No announcement yet.
  • Filter
  • Time
  • Show
Clear All
new posts

  • Seeking advice on a three-level model

    Dear Forum-goers,
    I am investigating a moderation effect in a three-level dataset. The data are in the format of a three-wave unbalanced panel with observations nested within employees, nested within organizations. I am examining the effects of a dichotoumous moderator policy "P" at the employee level on a the effects of a continuous employee-level variable "X." The outcome is an employee-level variable, "Y." I have all variables at three waves, but for various reasons I have so far used the data as a pooled cross-section

    I have thus far modeled the data using xtmixed as follows (omitting controls):
    xtmixed Y c.X##i.M i.year || org_id: || employee_id:
    I have had a few comments made to me about this strategy and I want to ensure I am interpreting my model correctly. My understanding is that by including random intercepts for organization and employee levels, the first part of the model is at the within-person level, and while I do not explain any person-level or organization-level effects, such effects will not bias standard errors in the estimates of the within-person effects. Is this correct? Or am I misinterpreting? I am not interested in the higher-level effects but I want to appropriately account for variation in Y at those levels in my model.

    Some have also suggested to me that instead of this approach, to fully utilize the panel structure of the data I should use SEM to examine within-person changes. Am I correct that I could use gsem to account for the organization-level nesting and also include the interaction term? I have never used gsem and am unsure I would be capturing what I want with what I have come up with. I am thinking it would look something like this (again, omitting controls and measurement models; Y is a latent variable while X and M are directly observed):
    gsem (Y_T3<- c.Y_T2 c.X_T2##i.M_T2 M1[org_id_T3]) ///
        (Y_T2 <- c.Y_T1 c.X_T1##i.M_T1 M1[org_id_T2]) ///
        (Y_T1 <- M1[org_id_T1])
    Finally, I am limited to Stata 14 due to restrictions in data access. Thank you for any guidance you can provide. It is much appreciated.