I built a two levels hierarchical data like parents in families. I want to summarize my data. I have 100 parents nested in 34 families and I used variable such as parents gender and age in level 1 (individual level) and family income, family structure as level 2 (family level variables).
If I want descriptive statistics for each level i.e., individual and family level, should I built two table: 1 with level 1 stats using 100 as denominators for means and a another table with level 2 variable using 34 as denominator for calculations?
My head tells me to check with survey Package in r for complex survey design.
Can any one help me?
Thanks
Hello Abdoulaye Ouattara ,
cosidering the description of data, I assume that the data is not perfectly balanced, or are 3 parents contained in every family?
To check means at each level, it is vital to use least square means, if the data is not balanced.
Fortunately, smart people alreade took care of it, the package {emmeans} provides the acquired solution for that.
library{emmeans}
e.g.
(em
You probably could use the Maximum likelihood (ML) and restricted maximum likelihood (REML) estimation. Type I and Type III sums of squares can be used to evaluate different hypotheses. Type III is the default.
Do you suspect there to be correlated or non constant variability?
Then you can use the GLM Univariate or GLM Repeated Measures procedure.
Also you can alternatively use the Variance Components Analysis procedure if the random effects have a variance components covariance structure and there are no repeated measures.
You could use simple functions such as mean() in R to compute descriptive statistics, I mean you do not need to use that package.
I am just curious, why do we have 34 families but with 100 parents? I thought a family has maximum two parents?
Hello Trung Dung Tran
I think this can create misleading results, as a different number of observations might put a heigher weight on some group.
For example, you have two treatments, one with 100 obs. with a mean of 0, second treatment with 5 observations, with mean of 1.
If one applied mean among both treatments, the value would be very close to 0, insted of 2.5. Thats why to use least square means.
BR, Marian
Marian Schönauer Maybe I am not aware of something complicated, but for me, for example, if we would like to see average family income, we can do mean(family income) in R.
I do not want to complicate things further but it might help to know that the mixed model results will not be the same as the usual least squares or arithmetic average as they will be precision weighted , which I think is a good thing; see Article Do multilevel models ever give different results?.
- Badges
- Science topic
- Similar topics
- Social Science
- Media
- Social Media