Suggestion for adapting the parameter aggregation

[KarolineHuth]

Hi Jonas,

in some recent work, we looked at the nodewise estimation approach for networks https://osf.io/preprints/psyarxiv/nmdxe. In the manuscript, we looked at the asymptotic properties of different methods of aggregating the parameters. In there, we show that merely averaging the two parameters leads to an asymptotically biased estimator. Instead, one should multiply or rescale the nodewise parameters. To avoid the bias in MGM, we recommend updating the R code that aggregates the edge weights in your script. If adopting the multiplied approach, one can obtain the edge weights with

sgn(weights_matrix)sqrt(weights_matrixt(weights_matrix))

Happy to discuss further.

Best, Karoline