270733_en-US

liliconcom

best charts for troubleshooting time series events

All, 

I'm developing a workshop to increase the use of JMP as an interactive data tool.  I see lots of folks in my org using it to create useful graphs for ongoing reporting, but I would like to see it used more effectively for troubleshooting and data exploration. We have lots of time series data.  Beyond the usual overlay trends and scatterplot matrices, I like variability plots and parallel plots.  I also plan to talk about lag plots and residual plots for ARIMA models.  Are there other plots you like to use for data exploration of time series?  Thanks.

Re: best charts for troubleshooting time series events

@a_mitlehner without knowing more about what you're looking at and how well it fits the techniques you've chosen, I'd suggest starting with standard residual plots to check that the assumptions underlying the methodology you're using are met. Personally, I find time series to be a bit restrictive, requiring ad-hoc differencing and the application of transfer functions to capture non cyclical changes. If you're working with multiple series, what about using Neural Network, or FDE to find curves that don't fit the general trends or fit the underlying model poorly. You could probably do the same with time series by saving the residuals and looking at them as a function of time or exogenous inputs.

Re: best charts for troubleshooting time series events

A bubble plot might be a good one.

Another one I’d suggest is that if your data is monthly or quarterly X11 decomposition method can be applied to to extract and plot trend and cycles/seasonal components  (see an example below). X11 is available in JMP's Time Series platform. (https://www.jmp.com/support/help/en/15.2/#page/jmp/differencing-and-decomposition.shtml#ww545636)

If you're not familiar with X11 there is a recording https://www.jmp.com/en_us/events/ondemand/mastering-jmp/time-series-analysis-and-forecasting.html