Announcement

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

  • What is the syntax for using 'init' with Latent Variables in SEM to address issue with 'initial values not feasible' error r1400?

    I am working with longitudinal data and using SEM and have been receiving the 'initial values not feasible' error.. I have re-scaled some of my variables according to the advice given in this forum to other inquiries of this type, and got some success with somewhat simplified versions of my model - using cross-sectional data and using indicators/observed variables only, rather than latent variables along with indicators/observed variables. The Stata manual states that the problem with 'initial values not feasible' can be addressed by the following:

    Specify initial starting values as

    You need to give that variance a larger starting
    value, which you can do by typing
    . sem ..., ... var(e.F, init(1))l
    or
    . sem ..., ... var(e.F, init(2))
    or
    . sem ..., ... var(e.F, init(10))


    Also the following:
    To set the initial value of a path coefficient,
    add the init() option where the path is specified. Say the original sem command included y<-x1:
    . sem ... (y<-x1 x2) ...
    If you wanted to set the initial value of the path from x1 to 3, modify the command to read
    . sem ... (y<-(x1, init(3)) x2) ...

    I was unable to find an example of a SEM script that shows 'init' being used.

    My SEM script has the arrows going in the opposite direction and looks something like this:

    sem(LatentVar> ObservedVar, )..... (LatentVarA-> LatentVarB, ) .... latent(LatentVarA LatentVarB .....) method(mlmv) nocapslatent

    In this context how should the 'init' option be used? Can you please provide examples?
    Last edited by sveera; 19 Mar 2017, 06:34. Reason: added tags
Working...
X