Performance analytics - Data from last 30 days

[Blair5]

I'm creating an index of several indicators. The data is for over the last 30 days. I created all of the indicator sources to look at the last 30 days and found a post (can't find it now) stating that indicator sources should almost always collect for 'opened on today'. So I recreated a few of the indicators to test this out and am getting different results. What is the proper recommendation? Since this in development, I want to make sure it's done correctly. Don't want a year to go by and find out it was set up incorrectly. See example of differences in attached.

[Josh Cooper]

As Adam stated - what I would do is take your regular daily indicator and use it in a formula with a time series applied.   I.e. Create a formula for "Incidents created in the last 30 days" with the formula set to "Number of new incidents" and pick "30 day Sum" for the time series.

If you can't get exactly what you want there, you could set it up as an indicator source, and have more control over the specific time series, filters, etc.   The drawback is that PA uses the Indicator Source to build a temp table, so depending on how large the table is, 30 days of records may be quite large, which may slow down your data collection.

[Uncle Rob]

Is there a big advantage to that over collecting a daily indicator and doing a 7 or 30 day running sum / avg?

[Adam Stout]

If you collect the scores for 7 and 30 days, you'll be duplicating the same data and can't recombine it later (for instance if you want a 28 day average in the future).   If you do daily, you can aggregate as desired at any time in the future.

Collecting the 30 days will also take longer for the collection job as there is 30 times the amount of data.   It may not be material, but it does take longer.   For a larger instance, but maybe an issue.

[Uncle Rob]

I guess I'm just looking for the use case where a pre-aggregated 7 / 30 day indicator would be superior to a daily indicator that's just using the 7 / 30 day time series.   Does that make sense?

[Adam Stout]

I would normally lean toward the daily with the time series set to do the aggregation (that is why the time series exist).   But there maybe use cases for the doing the rolling collection each day.   The biggest expense is the job run time, if that is not an issue you can do it.   I would recommend not collecting records since that will be redundant.

What do you see as the downside of using the time series?   I'm not sure of why we wouldn't use them.