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

  • Mediation Effect for categorical DV, ordinal MV and ordinal IV in Stata?

    Dear Statalist experts,

    I’m looking for some advice that will help me with a research project im currently working on. The question I’m asking is whether Stata 14’s sem or gsem command 1 can be used to calculate indirect effects for a categorical mediator. If not, could anyone of you please point me to an alternative source, e.g., a Stata ado, I could use.

    For a better understanding of my question, please consider the following: The core of the theoretical model I have in mind looks like this x -> y1 -> y2 and x -> y2, whereby
    • variable x represents a respondents social origin (5 categories/groups).
    • variable y1 stands for the respondent’s parents agreement to a specific question about education-related attitudes and is available in form of 5-point Likert scale.
    • variable y2 represents the respondent’s answer to the same question his/her parents have been asked previously. Also available as a 5-point Likert scale.
    What I'm trying to do is:
    1. Find out whether the respondents and his/her parents attitudes vary according to their social origin/class position. Basically a simple group comparison to answer the question if people from higher-status backgrounds hold higher positive attitudes towards education than people from lower-status backgrounds.
    2. That’s where mediation analysis comes into play, assess whether the respondent’s attitude towards education is entirely related to his/her social origin (x -> y2) or mediated through his/her parents attitudes based on their social origin (x -> y1 -> y2).
    Sociological literature provides evidence for both mechanisms and I’d like to learn more about the underlying mechanisms behind attitude transmission.

    I already consulted the Stata SEM manual but couldn't find any advice regarding a categorical mediator. All examples within the manual use a continuous mediator. However, I know that only the gsem command will do the job when working with a ordinal outcome variable.

    I am writing here to ask if anyone's done it before with Stata. Do any of you happen to know any papers regarding this model?

    Any tips will be graciously taken and extremely helpful.

    Thanks for you help.

    1 – I asked for advice for both commands because some folks argue that one should treat Likert scale data as continuous and use the regular sem command with ML estimation but apply Satorra-Bentler correction while others say I must be done with gsem.

  • #2
    GSEM looks like your best option. An alternative would be to treat the discrete variables as continuous.