Section Assignment 3

Reproduce a figure with your agent

Graded: Pass / No-Pass Due: before your Section 4 meeting (exact deadline on Canvas) Submit: one link on Canvas

Back to the lab

What to do

Pick one figure, then reproduce it with your agent using the same steps as in the lab: describe the figure and its likely source, ask for a download script and a plot script, read the report, and check the result against the original.

Option A · A figure from our sessions

Any figure from a POLISCI 158 session, politics or tech track. Slides and handouts are on the course website. Its data must be public: the figure's source line tells you where the numbers come from. Skip figures built on private data (for example, payroll records), as well as headlines, screenshots and diagrams.

Option B · Today's demo figure

Follow the lab: family wealth by parents' education, from the Federal Reserve's Survey of Consumer Finances, 1989 to the latest survey. Compare your figure with the dry-run figures in the guide.

Option A: not sure which figure? Some that work

Session 2: productivity vs. typical pay since 1948 (BLS); union membership over time (BLS); the top federal income tax rate since 1913
Session 4: the growth of AI lobbying (congressional lobbying disclosure filings)
Session 5: where chatbots send voters (a regulator's published test results)
Session 6: what predicts a favorable view of AI (PPIC Statewide Survey, as a coefficient plot)

Data, not pixels. Your numbers must come from the original data source, downloaded by a script. Don't let the agent read values off the picture of the figure: that's tracing, not reproducing.

What to submit

One link to your GitHub repository. It should contain four things:

ps158-reproduce/ ├── get_data.py # downloads the data from the original source ├── plot.py # redraws the figure from that data ├── figure.png # your reproduction (any file name is fine) └── README.md # a few lines: which figure, where the data come from

The data files your script downloads can stay in the repository too.

The README can be short: a few lines saying which figure you reproduced and where your data come from. Anything else is up to you, for example how close you got or what you would do next. Your agent can write it.

To pass

The link opens your repository.
It has the four things above.
The data were downloaded from the source by a script, not typed in or read off the picture.

Bonus (optional): extend it

Once your reproduction matches, take it one step further. Add one more figure and 2–3 sentences in your README on what changed. For example:

Bring it up to date: extend the series to the latest year available
Break it down: add a group, a state or a country
Change the measure: a different statistic, or an inflation-adjusted version
Restyle it: remake it in the other course style

How to submit

Paste this on Canvas:

https://github.com/YOUR-USERNAME/ps158-reproduce

Want to keep your repository private? Contact your TA before you submit, so we can make sure we can see it.