[Taka Sakamoto]

I'm sorry I'm dense. I've read page 1, but could not tell. Thank you.

[Richard Williams]

Dear Richard:

I've read your handout and have a question. You give an example of sem on p. 9. Does this model also have to have a variable that influences X3 but not X4 and a variable that influences X4 but not X3?

Taka

See the diagram on p. 1. X1 directly affects X3 but not X4. X2 directly affects X4 but not X3.

To further clarify though, X1 indirectly affects X4 (X1 affects X3 affects X4). Similarly X2 indirectly affects X3 (X2 affects X4 which affects X3).

I've heard people say that X1 should be uncorrelated with X4 and X2 should be uncorrelated with X3. Not true. Although they do not have direct effects, they should be correlated because of the indirect effects.

This is basically a minimalist reciprocal causality model. Duncan/Haller/Portes is a somewhat more elaborate reciprocal causation model.

[George Ford]

That's pretty much what is happening, though SEM can allow all sorts of strangeness. The model in the video is simple and probably what you're after.

[Taka Sakamoto]

Thank you for explaining things to me. ① Going back to my question:

In the example you give on p. 9, does the model have a variable that influences X3 but not X4 and a variable that influences X4 but not X3?

② I have a panel data with about 30 years of annual data. It would be a long command if I adopt a cross-lagged model, right? It might make a cross-lagged model inpractical?

Thank you all for your help.

[Richard Williams]

In the example you give on p. 9, does the model have a variable that influences X3 but not X4 and a variable that influences X4 but not X3?

I think I answered that in post #17, but to be totally explicit, the answer is yes.

[Taka Sakamoto]

Thank you. So, I know it's gibberish, but if I estimate something like attached, it doesn't have a problem as a sem model?

USE_24.stsem

Thank you again for helping me out.

Taka

[Maarten Buis]

The errors need to be correlated. Right now your errors are considered the instruments, which just does not make sense.

[Taka Sakamoto]

Thank you, Maarten. I've learned a lot from this thread.

Taka

[George Ford]

x

[Taka Sakamoto]

I would like to ask if the model below has a problem. I understand I need to have a variable that affects dgdplev but not dlplev, and a variable that affects dlpev but not dgdplev. I don't know whether as long as I have the mentioned two variables, other variables can be added as necessary. Thank you in advance for your help.

USE_1.stsem

[Maarten Buis]

There is still no correlation between e4 and e5. That is absolutely crucial for such models, as stated in #22.

[Taka Sakamoto]

Thank you, maarten. I appreciate your kind help.

Taka

[George Ford]

Probably more for Maarten, but since dgdpleve and dlplev are directly related (the error term in on that relationship), can you then correlate the residuals? When I tried that, it would not estimate.