[Bug]: Time Series Analysis Stationarity tests

[wooyoung32]

JASP Version

0.18.3

Commit ID

No response

JASP Module

Time Series

What analysis are you seeing the problem on?

Stationarity tests

What OS are you seeing the problem on?

macOS Intel

Bug Description

The results of the stationarity test from JASP appear to differ from the results obtained from Python.

Expected Behaviour

These are the results from the Python with the same data set.

Steps to Reproduce

1. Go to Time Series tap
2. Run Stationarity test
   Pennsylvania monthly offences data_2013.csv

...

Log (if any)

No response

Final Checklist

• I have included a screenshot showcasing the issue, if possible.
• I have included a JASP file (zipped) or data file that causes the crash/bug, if applicable.
• I have accurately described the bug, and steps to reproduce it.

[tomtomme]

@wooyoung32
thx for the report.
I can confirm the results you got with jasp for version 0.18.3 and current 0.19 beta.
The python results may differ because of different "truncation lag parameter"? For JASP this is between 5 and 4, while the pythons output shows "lags" of 6 to 13? Just an observation. I have no clue regarding this analyis and maybe truncation lag and lag is something completely different.

But if I am correct, then please do not be offended by the following question:
In JASP I can alter the truncation lag by filtering the data (shorter data = shorter truncation lag). Are you sure the data processed by python was not a longer time span then?

[wooyoung32]

@tomtomme
I've been using the same dataset for both JASP and Python, but I've noticed that the two programs display different test statistics and p-values for all three tests.

When using Python, "lags" refers to the number of lags needed to achieve stationarity through differencing. However, I'm not sure if "lags" means the same thing in JASP, as I couldn't find a detailed explanation in the program. If it has a different meaning in JASP, I would appreciate it if you could clarify it for me.

[tomtomme]

@sophieberkhout
Can you clarify the meaning of "lags" in JASP and how the different results may have occured?

[sophieberkhout]

It seems that R and Python use different default arguments for these functions.

For example, for the ADF test, JASP uses the R function adf.test from the tseries package. The truncation lag parameter represents the number of lags used in the regression and the default value is computed from a suggested upper bound based on the length of the series; you can find the documentation here: https://www.rdocumentation.org/packages/tseries/versions/0.10-56/topics/adf.test.

Python uses a different method to obtain the number of lags. Python should provide similar results once you set the maxlags argument to 5, the autolags argument to None, and the regression argument to "ct" (see also the answer in this discussion: https://stackoverflow.com/questions/78177755/why-augmented-dickey-fuller-results-are-different-in-r-and-python).

I could add an option to manually pick the truncation lag parameter, although I am not sure how desirable that is as both packages seem to use an algorithm for this.