Design Exercise 3: Stories from Census Data

November 4, 2026 (Last Modified: November 5, 2026)

This design exercise has you create visualizations from the new census data set. You will turn it in on Canvas as Design Exercise 3: Census (aggregations) (due Fri, Nov 13).

Preliminary version: the questions are ready so that you can begin work. The details of grading is not yet ready. The Canvas hand-in form is not yet ready. Details of the feedback activity and grading are subject to change.

At A Glance

This assignment asks you to make 4 visualizations from the census data set (the new census-composite version). Part A is a question we provide (where do migrants go?). In Part B you pick a question and answer it with two different designs, then compare them. Part C is a second question of your own.

You will turn the assignment in on Canvas at: Design Exercise 3: Census (aggregations).

On-time submissions (by 12:15pm, Friday, November 13) will have the opportunity to participate in a feedback activity. Students will be able to give and received critiques over the weekend, and revise their assignment by the cutoff(11:59pm, Wednesday, November 18). Students may earn extra points by providing critiques and by explaining their revisions.

Assignments will be rewarded for the effectiveness and quality of the visualizations and the quality of the questions/stories that these visualizations tell.

This assignment asks you to create visualizations that “tell good stories” (that is, help the viewer with some understanding task) from the census data that you worked with in Design Warmup 3: Working with Census Data. The data, and how to get it, are described at the census data set. Read the data’s README: it explains which years each variable covers and how to combine counties correctly.

As with DE2, “interesting” is subjective, and part of “interesting” is “leads to an interesting visualization.” Our goal is to make visualizations, not necessarily to do good social science: document what you observe, rather than trying to draw causal conclusions. What are we looking for? gives a sense of what we are looking for.

You will turn the assignment in on Canvas at: Design Exercise 3: Census (aggregations). Each question is listed below, in the description of the part it belongs to. You should prepare your answers (and images) offline.

Some Ground Rules…

You need to create static visualizations: things that work in “print” (or just as an image file). If you create something interactive, we won’t see the interactivity.
Each visualization must be a single “picture” (one page, preferably PNG; a PDF can only be 1 page). You can put multiple charts together into a single image, but it must fit on a page/screen.
You may use whatever tools you like. Maps are allowed but never required; the region and division columns give you a geographic grouping without a map.
We aren’t doing the actual social science here: these graphs are descriptive. We are showing (at most) correlation, not causality.
Please don’t reuse questions we have already posed (including the ones in Design Warmup 3: Working with Census Data). You may use the questions you posed in your answer to DW3.
You should not add data. You can use knowledge you have (for specific things, like events), but don’t augment the data by joining it with an outside source. One exception: you can use geographic information (for example, to make a map).
We are not asking for your intermediate files (programs, Tableau workbooks), but we do ask you to explain your process. We may ask you to show your work later, so please keep them.

AI use on this assignment

You may use AI tools to help you, for example to write or debug code, or to walk you through the steps in a tool. You should be the one choosing the questions and deciding how the visualizations show them. Tell us how you used AI in the Summary Questions. The course policies allow for a range of AI uses, provided that the use is disclosed. We appreciate clear, specific answers: they help us understand how students are using AI.

AI Policy Details for Design Warmups and Design Exercises

I am giving you (students) broad permission to use AI tools/agents, and relying on your judgment to use it in a responsible way that does not get in the way of your learning. Rather than trying to specify what is or isn’t acceptable, I ask that students use their judgment, honestly disclose their usage, and reflect on it.

The broad range of AI use makes a simple policy hard. When you create visualizations, AI tools can be anything from an assistant on using visualization tools to an automated approach to creating visualizations. You might try everything from “make the visualization specified in this assignment” to “tell me which button to push in Tableau to change the marks from red to purple.”

The AI tools can also do other aspects of data analysis. You can ask an AI agent “make 5 different visualizations from this data set showing different aspects of it, and explain what each one shows.” Or even “find something unusual in the data.”

