Announcement

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

  • Mediation problem

    Hi all,I was investigating whether political identity would affect income through social networks, so I constructed an interaction term and got the following results; the first column is the regression result with the interaction term, and the second column is the normal fixed effect. But the interaction term is negative, can it be used as an intermediary variable and research the mediation mechanism?Thanks
    Click image for larger version

Name:	111.JPG
Views:	1
Size:	72.2 KB
ID:	1639370

  • #2
    Here you are confusing interaction and mediation. What you present is an interaction. If you want a mediation (what seems to be the case in your research question), you need to compute two nested models. Then first without your mediator and the second including it. An example is given here: https://stats.idre.ucla.edu/stata/fa...e-sem-command/
    Best wishes

    (Stata 16.1 MP)

    Comment


    • #3
      Originally posted by Felix Bittmann View Post
      Here you are confusing interaction and mediation. What you present is an interaction. If you want a mediation (what seems to be the case in your research question), you need to compute two nested models. Then first without your mediator and the second including it. An example is given here: https://stats.idre.ucla.edu/stata/fa...e-sem-command/
      Thank you @Felix Bittmann ,But I found out that when I used "relation" as an mediation variable, it was found to be significantly negative. This is strange. Does it mean that the mediation variable is incorrect?

      Comment


      • #4
        relation
        Click image for larger version

Name:	222.JPG
Views:	1
Size:	70.7 KB
ID:	1639399

        Comment


        • #5
          What you need to compare is the coef of party between M1 und M3. As you see, the coef is not getting smaller in M3 (0.276 vs 0.283) so there is no mediation effect present.
          Best wishes

          (Stata 16.1 MP)

          Comment


          • #6
            Originally posted by Felix Bittmann View Post
            What you need to compare is the coef of party between M1 und M3. As you see, the coef is not getting smaller in M3 (0.276 vs 0.283) so there is no mediation effect present.
            Thanks!@Felix Bittmann , and I still have a small question, because I also made a moderating effect, treating the “relation” as the moderating variable in party member to income and the moderating variable in trust to income; but I found that in the centralized model (3), the interaction term it has become less significant, and it is negative, can I still have a moderating effect? thanks!
            Click image for larger version

Name:	111.JPG
Views:	1
Size:	73.9 KB
ID:	1639460

            Click image for larger version

Name:	112.JPG
Views:	1
Size:	76.4 KB
ID:	1639461

            Comment


            • #7
              You already see that a moderation (interaction) is present in your first post since the interaction variable is highly significant. This means that social networks work differently, depending on the party relation. Since you have to continuous variables, this is however quite difficult to interpret. But all in all, I wonder what is your theoretical model. I am still not sure why you chose interaction or mediation. This is a huge decision in your research and should not be driven my the size of your coefs.
              Best wishes

              (Stata 16.1 MP)

              Comment

              Working...
              X