500 from lineplot when a proportion value is unknowingly filtered out

[bobular]

GEMS1
Line plot
x-axis: Age group
y-axis: Case or Control case/case+control
overlay: Lost earnings due to care

or Borrowing, or is seems any variable uner Current care expenses.

What does work?

As above but with (three scenarios listed below)

• y-axis: numeric variable, like Age.
• y-axis: Participant->Sex female/male+female
• overlay: Participant->Sex

It's not obvious to me what the pattern is!
I couldn't get a similar 500 in other visualizations using the Lost earnings due to care overlay

[danicahelb]

overlaying "Lost earnings due to care" works when a different y-axis variable is selected. See:

[danicahelb]

If I add a filter to only include data where lost earnings due to care = yes, then the plot is empty. I would expect the plot to still have the same blue line

[danicahelb]

I can also see that only cases have data remaining when i filter by "Lost earnings due to care" = yes. maybe this has something to due with the error? (Actually, I can see from the definition that "Lost earnings due to care" was only assessed for cases)

[danicahelb]

I also get the error message when i add "Sunken eyes" as my overlay. Sunken eyes was also only assessed for cases

[dmfalke]

Looks like this is probably a backend issue, so moving it to EdaDataService

[asizemore]

Error is in plot.data. Moving there!

[asizemore]

Pushed to b58. Issue is that there is no overlay data for Control participants, which means the proportion is basically Case/Case. Our plan is to still err, but return a nice error to the user. Will require work from front and back end.

[danicahelb]

link to 4/15 convo in Slack #eda-map-client-imp
https://epvb.slack.com/archives/C012ZK4Q5CZ/p1650035476658289

[danicahelb]

@asizemore this is the ticket we discussed in 6/1 data viz meeting.
we've decided that -- for proportions -- we will allow numerator=denominator and we’ll return all 1s.

[asizemore]

@danicahelb the issue in this ticket is that the 500 is being thrown not because the num and denom are the same but because there are no samples left with the value of Control. Effectively the lack of data makes it Case/Case, but the user did not choose that. It's a very rare situation. Still, feels like we should somehow alert the user that they subsetted away an entire portion of the denominator? Without an alert, the user would just see a line plot with 1s and nulls and not necessarily know why.

In general being able to select the same num and denominator (for example in the screenshot below) is a different issue. I think the below is more what we talked about at the data viz meeting last week and i can implement a fix if it's a must have for b58.

[danicahelb]

Hey @asizemore
as we discussed in the 6/13 EDA UX meeting, if the user filters out data that effectively causes the numerator and denominator to be the same, please do not include an error message and allow the y-axis proportion to equal 1 (so there will be a straight line at y=1, which will give the user a clue that something is up).

Lets start here for b58, if we get feedback that it is confusing we can address this confusion later. thanks!

[asizemore]

Sounds good @danicahelb . This starting point should be addressed by the linked PR.

[asizemore]

Deployed to qa and feat. moving to ready for testing

[sweverschulman]

Functionality that when user makes the proportion = 1 to no throw an error but to plot y = 1 seems to be working. Functionality was also tested with PERCH study.

[moontrip]

@sweverschulman Thank you for your tests and confirmations! 👍