Time series analyses (stats) - anyone else doing them?

#1

Hey everyone -

I tried searching the forums here but didn’t find much - is anyone else doing time series analyses? I need to do stats, but here’s what I’m running into (and hence am curious how you’re doing it):

  1. If I do it at the dashboard level, I need tons of columns - one for the moving average, then one to copy the last MA into the current row, then one for the percent difference between those MAs, then another for … it goes on and on. I can’t do that for a dozen metrics. ALSO, at this level, I can’t hide the historical data (30+ days), which I only need to do the aforementioned calculations.

  2. If I do it at the data cube level, I can get all the above done really easily, but then I can’t aggregate on the dashboard level. That is, when you use the expanders on a dashboard, the choices are to sum, avg, etc. to aggregate, and none of those are mathematically accurate, so what’s shown is incorrect (you can’t avg or sum percent changes).

I’m not sure where to go here. I spent 2 days playing with every option I could find, but it didn’t seem to work right. Can anyone share how they do time-based stats or ideas as to how to tackle mine?

#2

Hey Ken,

This is an interesting topic. I am interested in what types of hints Dundas might provide. Can you tell me about what type of time series you are conducting? Reason I ask is I do a lot of time series work, particularly in forecasting, but it is all in R, C# and SQL.

I have wanted to move some of it over to Dundas but I haven’t gotten around to it just yet. With the flexibility of adding python code to your data you might be able to add some time series modules to your data, but to be completely honest, I haven’t done much leg work in it yet.

1 Like
#3

I don’t know the specifics of the analysis, but indeed formulas can and should be chained together to get to the final result.

In case you want to place a value from a previous row on the current row, you can use a placeholder that points to the specific measure you need, with an offset value of -1.

You can actually access offset or aggregation properties by clicking on the placeholder itself, after it has been defined as a Calculated Measure.

1 Like
#4

Regarding using formula placeholders with offset, you may find the relevant documentation here:
https://www.dundas.com/support/learning/documentation/analyze-data/formulas/adding-formulas#h3-2-placeholders

#5

Hi Jay, thanks for the reply, but you cannot have a placeholder with more than 1 offset in a formula. That is, I cannot pull in $date$ with a (-1) offset and $date$ with a (0) offset to get percent change.

#6

@ken Glad to help. Regarding using different offset, you may create multiple formulas. For example, create one with offset, then call this formula inside another formula.

#7

You can use two placeholders for the same measure, each one with a different offset value.

#8

no, unfortunately you can’t, Adrian. If you change the offset on one of the (identical) placeholders, the other one changes, too.

#9

@jay.gong, yes, I understand that’s possible, but with a dozen columns to do this to, it will quickly get unwieldy and impractical.