Zotero Chunk RAG

effects changed across the different conversation tasks (positive,
neutral, and negative) and across the conversation role (talking,
listening, and freely talking). However, the models fitted previ-
ously (i.e., Models 1–7) are not nested within these two new
models. Therefore, statistical fit cannot be compared between
them, and they are not included in Table 1. The first model tested
if the cross-partner effect changed during different conversation
tasks (-2LL  4367.4), and the second model investigated differ-
1 These models can also be fitted as structural equation models, which
can include a residual covariance. We did not estimate this parameter for
the current model within the mixed effects framework, although a test
showed that estimating this residual covariance within a structural equation
model had virtually no effect on the parameter estimates.
Comparisons of Model Fit for Different Tests of Coregulation
Model
	2 LL
df
-2 LL
p
1
4393.8
7
—
—
2
4387.3
8
6.5
0.01
3
4387.3
9
0.0
1.00
4
4377.1
10
10.2
0.01
5
4373.1
11
4.0
0.04
6
4368.1
12
5.0
0.02
7
4368.1
13
0.0
1.00
Note.
Models 1–7 are described in the text. 	2 LL  the models 	2
log-likelihood; df  the number of parameters estimated in the model; -2
LL  the difference between the prior and current model; p  the
probability of -2 LL, with p  .05 indicating better fit of the model with
more parameters; df  degrees of freedom.
Thi