Final Within subjects design

Question 1

In terms of variance your regression assumes that

Answer 1

There is no variance on your x

Question 2

Another assumption that no longer applies here is

Answer 2

Independence of data

Question 3

Composite variables

Answer 3

Are another form of aggregation

Question 4

When is it good to use a compound scale

Answer 4

To deal with poor properties of a likert scale

Question 5

Simple form of aggregation assumes that______

Answer 5

each question contributes equally – they do not!

Question 6

Issues with power analysis

Answer 6

Rules of thumb, problematic because it requires you have an idea of how big an effect you are going to see, how much noise

Question 7

Why are rules of thumb different depending on the field?

Answer 7

Because they depend on the effect size and expected variance

Question 8

This is not a within subjects design

Answer 8

A cross sectional design (is between)

Question 9

Is a way of treating pre-post data

Answer 9

Using a differential score (careful with Lord’s paradox)

Question 10

Lord’s paradox

Question 11

You have to do this to compensate for random assignment missing here

Answer 11

Counterbalancing

Question 12

Counterbalancing

Answer 12

You make sure all your subjects experience all conditions in a random order

Question 13

This is the ideal form of counterbalancing

Answer 13

Full counterbalancing
ABC…
ABC, ACB, BAC, BCA, CAB, CBA

Question 14

Reverse counterbalancing

Answer 14

Middle levels never experienced as first or last
Any two levels always experienced sequentially

AB—-BA
ABC—CBA

Question 15

Latin square or William Latin square

Answer 15

Another type of couterbalancing

Every level is presented in every order
ABC
BAC
CBA

Question 16

Is the only way to get an A after B half the time and B after A half the time

Answer 16

A latin William’s square not the standard

Question 17

Simpliest repeated measures design

Answer 17

No predictors/IVs, but each person is measured repeated times

Now the average performance is the intercept

Question 18

In a within subjects design the subject

Answer 18

Subject experiences multiple levels of the continuous predictor

Question 19

Random effects

Answer 19

Each subject has its own slope and intercept. Random because each subject was selected randomly

Question 20

If you have longitudinal within-subjects and you are interested in changes in slope/intercepts

Answer 20

ANOVA/regression analysis

Question 21

If you have longitudinal within-subjects and you are interested in pre/post differences

Answer 21

We use pre as a predictor. You don’t want a significant interaction here. This is and ANCOVA approach

Question 22

Nested

Answer 22

Means that things cannot be factorially combined. The best way to treat this is as a x level design with imbalance in the conditions

Question 23

These are the 4 most historically common methods to analyze repeated measures data

Answer 23

1. Average across observations and analyze as between-subjects design
   2.Univariate ANOVA design
   3.Multivariate ANOVA design
   4.Multilevel Modeling

Question 24

Average across observations and analyze as between main issue is…

Answer 24

You create an aggregate doing this and the problem with an aggregate is you miss information about variability, how big was the sample size

Question 25

Univariate ANOVA approach problems

Answer 25

Problem 1. Univariate repeated measures ANOVA. The issue is that softwares assumes "sphericity/circcularity" basically how much things are covarying with each other. Problem 2. All your predictors have to be categorical Problem 3. Can't handdle missing data. Propense to Imputation

Question 26

Imputation

Answer 26

Is an educated guess of what a missing value's value will be

Question 27

Multivariate ANOVA approach

Answer 27

You have multiple measures/predictors and you treat them as a single big factor of Y. Problem 1. time-consuming to run. Problem2. Does not make any assumption about variance and covariance

Question 28

Multilevel Modeling

Answer 28

You can specify what your variance/covariance assumptions might be Can include continuous and categorical predictors Can handle missing data You can do GMM (Generalized Multiple Models). You can specify binomial outcomes, count variables, gamma distributed variables

Question 29

Rule of thumb in a withing subjects design for sample size

Answer 29

The sample size doesn't depend on the number of levels of within subjects variables. That only applies to between subjects

Question 30

Nested Designs

Answer 30

Are not factorial designs. Because different levels of the variable can take different values upon group conditions. Some levels are nested in one condition and you have to specify that to do an analysis but you won't get the result of an interaction

Question 31

Issues with aggregation

Answer 31

No information about the variability was there for that particular aggregate. How big was the sample size. Information is lost

Question 32

These are all names of Multilevel modeling

Answer 32

Hierarchical linear Models (HLM) Random Coefficient Regression Models Mixed Effects Modeling