Allow the user to create bar charts in the DAD

[jeancochrane]

Overview

Closes #96, closes #97

Add logic for displaying bar charts for surveys in the data analysis designer.

Demo

Notes

There's officially enough data processing logic in the DAD (particularly in charts.js) that it would make sense for us to have tests. I didn't want to have to add the task of configuring a whole JavaScript test runner to the scope of this PR, so I opened up #131 instead.

Testing Instructions

• Make sure you have surveys with questions and responses of each question type:
  • A custom Number (e.g. "How many fingers do you have?")
  • A preset observational question (e.g. gender_observational)
  • A preset intercept question with dropdown responses (e.g. "What is your gender?")
  • A preset intercept question with free response (e.g. "How old are you?")
• On the Collected data view, try making charts for each question type and confirm that they look right

[beamalsky]

Very fun to play with this, and this PR accomplishes a lot! A couple comments big and small below.

[beamalsky]

I made a float field and submitted 3 values:

Here's the chart it made me:

What I'd expect to see here is a chart with an X axis of response values and a Y axis of how many times that response was submitted. Is there any case where an integer question would show anything other than a bar chart with a single bar? The single median value doesn't do much as a bar chart for me, and would be better displayed as a single number.

[jeancochrane]

The use case I'm thinking of here is specifically one where the user wants to compare the median of a survey to the medians of past surveys (not currently possible, but planned in #102). In that case, the chart would show a series of bars that are comparable.

Showing the count of each response seems dangerous to me since it's basically a free-response question and we could wind up with hundreds of slightly different responses. A variation of that approach might be to bin the data in some way and compare distributions, but then we'd have to know up front what the bins are, or bin them by e.g. quantiles and risking having quantiles that don't make sense if the data are very close together.

This seems like a case where a box plot could be a good compromise: in the single-survey case it would at least give a sense of the distribution of the answers, and in the multiple-survey case it would allow for more detailed comparisons. What do you think?

[beamalsky]

All makes sense! I can't quite picture the UI for #102 yet, but I'm comfortable leaving this question open until you get to that issue. For single surveys a box or scatter plot makes sense to me.

[jeancochrane]

@beamalsky I believe I addressed all your comments -- once tests pass, this should be ready for a second look!

[beamalsky]

Just one small thing:

For this age element, could the percentage have fewer decimals and a % sign at the end?

Otherwise this looks great to me and ready to merge!