero can say you are 95% confident that its a non-zero difference between those groups (yes significant, ones that do have zero should be not significant and should reflect that in the comparisons)