Effect Modification (Interaction)

Question 1

What is a modifier (Z)?

Answer 1

A variable that alters the relationship between the X1 and Y

Question 2

What is the interaction term?

Answer 2

The cross-product between X1 and the modifier

X1 x Z

Question 3

If β3 has a p<0.05, what can you conclude about the modification?

Answer 3

There is a significant effect modification.

There is a significant interaction between X1 and the modifier Z

Question 4

What are the main effects of the interaction interpretation?

Answer 4

β1 and β2

(effect of X1 on Y when Z=0, and effect of Z on Y when X1=0)

Question 5

If there is an interaction, how do you calculate the effect of X1 on Y?

Answer 5

β1 + β3 x Z

(Z = 0 or 1 if coded, use average if continuous)

Question 6

How do you interpret a negative interaction effect?

Answer 6

Effect of one variable decreases as the other variable increases

Question 7

How can you present continuous x continuous interactions?

Answer 7

▪️Table with quartiles
▪️Graph plotting quartiles

Question 8

What is an outlier?

Answer 8

An observation that lies an abnormal distance from other values in a random sample

Question 9

What are influential observations?

Answer 9

Outliers with large influence on the fitted regression model

May distort the actual relationship

Question 10

How can you spot outliers in your dataset?

Answer 10

▪️Sort by ascending/descending
▪️Look at minimums and maximums
▪️Plot on graph e.g. regression variable plots

Question 11

What are mild outliers?

Answer 11

Outliers that lie between 1.5 and 3 times the IQR (below Q1 or above Q3)

(between inner and outer limits)

Question 12

What are extreme outliers?

Answer 12

Outliers that lie more than 3 times the IQR (below Q1 or above Q3)

(beyond outer limit)

Question 13

What is the lower outer fence?

Answer 13

Q1 - (3 x IQR)

Question 14

What is the lower inner fence?

Answer 14

Q1 - (1.5 x IQR)

Question 15

How can you find outliers using standardised residuals?

Answer 15

Any observation with an absolute standardised residual larger than 3

Question 16

What is DFBETA?

Answer 16

For any given observations, measures the change in the estimated coefficient (β) due to deleting the observation (βwith - βwithout)

Standardised DFBETA ÷ SE(est β)

Question 17

What is DFFIT?

Answer 17

For any given observation, measures the change in predicted value (Y) due to deleting that observation

DFFIT ÷ SE(Y)

Question 18

What absolute value of DFBETA would suggest an influential observation?

Answer 18

> 1

Question 19

What absolute value of standardised DFFIT would suggest an influential observation?

Answer 19

> 1

Question 20

Why should you consider outliers?

Answer 20

Outliers can:
▪️Increase error variance
▪️Reduce power
▪️Decrease normality
▪️Violate assumptions of sphericity
▪️Alter odds of type I and II errors
▪️Bias/influence estimates

Question 21

What should you do if an outlier is a natural part of the population you are studying?

Answer 21

Keep it

Question 22

What should you do if an outlier is a measurement or data entry error?

Answer 22

Correct it if possible, if not remove it

Question 23

What should you do if an outlier is not part of the population you are studying?

Answer 23

Remove it

Question 24

What should you do when you remove an outlier?

Answer 24

▪️Document exclusion and explain reasoning
▪️Attribute a cause for removing it
▪️Potentially perform analysis with and without, then discuss difference