9. Independent groups One Way ANCOVA

Question 1

Why is a confound a problem?

Answer 1

Reduces the power of the effects

Question 2

Why do you introduce a covariate?

Answer 2

To control for/ partial out the effect of a confound (and increase the power)

Question 3

What are the 3 general purposes of using an ANCOVA?

Answer 3

Noise reducing
Descriptive model building
Step-down analysis

Question 4

What type of multiple linear regression model would you use with ANCOVA?

Answer 4

Hierarchical (entering the covariate in Block 1)

Question 5

Total Variability =

Answer 5

SSB + SSW

Question 6

Df of SSB?

Answer 6

k-1

Question 7

Df of SSW?

Answer 7

N-k

Question 8

What are the 3 additional steps for an ANCONA?

Answer 8

1. Regress the DV on the CV
2. Remove linear effects of CV in DV and IV group means
3. ANOVA on Adj values

Question 9

What are the two primary cautions/ reasons not to use an ANCOVA?

Answer 9

Dependence of CV and IVs (they must be unrelated)

Multicollinearity

Question 10

When using an ANCOVA, why must the CV and the IVs be unrelated?

Answer 10

Otherwise removing the effect of the CV would remove some of the effect of the IV on the DV

Question 11

What are the 3 assumptions of a one way independent groups ANOVA?

Answer 11

Normality
Homogeneity of Variance
Independence

Question 12

What are the 5 assumptions of an ANCOVA?

Answer 12

Normality
Homogeneity of Variance
Independence

Linearity
Parallelism

Question 13

What is the assumption of Linearity?

Answer 13

The relationship between the CV and DV are linear (otherwise the statistical power is reduced)

Question 14

How do you test for linearity?

Answer 14

Pearson Correlation must be sig. (e.g. they have a linear relationship)

Question 15

What does it mean when a CV does not have a sig. r with the DV?

Answer 15

It is actually not a confound/ Covariate

Question 16

What does parallelism mean?

Answer 16

The slope of the CV should be the same as the slope for each level of the IV (given by the b coefficient or the scatterplot)

Question 17

What do we need to check if the b coefficient (or the lines on the scatterplot) aren’t parallel between the CV and the IV levels?

Answer 17

The significance of the interaction term (All IVs * All CVs combinations)

If it is non sig. then they are parallel = same effect on all levels = good!

Question 18

The df of freedom in an ANCOVA are? (3)

Answer 18

Between: k-1
Covariate: number of CVs
Within: N - k - number of CVs

Question 19

What do we need to do in an ANCOVA if we find a sig. main effect for two levels of the IV on the DV?

Answer 19

Conduct a pairwise comparison (adjusted for the effect of the CV)

Question 20

What are adjusted means?

Answer 20

Means of the IV levels that have been adjusted for the CV

Question 21

What can’t be conducted with an ANCOVA?

Answer 21

Post-hoc tests

Question 22

What conventions do we use to analyse the effect size of eta squared?

Answer 22

.01 = weak 
.09 = moderate 
.25 = strong

Question 23

What is the formula to compute effect size for contrasts or pairwise comparisons?

Answer 23

sq rt of
t sq /
t sq + df

Question 24

What assumptions do you check for between subjects ANCOVA?

Answer 24

Normality, Linearity and Parallelism

Homogeneity of Variance
Independence

Question 25

What assumptions do you check for within subjects ANCOVA?

Answer 25

Normality, Linearity and Parallelism

Sphericity/ Compound Symmetry
Dependence

Question 26

What assumptions do you check for mixed model ANCOVA?

Answer 26

Normality, Linearity and Parallelism

Homogeneity of Variance
Sphericity
Independence
Dependence

Question 27

What should you do if in an ANCOVA the assumptions are not met?

Answer 27

Block the CV (break into 2 categories e.g. high or low)