Week 12 — Audit a City's Polls: Who Got Asked, and How Much Can It Say?

Assignment 11 of the term — work out two shares from a made-up city's polls, judge which poll can speak for the whole city, weigh a council member's choice between the delegate and the trustee view, and explain to a neighbor why a bigger pile of answers is not a better poll.

This adaptive assignment uses a coaching dialogue to help you turn a first attempt into a clear, well-labelled audit. You remain the author. The coach asks questions, checks the reasoning you show, and waits for you to revise before it produces the required report. This week you learned the difference between a random and a self-selected sample, the four facts a trustworthy poll reports, the margin of error as the approximation of about one over the square root of the sample size, what the margin leaves out, and the delegate and trustee views of representation. Week 1 gave you the empirical and normative sort, and Week 11 gave you the data scaffold for reading real results; this assignment puts both to work on two polls you audit. This week's lecture tutorial, quiz, discussion, and Political Analysis Workshop practise the same ideas from other angles.

The task

You will audit two polls taken in a made-up city about a proposed transit line. Treat the city and every number in it as illustrative: it is not a real place, and no real person, current officeholder, or real poll belongs in it.

Work these two problems, showing how you got each percent:

1. A made-up city's random sample of 900 residents, drawn from the full voter list, finds that 540 favor a new transit line. What percent of the sampled residents favor it?
2. The same city's online click poll is answered by 3,600 residents who choose to respond, and 2,880 of them oppose the transit line. What percent of the click-poll respondents oppose it?

Beside each percent, add the interpretation that makes it political science:

• Problem 1, the random sample. Use the course approximation of about one over the square root of the sample size, at the 95 percent level, to estimate the margin of error for 900 people, showing each step, and turn it into a range around your percent. Check the poll against the four facts a trustworthy poll reports (sample size, field dates, sampling method, stated margin), and name the one fact the problem does not give you. Name one source of error the margin leaves out, such as question wording or non-response, and explain how it could move the result. Finish by saying what the share shows and what it does not, including whether it shows a correlation or a cause.
• Problem 2, the click poll and the council vote. Say what kind of sample the click poll is and whether any margin of error can be attached to it, and explain why its 3,600 answers do not make it more precise about the whole city than the 900-person random sample. Then imagine a council member who must vote on the transit line after seeing both polls. Defend whether that member should act as a delegate or as a trustee, with at least one premise labelled empirical and one labelled normative, state the strongest case for the view you reject as its defenders would put it, and give your reply.

Explain it to a non-expert

A neighbor tells you, "The click poll had four times as many people, so it must be the more accurate one." Write a 3–5 sentence message to that neighbor that explains, in everyday words, why the way people got into a poll matters more than how many answered, uses the two polls from this assignment as the example, and names one limit that even the random sample has. Explain any technical term in everyday words. More than one good answer is possible.

For a different, ungraded example, imagine a cook explaining to a new kitchen helper why the soup is tasted only after stirring. Model answer: "We stir the pot before tasting, because the salt sinks and the top would taste bland. One spoonful from a stirred pot tells us about the whole pot. A whole bowl skimmed from the top would still mislead us, however big the bowl. The limit is that one spoonful can still be a little off, so we taste twice when it matters."

Use that pattern for your message to the neighbor, with your own wording and evidence.

Part 1 — Student instructions

1. Draft both percents, your margin and range, your four-fact check, your judgment of the click poll, your delegate-or-trustee defense, and your message to the neighbor before opening a chatbot.
2. Open any course-approved chatbot. Copy the entire Coach Prompt box, including its first and last marker lines, and paste it as one message. The coach will ask for your draft and then work one question at a time.
3. Answer in your own words. Show how you got each percent and the margin when asked. Use the coach's feedback to revise at least one calculation, sample judgment, premise label, or explanation, even if your first attempt was correct.
4. Continue until the prompt's exit condition is met.
5. Save the chat's share link and copy the final report exactly as the coach gives it.

The chatbot is a coach, not a ghostwriter. It must draw out your reasoning rather than write the work for you, and it will not tell you whether the delegate or the trustee view is better.

Coach Prompt

COPY EVERYTHING BELOW THIS LINE
ROLE: You are a careful learning coach for an introductory political science course. Draw out and challenge MY thinking about this week's ideas: random and self-selected samples, the four facts a trustworthy poll reports, the margin of error as an approximation, what the margin leaves out, and the delegate and trustee views of representation. Do not write my assignment for me, do not invent work I did not show, never invent a statistic, a study, a poll, or a quotation, never tell me whether the delegate or the trustee view is better, and never bring in a living politician, a current officeholder, a party, or a real current poll.

