HI all,
I have been working with daily diaries data ( 6 emotions of which 3 positive and 3 negative) were assessed for 30 days. These observations (1level data) are then nested within subjects (2nd level data) who are further nested in three groups (approx.50 in each group, 3rd level data).
What am I struggling with is finding out how to manage the time covariate in 1level data because the daily observations were further nested in odd and even days and then three groups of people toward who the participants were asked to indicate the feelings for.
In short,
on odd days a person was asked to indicate 6 feelings ( disgust, empathy, anger, sympathy, regret, respect) toward x1, s1, s2, s3, s4, s5, a1
on even days indicate the same six feelings toward x2, s6, s7, s8, s9, x3
As you can see, x represents one type of group of representatives, s second type and a1 stands alone.
My role is to find out if there is any difference in intraindividual variability of emotions between the 3 groups.
I will appreciate any tip how to manage all this complexity. Thanks so much!
Hello Milena Ferencikova ,
if I see this right, you should performa a comparison between variances.
for example, by means of a F-test.
This test compares the variances for groups or values.
1. Since you are working with ordinal scaled data, I would suggest to calculate means for each group of feeling.
2. then, I would compare the variance of these mean values between the groups (1: even days, 2: odd days) (F-test, also Levene-test possible)
3. test statistics will show if there is equal or unequal variability between days, for each group of feeling.
If done so, nested effects would not be crucial to include, since variance is compared hereby.
I am not sure if this addresses what you wanted to know - maybe you could explain the research question in more detail, in particular among which groups you want to compare.
HTH, Marian
Dear Milena Ferencikova,
I think more information is needed about your data to be sure, but if you want to build a linear mixed model, then with the assumptions that
1) your DV (IIV, however you define it) is across all 6 emotions and you are not interested much in any subgroups there (e.g., based on valence);
2) representatives are something meaningful and are grouped in 3 types (x, s, a) which are conceptually random (e.g., are positive, neutral, negative pictures; I'm really generally guessing what representatives are in your data);
and 3) the day (odd/even) is somehow meaningful to you and you expect it may play a role in the data,
the model would look like this:
IIV ~ group + day + (1|person/representative)
Day would be a 2-level categorical predictor (odd, even), person would be the subject ID, and representative would be a 3-level predictor (x, s, a). The day is conceptually a random term, but it only has two levels so it makes more sense to add it as a fixed one and consider it a covariate (but you can try to add it in the random term as (1|person/day/representative)). If you're not sure if it should be a part of a model given the data structure and design, just check whether it improves the model's fit.
I don't know how your IIV is calculated, but you can also look into ICC (interclass correlation) as a way to grasp emotion diversity/complexity.
Hope that helps, good luck!
- Badges
- Science topic
- Similar topics
- Psychology
- Emotion