Higherarchial anova Flashcards

1
Q

What is a factorial anova DESIGN

A

one thing they all share - all levels of one independent variable appear at all levels of the other independent variables; applies to both fixed and random effects ANOVA

what we refer to when factors are crossed.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

what is a hierachrial/nested ANOVA design

A

A design where each level of one of your factor/variable appears only at a single level of another; factors are nested

  • A design in which factors are hierarchically built up.
  • or one factor that’s nested within another.
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

provide an example of a nested factorial design

A

One method implemented in 3 schools. means the factor school is nested within the factor method.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

goog critique for paper?

A

maybe you read it and think oh this should have been analyzed as a nested design but it was analysed as nested?

the problem of this is that if you treat is as a factorial design people will get interaction effects which is meaningless. and secondly, it doesn’t take into account this confound of everything not being combined with everything else. get a bias ANOVA results.

anova might tell. yout heres something significant even though there might not be

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

how many levels down max should you go with a nested ANOA desgin

A

2 levels should be the limit

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

what is the logic of ANOVA?

A

that we are comparing variability(means squared) due to some experimental manipulation with variability(means squared) due to error

this is the basis of the f ratio

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

why is the f ratio adjusted in random factor ANOVA designs?

A

to account for the added variability we get from having random factors

(using interaction terms as error terms)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

nested anova and f ratio - what 2 issues do we now have to consider?

A
  1. each of the nested factors is only combined with certain levels of other factors - don’t have a total combination of everything with everything.
    * means we do not have interactions, doesn’t make sense to assess the significance of interaction terms
  2. in nested designs the difference between the two kings at the top of each level is confounded with the difference between king one’s assistant factors on one hand and the difference between king two’s assistant factors on the other.

We have to consider both of these while doing the analysis

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

factor ___ is nested within factor ___

A

story number is nested within story type

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

story example. Remember the first problem to consider with nested designs:

  1. each of the nested factors is only combined with certain levels of other factors - don’t have a total combination of everything with everything. Mean’s we do not have interactions doesn’t make sense to assess the significance of interaction terms

how do we analyse the data, taking this problem into account?

A
  1. We compute the main effects for each story type
  2. compute the effect of story number nested within story type

(2) Not to be confused with an interaction effect. It simply tells us whether there is a difference across exotic stories, and separately, if there is a difference across familiar stories

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

story number example. Remember one of the issues we need to consider with nested designs

  1. in this example, the effect of story type is confounded by differences among stories 1-3 and stories 3-6.

For stories 1-3 the third story’s mean is higher than the first two. we need to now if this is because the story is exotic or one story (regardless of exotic or not) just more memorable than the others

how can we tease this apart in analysis?

A

we can consider this confound in our statistical analysis by TREATING NESTED FACTORS (the story numbers) AS RANDOM FACTORS

takes care of the counfound between the top level variable and second level variable (in this example)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

so in nested ANOVA we treat the nested factors as random. what about the top level factors? what about for this specific example?

A

Need to decide whether to treat them as fixed or random

  • based on experimental design and your goal in terms of how you want. to interpret the results/generalise

for story example

  • If the levels of story type – chosen randomly
  • And goal is to draw general conclusions about familiarity – then story type should be considered random

OR

  • If levels for story type are chosen purposely
  • And all you are interested in is familiar vs exotic
  • Then treat it as fixed

In this example story type is considered random

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

when should you check for homogeneity of variance?

A

always! whether you have fixed effects or random effects you always have to check!

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

what does the levene test tell us and do we want it to be significant?

A
  • insignificant result tells us the variance or error variance is homogeneous or equivalent across all conditions.
  • If however, the levene term is significant it tells us the variance or error variance in at least one of or conditions differ significantly with at least one of the other condition.
  • This would be a problem.
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

output of the nested ANOVA table - how could you interpret the last source results conceptually?

A

the results of story number nested within story type

And for this example both of these effects are significant. Based on this we would reject the null hypothesis. Shows there is significant difference across the conditions.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

The ANOVA told us there was a significant difference across conditions. Using this and the plot explain what was gwarning?

A
  • exotic stories are recalled significantly better than familiar stories
  • there is significant differences in recall WITHIN stories 1-3 and WITHIN 4-6. this is a significant effect of story number nested within story type
17
Q

if the ANOVA has told us there was a significant nested effect of story number WITHIN story type how could we follow this up?

A

Whenever you have a significant nested effect you break up the analysis in a follow up the analysis into something simpler - e.g.,

  • 2 one way anovas - one comparing stories 1-3 and the other comparing stories 4-6
  • or post hoc t-tests
18
Q

why would it be diabolical to compute a large one-way ANOVA of stories 1-6 in a follow-up analysis inspecting why stories 1-3 differed significantly between each other and why stories 4-6 differed significantly between each other

A

because you wouldn’t know if any differences picked up are due to story number or story type

19
Q

What is the point of Scheffe and Games Howell tests?

A

These are both post hoc tests.

  • scheffe
  • Games Howell - doesn’t assume homogeneity of variance (equal variance) - (use this if the leverne test is significant (bad))
20
Q

what is homogeneity of variance mean?

A

the population variances (i.e., the distribution, or “spread,” of scores around the mean) of two or more samples are considered equal.

21
Q

Written report of hierarchial/nested ANOVA. What do we include?

A
22
Q

What if our overal nested ANOVA analyses gives us the option to run post hoc tests, should we use this?

A

NO! only use post hoc tests after you have simplified/broken down the analysis

if you do it during the overall analysis you have the confound of the nested variable confounded with the top level variable. Not a meaningful comparison.

let’s say we see a difference between story 1-5 we wouldn’t know if this is due to stories 1-5 being different from another or if it’s due to one being exotic and one being familiar.

23
Q

with nested ANOVA do we calculate estitated marginal means

A

no

while we can compute these it doesn’t make sense for us to do so.

But there is an issue with the confound. While SPSS will compute the EMM for you it won’t take into account the intrinsic meanings of story type and story number.