For me, there is no clear line between “AI does it for you” and “AI is helping you do it by taking care of the boring bits.” So, I ask you to use your judgment: use AI tools in a manner that fits the spirit of the assignment, where you are getting the experience of designing the visualizations and practice at making the decisions and thinking about how to effectively show the things that you want to show.

Having AI do too much for you gets in the way of you experiencing the designing and creating, which I think is essential for learning.

Aside: Learning about AI tools

The goal of this class is for you to learn about visualization. You might use AI tools to help you make visualizations, to design them, or even to learn about them.

There is a different goal: learning about AI tools. Rather than learning about visualization, the goal could be to learn how to use AI tools to efficiently do work (for example, assignments asked of students). This is not the goal of this class.

Asking your AI agent to write programs to process data, or even to help write programs to make visualizations you describe, is generally acceptable.

My goal is that you should learn about visualization. You can actually learn a lot by using an AI agent and, when it does something for you, asking why.

You should be involved in the choices of how visualizations are created. Generally, this means making the choices of which environments/tools are used (e.g., to use Python and a particular library). However, you might ask an AI “We are going to make some (describe visualizations) by writing Python programs. What visualization library might you recommend?” (I use Python as an example: AI agents seem to know about many languages and libraries. They are also good at reading the tutorials for interactive systems like Tableau.)

So, I am going to trust students to use AI responsibly. This means following the main rules AI Policies. But some specifics:

I will explicitly ask if and how you used AI tools for the assignment. Please be specific about what tools you used and how you used them.
I would like your prompts (there will be a separate question for that). Please provide them as much as possible. This is mainly for me to learn from.
I will ask “What did you do to make sure that the AI tools did not get in the way of your learning.”

If the goal is to learn about visualization, trying to one-shot prompt to get an AI agent to do the assignments is unlikely to give you the experiences that help you learn about visualization design. If the goal is to see if current AI models are capable of doing these assignments, trying the prompts yourself is a waste of your tokens: the course staff has tried it, and has confirmed that the AI tools are good at it. You won’t learn much by doing it.

Further clarifications to the AI policy are coming soon.

Explaining your designs

For each visualization, we ask what it makes easy to see and what in your design makes it easy. Name what you did, not a general principle. “I highlighted the Wisconsin counties in orange so they stand out from the rest” is what we’re after; “position is the most accurate channel” isn’t. For some, we also ask about your most important or unusual design decision. A few sentences is usually enough; longer isn’t better. Explanations earn credit for what we can see in your chart: an unusual choice whose effect is visible is great; a claim we can’t see in the chart doesn’t count.

We value clear and concise answers. Excessively long answers may earn fewer points.

Your visualizations should stand alone. A viewer should be able to figure out what your intended story is. Hint: titles, captions, annotations, and design that focuses the viewer’s attention on the story can help.

Ask the course staff if anything is uncertain.

Part A: Where Do Migrants Go? (Staff-Provided Question)

Do people who move to the US from abroad (international migrants) go to the same kinds of places as people who move within the US (domestic migrants)?

“The same kinds of places” can mean many things: the same parts of the country; places that are similarly urban or rural; places with similar jobs, incomes, education levels, or politics. Pick the sense you think tells the most interesting story and make one visualization that answers the question in that sense.

Question 1: Your visualization.

Question 2: Which sense of “the same kinds of places” did you choose, and why? What is your answer? (2–4 sentences)

Question 3: What is the main thing your visualization makes easy to see, and what specifically in your design makes it easy? (2–4 sentences)

Question 4: What was your most important or most unusual design decision, and what alternative did you consider? What would a reader lose with the alternative? (If it’s the same thing you described above, focus on the alternative.) (2–4 sentences)

Part B: One Story, Two Designs (Your Choice)

Pick a question (or story) about the census data, and answer it with two visualizations that tell the same story in meaningfully different designs: not the same chart restyled, but a different approach (for example, a different chart type, a different aggregation, or a different arrangement). Put the one you think is better first.

Question 5: Your question or story. (1–2 sentences)

Question 6: Your first (preferred) visualization.

