Hi,
Not sure whether anyone can help, but I've encountered some problems with my Interrupted Time Series Analysis. I originally applied a Dynamic Linear Model to the data, from which I hoped to be able to create a reliable counterfactual scenario. Whilst this worked for some of my data it has failed in regards to other parts.
I can undertake a similar predictive step with an ARIMA, but also wondered how such a model would ordinarily be used in ITS analysis? In the example of general regression models as ITS can be represented as either the difference in slope and intercept of the two segmented time series or the difference between the predicted counterfactual and modelled observations.
Has anyone used ARIMA for such?
And if so.......
Is it possible to quantify an intervention effect based without the counterfactual (i.e. replicating the general regression model outputs)?
Any help would be hugely appreciated!
Your question is rather vague. One approach that you may find helpful depending on your interuption (i.e. research question) is described in the attached. HTH David Booth
I agree with David Eugene Booth that your question is vague. Probably intervention analysis of Box and Tiao (1975) can help. But the types of interventions (additive outlier, level shift, temporary change, innovational outlier) is perhaps too limited. A very long time ago (1981), I proposed in a conference book, see https://www.researchgate.net/publication/241768202_On_an_alternative_model_for_intervention_analysis, a more efficient implementation. As far as I know it has not been implemented elsewhere than in my own software package Time Series Expert.
Thanks for the answers both David and Guy. Really useful to have some alternatives to look at. I've sort of reached a point of no hope with it all! Thanks again
John Rahilly I really liked Guy Mélard suggestion. We have done some work exactly like he suggested for AR1 models you may wish to look at some of our papers, see: https://scholar.google.com/scholar?hl=en&as_sdt=0%2C36&q=Booth+Wright+Hu+outliers+short+time+series&btnG=
best of luck, David Booth
John, can you clarify what you call an interrupted time series. If it can be considered as a time series with missing data, there are other possibilities. See e.g. the paper by Gómez et al. (1992) available on ResearchGate: Article Computing Missing Values in Time Series
. I should add that their method is implemented in Tramo-Seats and in JDemetra. I presented the included poster with mini-paper on high-frequency time series, which includes the problem of missing data, of course, where you can find a short bibliography. I should have a presentation focusing on software aspects but I need to find it. Anyway it is work in progress.
Please check out this paper - Article Interrupted time series analysis using autoregressive integr...
It has shown the use of ARIMA model in ITS analysis with specific details.
I hope it helps.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size