I am looking for a suitable dataset for causal inference in time-series data.

The biggest issue I am facing at this stage is to find a dataset suitable for experiments, in the sense that the causal relationships among the variables are well-known and hence they can serve as a validation for the method I am implementing.

Is there any available open dataset of the above-mentioned sort?

The only thing I have been able to find on the internet is the NCEP reanalysis dataset in which some of the causal relationships are clear, but many others are not.

Furthermore, I tried to construct a synthetic dataset but I am not sure that it will recover the complexity I need for accurately testing my method, do you have any suggestion about it?

  • I might be misunderstanding, but there is a lot of data online, and a lot of it shows time series dependence. Even something as simply as hourly/daily temperature should show a time-value correlation. – user3856 Oct 8 '19 at 15:49
  • The point is: I need a dataset in which causal relations are labeled in some sense (eg. by a physical model) because it seems to me that if I am the one assuming a causal relation (even though it seems to be clear) I will introduce some bias I won't be able to control. Moreover, there is not a precise definition of causality, hence I cannot label the data myself using, for example, correlation. – Symòn Oct 9 '19 at 7:36
  • What are you using to produce the inferred relationship? Knowing what you should get doesn't always lead to introducing bias - I know I've got results that surprised me before. – Lio Elbammalf Oct 9 '19 at 9:31
  • I am trying to combine different aspects of causality (for now, Granger causality, convergent cross mapping and conditional independence test) using a measure of fitness for bipartite graph – Symòn Oct 9 '19 at 9:57
  • Where in the process of, for example, implementing Granger causality do you foresee your bias coming into play? – Lio Elbammalf Oct 9 '19 at 12:42

I found a dataset which seems to be exactly what I was looking for

Causal Effects in Time Series

It provides sales' data for 100 products and 1000 promotions, moreover it is endowed with a 1000 x 100 matrix in which each element (promotion, product) is a number describing the causal relation.

  • please mark this answer as "accepted"! – philshem Jul 6 '20 at 10:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.