hi, i have multiple observations per person over the course of a week (i.e. observations within persons). I'd like to compare the effect of a level 1 predictor variable (task difficulty) on two level 1 dependent variables (concentration and anxiety). I'm interested in whether task difficulty is differentially related to the two dependent variables. Would such an analysis be possible in Mplus? If so, which modules would I need? thanks
Sure, participants carried around beepers for a week, and were randomly paged 8 times a day. Each time they were paged, they completed a short survey which assessed their current mood as well as what they were doing when they were beeped. So I'm interested in the within-person relationships between the variables mentioned above, thanks.
So repeated measures across time (level 1) within each person (level 2), where on level 1 you have 2 DVs and 1 IV? Given that you have many time points (8x7) this is hard to do using the single-level, multivariate, wide-data approach we usually recommend, and instead 2-level modeling would be used. That requires the base+multilevel add-on module.
Given I don't want to examine how the two relationships (task difficulty>>concentration and task difficulty>>anxiety) vary across time points, but rather to test whether the two relationships are different from one another, I'm having trouble understanding how to adapt the growth model presented in example 9.16 for my purpose.
Example 9.16 is relevant because it shows how an analysis is done when the data are in the long versus wide format. When data are in the long format, it is necessary to have person as the cluster variable. So let's say you have, two variables that you measured three times and you want to look at their relationship. Your data would look like:
Person y x Time 1 8 4 1 1 9 4 2 1 10 5 3 2 4 3 1 2 4 4 2 2 6 5 3
In the VARIABLE command, you would have CLUSTER=PERSON; as in Example 9.16.
With this situation, you have two alternatives. You can use TYPE=COMPLEX with y ON x in the MODEL command. Or you can use TYPE=TWOLEVEL. Then in the MODEL command, you would have y ON x on the within level. On the between level, you can estimate the intercept variance of y.
In example 9.16, is it possible to fix the covariance structure according to an observed variable? I have a sample of 2 children per family with varying degrees of genetic relatedness. I would like to do a multilevel model to test mediational hypotheses. Basically, for full siblings, half-siblings and unrelated siblings, I would like to use that covariance structure.
The variance of the random effect would reflect genetic effects, the random mean the shared family variance and the residual at the within level would be non shared environmental variance. Should the time variable be their degree of genetic dissimilarity (so that greater numbers reflect less covariance).
This wouldn't show me the variance of that relationship, would it? The theory is that the genetic variance is equal to the variance of a random effect, which is similar to something done with example 9.16 if I understand correctly.
Ex9.16 is a growth model and I don't see how time would translate in your genetic context. I thought you wanted to correlate the 2 children differently depending on their relationship, which is ex5.23.
The variance of a random effect can be used to represent the degree of assocation between variables influenced by that random effect. That's the case in ex 5.18. See how 5.18 translates into 5.21, which then gets generalized into 5.23.
You understood corrrectly. Here's an exercpt of the paper I'm referring to, which might explain it better than I ever could.
"In some cases, for example spatial or time series models this assumption is relaxed. In time series models the correlation between two measurements made on the same individual at different points in time is modelled as function of the distance in time between the measurements. With genetic effects we have an analagous situation, where the genetic correlation between two individuals is modelled as a function of the genetic distance seperating the two individuals."
Luisa Solms posted on Wednesday, September 23, 2020 - 7:10 am
I am having repeated measures data (daily assessments of several predictor and outcome variables) and I am wondering if I took the right approach. I first conducted a null model (intercept-only), then added my control variables at the between-level (time-invariant variables, such as age and sex) and finally conducted a model with all explanatory (time-varying) variables. I am not interested in growth but mostly in the relationship between my time-varying variables.