|
Message/Author |
|
|
If I could get an answer to this question or some brief thought by tomorrow I would really appreciate it. I tried to contact you earlier but you were away. Sorry. There has been a lot of discussion regarding causal inference with regards to logistic and probit models . In particular interest to me was Judea Pearl 2010 The mediation formula. In my analysis I used the SE model for binary outcomes (in MPLUS). My mediator was also binary and I used bootstrap confidence intervals. How do the results from that analysis deal with the questions raised by Pearl? What is the interpretation of the estimate, does it have a causal interpretation? of any kind? |
|
|
David MacKinnon discusses causal inference and mediation in his book Introduction to Statistical Mediation Analysis. Perhaps this could be helpful. |
|
Kihan Kim posted on Sunday, March 27, 2011 - 6:12 pm
|
|
|
I ran the following SEM model, and the path coefficient from "presence" to "suspen" was negative. But when I looked at the correlations betweeen the items of Presence (i.e., w57 w58 w64 w69) and the items of Suspen (i.e., w36 w37), they were all positively correlated. Is it possible that the correlatiosn between two factors are posiitve, but the path coefficient sign is negative? How would I understand such relationship? Model: presence by w57 w58 w64 w69; immers by w66 w67 w68 w70; negef by w83 w84 w85; game by w29 w30; suspen by w36 w37; enjoy by w14 w16; enjoy on suspen; suspen on presence immers negef game; Thanks! |
|
|
The regression coefficient for suspen on presence is a partial regression coefficient. It depends not only on the relationship between suspen and presence but also the relationships among presence, immers, negef, and game. |
|
Back to top |
|
|