Forum Discussion
Relationship issue with Measure
Hi Experts
I have been looking at this problem for 3 days and cannot see why my table to the right does not show the same p-value for each of the month periods.
Scenario
If you select product id 72201092 which has a p-value of 0.013345 when you look at the table to the right for each period (date) the p-vlaue os 1.0000 when it should be the same for each data point.
Sample file. https://www.dropbox.com/s/jj62h2vxtepm5yd/Sample__Test%281%29.pbix?dl=0
it seems that there is no relationship between meausre and month field.
New Table
My Table = VAR OuterComplaints = COUNTROWS ( PMS_COMPLAINT ) RETURN ADDCOLUMNS ( VALUES ( PMS_FINANCIAL_PDS[Month Start] ), "CountComplaints", CALCULATE ( COUNTROWS ( PMS_COMPLAINT ) ), "Index", RANKX ( PMS_FINANCIAL_PDS, PMS_FINANCIAL_PDS[Month Start],,, DENSE ), "Outer Complaints", OuterComplaints )Something along these lines...
13 Replies
Hi Anonymous,
Nice dashboard and some advanced (to me) stats! I've tagged this with #statistics to get the attention of the stats geniuses.
When I click each month bar below, I notice that the p-value is 1. Wouldn't this be correct? That is, the correlation between month 10 and the dataset of month 10 only would be one.
Your data model and relationships look pretty good. I have a couple of suggestions.
- Your calendar table is monthly not daily so it can't be marked as a date table. Be sure to turn off auto date time for this report in settings (Options and Settings: options: Current File: Time Intelligence: Auto date/time). Doing this will reduce your file size.
- You have multiple transaction tables/fact tables connected to a shared lookup or dimension table with bi-directional filtering, which can create ambiguity. For complaint description, it's 1-1 but you can actually set it as 1-many or many-1 (put the many on the complaint table side). Not sure why you set Parts to bi-directional. If it's to cross-filter slicers, then there's a better way.
organized data model
Hi Freder!ck
When I click each month bar below, I notice that the p-value is 1. Wouldn't this be correct? - NO should be the value in the table for each data point
That is, the correlation between month 10 and the dataset of month 10 only would be one. - As above
Your data model and relationships look pretty good. I have a couple of suggestions.
- Your calendar table is monthly not daily so it can't be marked as a date table. Be sure to turn off auto date time for this report in settings (Options and Settings: options: Current File: Time Intelligence: Auto date/time). Doing this will reduce your file size. - Done may thanks
- You have multiple transaction tables/fact tables connected to a shared lookup or dimension table with bi-directional filtering, which can create ambiguity. For complaint description, it's 1-1 but you can actually set it as 1-many or many-1 (put the many on the complaint table side). Not sure why you set Parts to bi-directional. If it's to cross-filter slicers, then there's a better way. - You are correct on the bi-direction realtion ship
one quick question how do you tag a question to #statistics for future reference
based on your excellent feedback, if you take from PMS_TM1_Sales_vol Table column (month_date) and add that to the table it give you the right p-value but errors on the inidivdual data points shows 846 as a constant number for 12 periods.
As opposed to the actual value (see image below) - fixed one problem another one comes along.
Haven't been able to look at this a whole lot the past couple of days. The problem of a wrong value in a total is generally the result of evaluating a measure with no filters. The fix is either to do a SUMX of the measure over the distinct months, or to have an IF statement that evaluates the measure differently if there are no filters.
here is a more up to date file, replacing the older file from yesterday.
https://www.dropbox.com/s/w4p1z5ueibwrwxd/Sample__Test.%281%29.pbix?dl=0