Is Replication necessary for analys?

67 views

Skip to first unread message

ghz weng

unread,

Mar 10, 2021, 5:54:20 AM3/10/21

to IDTxl

Dear developer,

It seems that the number of replication influence the result of MTE much. For my real data, I have no replication, MTE only find few relationships in network(4 relationships among 33 processes). If I copy my data to create repliations(from (33,1448) to (33,1448,4)), MTE would find many relationships between processes.

So is the replication vert important?

Best wishes

p.wol...@gmail.com

unread,

Mar 12, 2021, 6:07:23 AM3/12/21

to IDTxl

Hi,

no, replications are not necessary per se. For IDTxl, replications are repetitions of your data collection process, i.e., repeated observations of the same phenomenon such as repetitions of an experiment (see also the tutorial on the data class). For such cases, IDTxl lets you provide 3D data consisting of processes, samples, and replications and uses all these samples for estimation. You can just as well provide only a single sufficiently long continuous observation without replications. This depends on your data.

If a relationship that is actually present can be found depends on things like how strong the relationship is, the amount of noise, the number of samples, etc. Generally, more samples help in finding a (weaker) relationship. Note that more samples means you collect more observations, but not copying your data. By copying your data you introduce artificial regularities in your data that are not actually present.