Question 7: What is the main thing your visualization makes easy to see, and what specifically in your design makes it easy? (2–4 sentences)

Question 8: Your alternate visualization, telling the same story with a different design.

Question 9: Compare the two designs. Does the alternate design make the main story harder (or easier) to see? Does it make different stories easy to see? Be sure to describe specific aspects of the design to support these claims. (a few sentences)

Part C: A Second Story (Your Choice)

Pick a different question or story from Part B (not the same finding shown another way), and make one visualization that answers it. We value variety in design: a different approach from other parts shows more range.

Question 10: Your question or story. (1–2 sentences)

Question 11: Your visualization.

Question 12: What is the main thing your visualization makes easy to see, and what specifically in your design makes it easy? (2–4 sentences)

Question 13: What was your most important or most unusual design decision, and what alternative did you consider? What would a reader lose with the alternative? (If it’s the same thing you described above, focus on the alternative.) (2–4 sentences)

Summary Questions

Question 14: How did you make these? What produced your final images? (For example “Tableau”, “my own Python with Matplotlib”, “Matplotlib code Claude wrote from my description”, or “Claude Code / Codex wrote and ran the code”.) If you changed the data beyond loading it (for example, combining counties or computing new variables), say how in a sentence or two. If you used any grouping or information that isn’t in the data (like a list of which countries are in which region), say where it came from.

Question 15: How did you use AI on this assignment? (The course policy allows a range of AI uses as long as they’re disclosed. We’re asking to learn how students work with AI, so clear, specific answers help us most.) If you didn’t use AI, say so and skip Question 16. Which parts (or question numbers) did it touch, and what did it do? (For example: wrote or debugged code, gave step-by-step help in a tool like Tableau, suggested questions, stories or designs, explored the data, looked things up, drafted or polished text, critiqued your work.) Which ideas, stories, or design choices came from the AI, and what did you do with them?

Question 16: (If you used AI) Share two or three prompts, copied as you typed them (any language is fine). Include the one that did the most work. For that one, say in a sentence or two whether it got you what you wanted, and why or why not. A full log file or share link is welcome but not required. (If you did not use AI) Say so, and describe in a sentence or two what you used instead when you needed help (for example, documentation, tutorials, examples from class, classmates, or trial and error).

Question 17: Something that didn’t go as planned. Pick one and describe it in a few sentences: (a) a design you tried and dropped, and why; (b) something in the data that surprised you or that you had to fix; (c) a choice you went back and forth on; (d) something an AI gave you that was wrong or misleading, and how you noticed; or (e) a time you overrode, questioned, or learned something from an AI suggestion. (A design that didn’t work is just as good an answer as one that did. We’re interested in what you noticed and what you did about it.)

The Defend Your Designs Phase

After the final cutoff, we will have a “defend your design” phase, where we will ask you questions about your designs. The exact mechanisms, and how it will count towards the grade, is still to be determined. We will announce the mechanism after the final cutoff for the assignment.

The Feedback Exercise

(Note: based on the results of the survey, we may make some adjustments to the timing)

If you turn in a complete assignment on time (by 12:15pm Friday)…

This deadline is strict: on Friday afternoon the course staff will assign you visualizations to critique. (exact mechanism to be specified). You will earn 1 point for each visualization that you provide “useful” critique for (you can only earn for up to 2 critiques). To be considered useful, the critique must:

be a good critique (i.e. the course staff thinks will meaningfully help the designer).
be timely (provided before end of day, 11:59pm Sunday, Nov 15), so the designer has time to consider it,
must not be redundant with other feedback provided. You can either be the first to provide feedback on a design, or you can provide different feedback that is useful beyond what others have already given.

The feedback mechanism will be a Canvas discussion so that you can see feedback to/from others.

Course staff will also contribute critiques to the discussions.

The idea is that you should be able to learn from doing the critiques, from critiques of your work, but also seeing the critiques to others.

