dyadic growth model

129 views
Skip to first unread message

Loren L. Toussaint

unread,
Jan 15, 2024, 6:39:14 PM1/15/24
to lavaan
Hi,

First time posting... I am attempting to run a dyadic growth model in lavaan using the following syntax.

m1 <- 'ip1 =~ 1*PSS.1.00 + 1*PSS_W2.1.00 + 1*PSS_W3.1.00
           sp1 =~ 0*PSS.1.00 + 1*PSS_W2.1.00 + 2*PSS_W3.1.00
           ip2 =~ 1*PSS.2.00 + 1*PSS_W2.2.00 + 1*PSS_W3.2.00
           sp2 =~ 0*PSS.2.00 + 1*PSS_W2.2.00 + 2*PSS_W3.2.00
                
         '
fit_m1 <- growth(m1, missing="ML", data=MERGED4)
modindices(fit_m1, sort = TRUE, maximum.number = 5)
summary(fit_m1, fit.measures=TRUE, standardized=TRUE)

This model runs fine, but in the reference below, I am told to implement a number of constraints on the latent intercepts and slopes. I have no idea how to do so. For reference, ip1 = intercept person 1, and sp1 = slope person 1, etc.

My dyads are indistinguishable, hence the need for the constraints. Any help would be greatly appreciated. And, if anyone has worksheets, calculators, etc. for adjusting fit indices I would be interested in that as well.

Reference:
Iida, M., Savord, A., & Ledermann, T. (2023). Dyadic longitudinal models: A critical review. Personal Relationships, 30(2), 356–378. https://doi.org/10.1111/pere.12468

Many thanks!
Loren

Terrence Jorgensen

unread,
Jan 16, 2024, 6:28:54 AM1/16/24
to lavaan
I am told to implement a number of constraints on the latent intercepts and slopes. I have no idea how to do so. For reference, ip1 = intercept person 1, and sp1 = slope person 1, etc.

You can see how to specify variance and intercept parameters here: https://lavaan.ugent.be/tutorial/syntax1.html
You can see how to place equality constraints by using the same label here: https://lavaan.ugent.be/tutorial/syntax2.html#simple-equality-constraints

Don't forget to specify the "inter-person residual correlations" mentioned in Section 4.2

if anyone has worksheets, calculators, etc. for adjusting fit indices I would be interested in that as well.

Adjusting fit indices?  If you mean that you need to specify equal intercepts and variances across partners in the independence model (to calculate the correct CFI, TLI, etc.), then you can just fit that constrained independence model yourself and pass it to the fitMeasures(..., baseline.model=) argument.

Terrence D. Jorgensen    (he, him, his)
Assistant Professor, Methods and Statistics
Research Institute for Child Development and Education, the University of Amsterdam
http://www.uva.nl/profile/t.d.jorgensen


Loren L. Toussaint

unread,
Jan 16, 2024, 11:37:51 AM1/16/24
to lav...@googlegroups.com, tjorge...@gmail.com
Hi Terrence,

Thanks so much for your reply. My model is a growth model, so I need to set constraints on the latent intercept and variance because of the dyadic nature of the data.  Text is below:

If a researcher has an indistinguishable dyadic design, they will need to apply a set of constraints and adjust the fit of the model (Kashy et al., 2008; Olsen & Kenny, 2006).

The first set of constraints is that the two means of the latent intercepts and the two means of the latent slopes need to be set equal. (both are latent)

Another set of constraints is that the variances of the intercepts, the slopes (both are latent), and the residuals (of the observed variables) need to be set equal across the two dyad members. 

A final set of constraints is that the intra-person and inter-person associations between the intercept and slope factors are set
equal for both partners. (these are estimated automatically by lavaan, so I don't know how to constrain them).

I'm still quite new to R, but I think the "growth" script is designed to help estimate growth models easily, but then makes it difficult to make adjustments when the dyadic growth model is employed. The links you shared are helpful, but I don't know how to employ these constraints when the elements that need constraining are latent or "implicit" in the model syntax.

Truly appreciate your help!

Loren



--
You received this message because you are subscribed to a topic in the Google Groups "lavaan" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/lavaan/0fz2JlqMpcY/unsubscribe.
To unsubscribe from this group and all its topics, send an email to lavaan+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/lavaan/ef42a960-858c-48d1-88fa-eee4b4be7497n%40googlegroups.com.


--
Loren Toussaint, Ph.D.
Department of Psychology
Luther College
700 College Dr.
Decorah, IA  52101
Voice:  563-387-1647
Fax: 563-387-1080
Email: tous...@luther.edu
Zoom Personal Meeting Room: https://luther.zoom.us/j/3526506636

Terrence Jorgensen

unread,
Jan 17, 2024, 8:20:39 AM1/17/24
to lavaan
My model is a growth model, so I need to set constraints on the latent intercept and variance because of the dyadic nature of the data. 

The latent intercept and latent slope are variables.  Each of them has an intercept and a variance.  Follow the links to see how to specify them.

The first set of constraints is that the two means of the latent intercepts and the two means of the latent slopes need to be set equal. (both are latent)

For example:

ip1 ~ mean_int*1
ip2 ~ mean_int*1
 
Another set of constraints is that the variances of the intercepts, the slopes (both are latent), and the residuals (of the observed variables) need to be set equal across the two dyad members. 

For example:

ip1 ~~ var_int*ip1
ip2 ~~ var_int*ip2

A final set of constraints is that the intra-person and inter-person associations between the intercept and slope factors are set
equal for both partners. (these are estimated automatically by lavaan, so I don't know how to constrain them).

You just have to specify them in your syntax.

## intraperson
ip1 ~~ i_s_Cov_intra*sp1
ip2 ~~ i_s_Cov_intra*sp2
## interperson
ip1 ~~ i_s_Cov_inter*sp2
ip2 ~~ i_s_Cov_inter*sp1

Also estimate interpersonal correlations of indicators:

PSS.1.00 ~~ PSS.2.00

And set the residual variances equal across partners:

PSS.1.00 ~~ evar1*PSS.1.00
PSS.2.00 ~~ evar1*PSS.2.00

I'm still quite new to R, but I think the "growth" script is designed to help estimate growth models easily

Only simple growth models.

but then makes it difficult to make adjustments when the dyadic growth model is employed

For anything more complex, just specify your model parameters in the syntax and use the lavaan() function instead of a shortcut that doesn't work for your situation.

Loren L. Toussaint

unread,
Jan 25, 2024, 3:58:03 PM1/25/24
to lav...@googlegroups.com, Terrence D. Jorgensen
Terrence,

My thanks is well overdue, but you saved me weeks of frustration. I followed your guidance and the model ran just perfectly! Well, I have lots of missing data and the matrix is not positive definite, but the model syntax was easy to write once you shared your wisdom!  Wonderful to have this group and thanks so much for all of your help!

With much appreciation,
Loren

--
You received this message because you are subscribed to a topic in the Google Groups "lavaan" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/lavaan/0fz2JlqMpcY/unsubscribe.
To unsubscribe from this group and all its topics, send an email to lavaan+un...@googlegroups.com.
Reply all
Reply to author
Forward
0 new messages