1

Which R package is most helpful for irregular chrono time series? I see Lubridate and xts used in Regular analysis over irregular time series

I understand that this is an open ended question and would appreciate insight into your choice. Thanks.

Community
  • 1
  • 1
user592419
  • 4,853
  • 8
  • 40
  • 65
  • 5
    You probably don't need to look further than `zoo` and `xts` – Andrie Nov 11 '11 at 10:04
  • What sort of analysis do you want to do on your irregular time series? There are packages that can fit models in continuous time, dynamic linear models and state space models can accommodate missing data or even, I am told, totally irregular data if parametrised correctly. Or you could fit ordinary regression models and correct for the lack of independence late (like sandwich estimator of covariance matrix). – Gavin Simpson Nov 11 '11 at 10:18
  • By 'Irregular Time-Series' do you mean quasi-periodic or stochastic? I am not very familiar with R, but from a theoretical stand point you will want to take two *combined* approaches. i. Establish a correlation time-scale associated with your time-series [a time-scale at with noise is dominated by physical processes]. ii. Establish the nature of your physical fluctuation - for a "irregular time-series" where I assume a changing amplitude as well as frequency you will want to veer away from FFTs. In this case look at Wavelet analysis. – MoonKnight Nov 11 '11 at 10:30
  • 2
    @Andrie **zoo** and **xts** just provide classes for handling irregular data. If the OP only wants to compute moving window stats etc., then I agree, these packages would be more than sufficient. If the analysis is more involved, say fitting a time series model then **zoo** and **xts** may not be suitable/required. It really does depend on what the OP wants. – Gavin Simpson Nov 11 '11 at 13:46
  • @GavinSimpson Really good observation. I found your comment most helpful, thank you. – Andrie Nov 11 '11 at 13:50
  • 1
    @Killercam: I *think* (but am not sure) that the OP is referring to irregularly **sampled** time series. At least, that's what the linked packages/posts are referring to. – Ben Bolker Nov 11 '11 at 23:33
  • Sorry for taking a while to respond guys. Imagine something like web views where you have the id of the user and the time they viewed it. The data is similar to that model. – user592419 Nov 12 '11 at 20:35

1 Answers1

3

You've not given much insight into your needs, so the best CRAN page to investigate is the Task View for Time Series Analysis.

Iterator
  • 19,943
  • 12
  • 71
  • 109