Hi Team,

Here's a bit of a head-scratcher...

I've got a measure that shows a downward trend when comparing month 04 against month 03. Strangely enough, the aggregated values are exactly the same, so I would have expected a grey dash for the trend instead.

The measure has eight records for one week in month 4 and two records for one week in month 3, like so:

Week   Month   Actual   Target  % of Target   # of Records
10     03      100%     90%     111.1%        2
11     03      .        .       .
12     03      .        .       .
13     03      .        .       .
14     04      100%     90%     111.1%        8
15     04      .        .       .
16     04      .        .       .
17     04      .        .       .

When running the Measures - Compare to Previous report and choosing month 04 from the Time Dimension tree, I see:

             Current % of Tgt    Current Actual    Current Tgt    Previous % of Tgt    Previous Actual    Previous Tgt  Variance  Trend
Measure Name  [111.1%]             100.00%           90.00%         [111.1%]             100.00%           90.00%         0.0%     down

Any idea why this is a downward trend? Even if it had shown an upward trend instead of a grey dash, I'd have bought it. 8 records at 100% is a "stronger" 100% than 2 records at 100%, right?

Please advise.

Joey

This message has been edited. Last edited by: Moogle,

Would you mind copying your two records from Month03 and eight records from Month04?

Just make sure we aren't miss-understunding

thnx Endre

Also, would you mind letting us know what is your time aggregation level set to from the measure loader?

Could PMF be comparing week 11 to week 10, instead of month 03 to Month 04?

Seth: It's set to a weekly time aggregation in the measure loader, and a default time level of weekly. The report I'm running is the prior vs current multi-measure report, but I've over-ridden the time level by clicking on month 04.

Endre: I'm not sure what you want me to do. do you want me to paste the source table records into the forum?

Joey,

Yes, if you could paste the detail data for the two weeks you menstioned.
If I understand this correctly, we should see 2 records for week 10 and 8 records for week 14.

thnx Endre

Hi Endre,

Here are the records from the Measures table in the PMF datamart:

 
MEASURE_ID	MEASURE_SERIES	DIMENSION01_ID	DIMENSION02_ID	ACTUAL	ACTUAL_NUM ACTUAL_DEN TARGET TARGET_NUM TARGET_DEN
11444085	         56	         230329	         145947	         100	1	  1	    0	  0.9	    1
11444086	         56	         230334	         229745	         100	1          1	    0	  0.9	    1
11444087	         56	         230334	         229750	         100	1          1	    0	  0.9	    1
11444088	         56	         230334	         229751	         100	1          1	    0	  0.9	    1
11444089	         56	         230334	         229752	         100	1          1	    0	  0.9	    1
11444090	         56	         230329	         146663	         100	1          1	    0	  0.9	    1
11444091	         56	         230334	         229786	         100	1          1	    0	  0.9	    1
11444092	         56	         230334	         229790	         100	1          1	    0	  0.9	    1
11444093	         56	         230334	         229793	         100	1          1	    0	  0.9	    1
11444094	         56	         230334	         229794	         100	1          1	    0	  0.9	    1

Here are the records when you do a preview in the measure loader:

Hub_1	Time	New Actual	Curr Actual	New Benchmark	Curr Benchmark	New Forecast	Curr Forecast	New Stretch Target	Curr Stretch Target	New Target	Curr Target
G.U.S.A./KNITS/141/037-54-1030	2009q01Aprw10	100	0	0	0	0	0	0	0	90	0
G.U.S.A./KNITS/141/037-54-1035	2009q02Mayw14	100	0	0	0	0	0	0	0	90	0
G.U.S.A./KNITS/141/161-54-1021	2009q02Mayw14	100	0	0	0	0	0	0	0	90	0
G.U.S.A./KNITS/141/161-54-1056	2009q02Mayw14	100	0	0	0	0	0	0	0	90	0
G.U.S.A./KNITS/141/161-54-1056T	2009q02Mayw14	100	0	0	0	0	0	0	0	90	0
G./KNITS/141/037-54-1030    	2009q01Aprw10	100	0	0	0	0	0	0	0	90	0
G./KNITS/141/037-54-1035	         2009q02Mayw14	100	0	0	0	0	0	0	0	90	0
G./KNITS/141/161-54-1021	         2009q02Mayw14	100	0	0	0	0	0	0	0	90	0
G./KNITS/141/161-54-1056	         2009q02Mayw14	100	0	0	0	0	0	0	0	90	0
G./KNITS/141/161-54-1056T	         2009q02Mayw14	100	0	0	0	0	0	0	0	90	0

Does anyone have any suggestions about this? I'm still curious and would like to be able to tell the client something by early next week.

Cheers,

Joey

My guess: Sounds like the time aggregation level used for trend for this particular report is hard coded to be at the aggregation level set in the measure loader.

To test, change the aggregation level in the loader to monthly. Reload. Then see if the monthly trend works. If it works after this change then it sounds like there may be bug in the Measures - Compare to Previous report -- OR your customizations to the report may be overriding the time aggregation level.

In your custom fex do you have a line like:

-SET &SORTACROSS='TIME_LEVEL03_VALUE';

If so, then you will override the time sort level regardless of whether you click on the month.

This message has been edited. Last edited by: SethW,

Nicely figured out, Seth!

So what can a poor PMF dude or dudette do to make the ACROSS sort more "soft" but still specify it should be Time?

  
-SET &SORTACROSS = 'TIME'; 

I do believe this will allow the time level to be set by a passed Time Range, or the context of a Time Dimension value (e.g., if the context is a Month, it's monthly, for example.) But Eric might jump on and correct me...

thanks

One other thing. The trend calculation code is by default intolerant of any deviation, even a 0.01 deviation will show up as up or down. It was assumed customers wanted absolute mathmatical accuracy. However, you can expand the calculation's range by changing a line of code in a_prevcurr2_i1.fex:

-DEFAULTH &PCT_RCHD_FLAT_&TARGET_FIELD_NAME.EVAL = 0.0

If you change the 0.0 at the right side of this setting to a value, PMF will expand its tolerance for non-zero disagreements. For example if you changed it to

-DEFAULTH &PCT_RCHD_FLAT_&TARGET_FIELD_NAME.EVAL = .1

PMF would see a variance of .1 or less to be equivalent to zero and would show a flat tolerance.

Be careful setting this...

Should we make this a setting (e.g., "Call variances of this amount or less 'flat'?" or some such?)

thanks

quote:

-DEFAULTH &PCT_RCHD_FLAT_&TARGET_FIELD_NAME.EVAL = .1

Hi Bob,

This worked very well to correct this issue I was experiencing. I don't know if this needs to be a global setting, because it has such a subtle impact. I only found the issue on one Measure on one report for one time period & time level.

By the way, I did try to load the Measure at a monthly level instead of weekly, but the problem persisted.

Thanks to Seth and Endre as well.

Cheers,

Joey