If you turn things in on time, you may re-submit your assignment. When you resubmit, you also answer Questions 18-20 below. You will get 1-2 points for answering these well. Note: even if you decide your assignment is sufficient, you can say “I didn’t change anything” (and give a few sentences on what you learned from reading and writing the critiques) and still earn 1 point (providing you really didn’t change anything). The second point is for actually improving your assignment.

In total, the feedback exercise is an opportunity to improve your visualizations, and earn up to 4 points (up to 2 points for critiques, 1 points for completing the resubmission question, and 1 point for a good answer to it).

Resubmitting

If you turned in on time, you can resubmit as many times as you like before the cutoff. You don’t need to redo the whole assignment. We grade the most recent answer you gave to each question. In a resubmission, answer only the questions you changed, and leave the others blank. For the questions you leave blank, we use your most recent earlier answer.

A new answer replaces your old one entirely. If you revise a text answer, give the whole revised answer, not just the changes.
To change an image, upload the new version. If you don’t upload one, we keep the one you submitted before.
Leaving a question blank keeps your old answer. You can’t remove an answer by leaving it blank.
Questions 18-20 work the same way. If you resubmit more than once, your most recent answers to 18-20 replace the earlier ones. So they should cover all of your changes, not just the latest round.
If you change a visualization, check that its rationale still describes it. If not, resubmit the rationale too.

Re-uploading everything still works.

When you resubmit, you should answer the following three questions:

Question 18: For each change: which question(s) changed (image, text, or both), what you changed, what prompted it (a specific critique, something you saw in others’ work or class discussion, or your own second look), and what it fixes. If you changed nothing, explain why you decided your original was sufficient. If time kept you from making a change you wanted, say what it was in a sentence; that’s useful for us to know.

Question 19: Describe one judgment call you made about feedback. Pick one: (a) a suggestion you decided against, and why; (b) a suggestion you didn’t fully understand or agree with, and what you did with it; (c) a suggestion you acted on, but differently than the reviewer proposed; or (d) if you didn’t get much useful feedback, a problem you found on your own second look.

What are we looking for?

A hint from a past version of the class:

… we are looking for explicit choices you made to emphasize your “story.” A “data dump” (just making a chart of some of the variables) is not likely to get you a good score. If you make explicit decisions - selecting subsets of data, highlighting particular points, arranging designs that emphasize certain aspects, etc. - this will be rewarded.

Some notes on “A Good Question”

For the purposes of this exercise, a “Good Question” has a non-standard (but still hand-wavy) definition: it will make for a good visualization. A good question/answer should be:

Interesting (non-obvious; ideally something the course staff hasn’t already seen; uniqueness is valued)
Data rich
Taking advantage of visualization (generally, the answer is complex; or a simple answer where the visualization gives context, supports it, or helps with follow-up questions)
Answerable from the data (you could actually make a good visualization to answer it)

For this assignment, what we are looking for in a “good visualization” includes:

Is the question/story interesting and clear?
Is the design effective (well adapted to the story/task)?
Do the details represent good choices (including color)?
Is the design appropriate for the data, including how the data is aggregated (counties, people, or rates; weighted or not)?
Do your explanations match what we can see in the chart?
Is the design complete (it has enough of a title and caption that it stands alone)?

Of course, you should avoid “bad” practices (for example, inappropriate encodings or aggregations).

Rubrics

We are developing rubrics and assessment mechanisms that reward students who produce effective designs and can stand behind their decisions. More details will be provided closer to the assignment due date.

Part A will be worth 30%, Part B 40%, Part C 30%. Each part will be scored independently.

The scores for Parts A, B, and C will be converted to the 90=A Scale. Other bonuses and penalties are added to that score. For the Summary Questions, insufficient answers may lose points; good answers, or interesting observations, may earn a small bonus. Participation in the feedback exercise will earn additional points (up to 2 points for giving critiques, 1-2 points for describing the improvements).

CS765 course material · AI use: pages.graphics.cs.wisc.edu/765-26/ai/de

GenAI Disclosure:

Claude helped develop this assignment: it made the census data set, checked that the Part A question has interesting answers in the data, and drafted this page.