I like the idea of using a dynamic Bayesian network to build a causal structure, however am unsure how to tackle time-series data where there is an irregular sampling resolution. Specifically, in a sport scenario where there are 2 teams and samples are events which occur irregularly. Ultimately, I would like to explore causal effects of interventions in this data.
Someone recommended the use of an SSM. To my understanding, when it is discretised, it could be represented as a DAG? Then I have a structure to represent these causal relationships.
Other workflows could be:
this library: https://github.com/jakobrunge/tigramite
using ARIMA to detrend the time-series data then use some sort of Bayesian inference to capture causal effects
using a SSM to create a causal structure and Bayesian inference to capture causal effects
making use of the CausalImpact library
Although I suggested 2 libraries, I like the idea of setting out a proper causal workflow rather than letting a library do everything. This is just so I can understand causal inference better.
I initially came across this interesting paper: https://arxiv.org/pdf/2312.09604 which doesn't seem to work with irregular sampling resolutions.
I'm quite new to causal inference, so any critique or suggestions would be welcome!
Many thanks!
