W8: LMM Part 1

Question 1

What do you use egltabe( c (“loneliness”, “sex”), data =d, strict = FALSE) for?

Answer 1

To get summary descriptive statistics

Question 2

When using egltable() and plot(testDistribution) for BETWEEN variables, there is a problem of weighted means. How do you solve that?

Answer 2

Remove duplicate IDs
* egltable(c(“loneliness”, “sex”), data =
dm[!duplicated( ID ) ] )
* plot(testDistribution(
dm [!duplicated( ID )]$loneliness) )

Question 3

When getting descriptives for continuous RM (within) variables, there is an issue of including total variance (between and within). How do you only calculate the mean of individual means (between units only)?

Answer 3

Create between and within variables using meanDeviations()
m[, c(“Bstress”, “Wstress”) := meanDeviations(dStress), by = ID]
then calculate as usual:
egltable(c(“Bstress”), data = dm[!duplicated(ID)])

Question 4

If we want descriptives of mean for all observations (regardless of between or within), do we use dstress or bstress?

Answer 4

egltable(c(“dStress”), data = dm)
includes both between + within variance (possible unequal weights participants)

Question 5

What are 2 ways to group calculation of between unit variance for continuous RM variables

Answer 5

Group by days
egltable(c(“dStress”), g = “SurveyDay”, data = dm)
* or indicate individual timepoints
egltable(c(“dStress”), data = dm[SurveyDay == 1])

Question 6

For descriptives of categorical (within) RM variables, what are they normally reported as?

Answer 6

Frequencies (N) or percentages (%)
e.g using overall variable: egltable(“Int_Fri”, data = dm, strict = FALSE)

Question 7

What are the 2 common estimators used with LMMs?

Answer 7

1. Maximum Likelihood (ML): population variance
   REML = FALSE
2. Restricted Maximum Likelihood (REML): population variance from sample, less biased
   REML = TRUE

Question 8

What is ML used for?

Answer 8

Model comparisons

Question 9

What are 2 main uses of random intercept models?

Answer 9

1. Model comparison (how much better complex model fits)
2. Calculate ICC

Question 10

“scaled Pearson residuals” from lmer() output is useful to identify what?

Answer 10

Outliers based on “Min” (lowest residual) and “Max” (maximum residual)

Question 11

Std. Dev of “random intercept” from lmer() output represents what for a random intercept only model?

Answer 11

The average difference between individual’s average score and population average score of variable

Question 12

Std. Dev of “random residuals” from lmer() output represents what?

Answer 12

Average difference between individual score and predicted score of variable

Question 13

How do you calculate confidence intervals for random effects?

Answer 13

Using profile likelihood confidence intervals
confint(x, method = “profile”, oldNames = FALSE)

Question 14

Do we need to check if linear association between predictor and outcome is appropriate for intercept only models?

Answer 14

No

Question 15

When interpreting that there are individual differences, do we refer to standard deviation of random intercept or random residuals?

Answer 15

SD of random intercept

Question 16

After seeing individual differences..
Assuming the random intercepts follow a normal distribution, we would expect most people to fall within..

Answer 16

1 SD of the mean
(Std. Dev. from random intercept + / - est. of fixed effect)

Question 17

What random / fixed components are shown in this:
lmer(dStress ~ dEnergy + (1 | ID) )

Answer 17

Fixed effect of dEnergy
Random intercept : (1 | ID)

Question 18

When there is a fixed predictor and random intercept involved in a model, how do we interpret the Std. Dev. of random intercept from the output?

Answer 18

It is the average difference between an individual’s estimated score and population average estimated score when predictor is 0

Question 19

When there is a fixed predictor and random intercept involved in a model, how do we interpret the estimate of the fixed effect from the output?

Answer 19

It is the mean/average score when predictor is 0

Question 20

What fixed or random components are shown in this:
lmer(dStress ~ Benergy + Wenergy + (1 | ID) )

Answer 20

Fixed effects of Benergy (average energy per person across days) and Wenergy (individual deviations from that person’s average energy)
Random intercept : (1 | ID )

Question 21

What does the Std. Dev. of random intercepts tell us from the output of this model:
lmer(dStress ~ Benergy + Wenergy + (1 | ID) )

Answer 21

The average difference between individual’s estiamted stress and population’s average estimated stress when Benergy and Wenergy are 0

Question 22

What does the fixed intercept tell us from the output of this model:
lmer(dStress ~ Benergy + Wenergy + (1 | ID) )

Answer 22

Average estimated stress score when Benergy and Wenergy are 0

Question 23

What does the fixed estimate of Benergy tell us from the output of this model:
lmer(dStress ~ Benergy + Wenergy + (1 | ID) )

Answer 23

Expected change in stress on average across days when average energy is 1 unit higher

Question 24

What does the fixed estimate of Wenergy tell us from the output of this model:
lmer(dStress ~ Benergy + Wenergy + (1 | ID) )

Answer 24

Expected change in stress that day when energy is 1 unit higher than individual’s own average that same day

Question 25

REML is less biased than ML because it doesn’t require the….

Answer 25

mean

Question 26

The fixed intercept from lmer() means what?

Answer 26

Average estimated y when predictor(s) = 0
* the mean of the random intercept

Question 27

What is the difference between Wald and Profile confint methods?

Answer 27

Wald can only calculate fixed effect confint (intercept)
Profile Likelihood can calculate both fixed and random effects