Week 8

Question 1

For continous variables, which information do we want to look at: M (SD) or N(%)?

Answer 1

Mean and Standard Deviation.

Question 2

With fixed effects in a random intercept model, what are you estimating between X and Y?

Answer 2

The overall association between X and Y for the WHOLE sample without allowing the strength of direction of the effect/slope to vary randomly by person.

Question 3

If you have a fixed effect in a random intercept model, are you assuming the effect of X on Y will be different or the same for everybody?

Answer 3

The same

Question 4

If you have a random intercept with a fixed effect, and say Y is stress and X is energy, what is the random intercept actually showing?

Answer 4

That the mean value of stress does vary between persons because it is random, however, the overall effect of energy on stress for the whole sample is fixed - this does not vary

Question 5

The Int_Fri variable contains information on whether a person interacted with a friend on a day. One = they did, Zero = they didn’t.

Does the 0 and 1 column here tell us that 64% of people interacted with a friend on any day?

Answer 5

No. It tells us that of the observations (days) in the survey, 64% featured an interaction with a friend

Question 6

In the Bstress variable, or between stress, the average scores of stress were firstly averaged across the days per person

(e.g ID 10 got 3, 2, and 5 across three days. Then divided that by 3)

THEN that score is ….?

Answer 6

Averaged across the sample for all people.

Question 7

What type of variable is being shown here and how has the data been dealt with?

Answer 7

The betweenstress variable (Bstress), which is avg. score per person across days, and for those we only want ONE row of data input per person (not duplicated operator being used).

Why? Because we want to avoid inflating the mean for the individuals who filled out more observations.

Question 8

To ensure that repeated baseline variables (seen in between person variables such as sex on loneliness measures, because often the things that are repeated are biological sex) do not contribute a higher weighting to the mean, what might we do?

Answer 8

We drop the duplicated ID by using thegltable code but als data= dm[!duplicated(ID)]

If you have long format data set and want to report descriptives statistics on variables that do NOT change across time points, this is one way to include non duplicated ID’s.

Question 9

When looking at descriptives of any between subjects variables in multilevel data need to ask yourself…?

Question 10

If the variable does NOT change within a person in a dataset, do you want the mean of all observations?

Answer 10

No, as people with more observations will count more towards the descriptives

Better to have the mean of the observations within the one ID.

Question 11

Will this variance in this WITHIN continous variable show variance in both between and within aka total variance?

Answer 11

No, just mean of individual means so variance will be between units only.

Question 12

In this within continous variable, are we seeing the mean of individual means, so between units only, or mean across all observations, so variance in both between AND within units also known as TOTAL Variance?

Answer 12

Yes! Variance in both between and within units, aka TOTAL VARIANCE.

Question 13

What could be a problem with taking the mean across ALL observations, not just people?

Answer 13

You include a lot of variance from people who provided more data. Potentially bias the results.

Question 14

In the mean of individual means, what will the variance be taken from?

Answer 14

Between units only

Question 15

In the mean across ALL observations, what will the variance be taken from?

Answer 15

Variance will be TOTAL variance taken from both between and within units

Question 16

How can you report variance on between units only?

Answer 16

Average by person and report descriptives on individual averages

or

Report descriptives on individual timepoints/assessments.

Question 17

How can you report total variance?

Answer 17

Report descriptives on the overall variable whih captures the total variance BUT may unequally weight participants (as takes into account variance from people who provided more observations)

Question 18

What are the two variance estimators for our LMM?

Question 19

Of the two variance estimators, ML and REML, which one would we use and why ?

Answer 19

Prefer to use REML as it doesn’t need to know the mean, therefore it is less biased, whereas ML estimates are biased.

REML is good if you need variance estimates, BUT because it transforms the data we can’t use for model comparison

Question 20

If you see a SD for a model is fixed at 0. And the model is normally distributed. What type of model can you assume you are looking at and why?

Answer 20

Because it is fixed at 0 we assume it is a fixed effect as 0 does not allow variation in the intercepts or for random effects.

So it’s just assumed you are looking at the estiamted mean across everyone.

If the intercepts are mixed effects then individual variation in the means(aka intercepts) is allowed for.