Background The psychonetrics package offers the possibility to estimate dynamic network models based on repeated measures data / panel data. For cross-sectional networks that are estimated using the estimateNetwork() function, there is a package called bootnet that can assess stability and accuracy of the estimated parameters. Is there a way to assess stability/accuracy of parameters for dynamic network models that have been estimated using the psychonetrics package?
My specific case: I have data from n=100 patients that were measured at 8 time points. The data are continuous and I standardized them. I used the ml_ts_lvgvar function and estimated a network model:
#Form model:
model.a1 <- ml_ts_lvgvar(data,
beepvar = "session",
idvar = "patid",
vars = vars,
estimator = "FIML",
standardize = "none")
#Estimate saturated model:
model.a1 <- model.a1 %>% runmodel
Furthermore, I applied a model search strategy (pruning). I am now trying to assess the stability of our edge-weight parameters by estimating bootstrapped confidence intervals.
#Prune model
model.a1_pruned <- model.a1 %>%
prune(adjust = "fdr", recursive = FALSE, alpha = 0.05)
So far in the literature, I have found this excerpt from Epskamp (2020): "To assess the stability of the estimation algorithm, I performed 1000 case-drop bootstraps (Epskamp, Borsboom, & Fried, 2017) in which I dropped 25% of the data and reestimated the model structure using the search strategy of Fig. 1."
However, as described in another post here on Stackoverflow: How do you assess stability in saturated network models that have used model search algorithms?, it is unclear how one can use the functions to assess stability/accuracy from the bootnet package when you have estimated the network model (with model search algorithms) using the psychonetrics package.
One idea based on the answer given to the post above, would be to create a custom estimationfunction to then use the estimateNetwork() function, and then run bootstrap using bootnet(). My question is how to create such a custom estimation function based on my model above (using the ml_ts_lvgvar function above and applying model search strategies)?