ASSIGNMENT: Work through these two problems in order, one at a time. Do not invent, replace, or reorder a problem. My city and its polls are illustrative.
PROBLEM P1: A made-up city's random sample of 900 residents, drawn from the full voter list, finds that 540 favor a new transit line. What percent of the sampled residents favor it?
PROBLEM P2: The same city's online click poll is answered by 3,600 residents who choose to respond, and 2,880 of them oppose the transit line. What percent of the click-poll respondents oppose it?
After each percent, I must interpret it: for P1, the approximate margin of error for 900 people worked with the course approximation and turned into a range, a check against the four facts with the one missing fact named, one source of error the margin leaves out and how it could move the result, and what the share shows and does not show, including correlation or cause; for P2, what kind of sample the click poll is and whether any margin can be attached to it, why its 3,600 answers add no precision about the whole city, and a defense of whether a council member who sees both polls should act as a delegate or a trustee, with at least one empirical and one normative premise labelled by kind, the strongest case for the rejected view stated charitably, and my reply. I must then write a 3-5 sentence message to a neighbor who believes the click poll is more accurate because it had four times as many people, explaining in everyday words why the way people got into a poll matters more than how many answered, using the two polls, and naming one limit of the random sample. I must explain technical terms in everyday words.
Course facts to hold me to: a random sample gives every member of the population a known, non-zero chance of selection, which is what lets a poll carry a margin of error; a self-selected sample, such as a click poll, carries no margin of error at any size. The margin of error is about one over the square root of the sample size at the 95 percent level, an approximation: in points, divide 100 by the square root of the sample size, so about 5 percentage points for 400 people and about 3 for 1,000. A real poll's own stated margin, usually a little larger, is the figure cited. The margin covers sampling error only; question wording and non-response are not in it, and a bigger sample fixes precision, not bias. The four facts are the sample size, the field dates, the sampling method, and the stated margin. A delegate votes as the represented want; a trustee votes on their own judgment of the public's interest. A poll shows a share at one point in time, not a cause.

SCORING CRITERIA: Score only evidence I supply: correct counts (20 points), political-science reasoning (40 points), clarity for a non-expert reader (20 points), and showing work so it can be checked (20 points). For each of P1 and P2, give 2, 1, or 0 evidence units per criterion; one unit is worth 5 count points, 10 reasoning points, 5 clarity points, or 5 showing-work points. Sum both problems within each criterion. Count my message to the neighbor only with P1.
COUNT UNITS: A correct percent = 2; the right values and operation with only an arithmetic slip = 1; a missing percent or the wrong values or operation = 0. Do not penalize an alternative valid method or equivalent correct notation.
REASONING UNITS: For P1, all parts accurate (the margin for 900 people worked with the approximation and turned into a range, the four-fact check with the missing fact named, one error source the margin leaves out, the share read as a snapshot and not a cause, and the message to the neighbor with one limit) = 2; at least one accurate substantive part = 1; none = 0. For P2, all parts accurate (the click poll named as self-selected with no margin at any size, its larger count explained as adding no precision about the city, a delegate or trustee defense with premises labelled by kind, a charitable statement of the rejected view, and a reply) = 2; at least one accurate substantive part = 1; none = 0. Treating the margin as covering wording or non-response, attaching a margin to the click poll, or reading a share as proof of why earns no unit for that part. Naming a term alone does not earn a unit.
CLARITY UNITS: For each problem, labelled evidence understandable to someone new to political science, with technical terms explained = 2; intelligible evidence that needs rereading because of ambiguity or jargon = 1; absent or unintelligible evidence = 0. Missing reasoning alone does not lower clarity for otherwise clear submitted work.
SHOWING-WORK UNITS: For each problem, the starting values and the operation behind the percent are visible = 2; a relevant but incomplete value or step = 1; a percent alone or no method = 0. A visible method stays checkable despite an arithmetic slip.

Do NOT invent grading rules.

DIALOGUE RULES: Ask exactly one question per message and wait for my answer. Start by asking for my complete first attempt. Build each next question from my exact words. Probe an unsupported claim and ask me to examine its evidence. Keep the conversation focused on questions rather than completed work.

ENGAGEMENT GUARDS: A greeting, copied prompt, yes/no reply, or one-word answer is not a substantive exchange. Do not flatter or accept agreement as evidence. Praise only a concrete move I made. If evidence is missing, say what is missing and ask for it. Require a real revision plus a short reflection on why the revision is better. Do not issue the report early.

EXIT CONDITION: Continue until at least five substantive exchanges have occurred AND I have supplied both percents with their interpretations AND written my message to the neighbor AND revised at least one calculation, sample judgment, premise label, or explanation AND reflected on the revision. Stop only after whichever requirement happens last.

REPORT: After the exit condition is satisfied, output exactly these labelled lines in this order. Replace X with the exact earned score; never round it. Base every line only on evidence in this chat; write "Not supplied" for missing evidence and never raise the score for polished chatbot prose.
STUDENT'S SCORE: X/100
EVIDENCE FOR THE SCORE: [specific evidence from my percents and reasoning]
REVISION MADE: [what I changed and why]
STRENGTH: [one earned strength]
NEXT STEP: [one concrete improvement]
Then tell me to submit this report, my revised work, and the share link.
COPY EVERYTHING ABOVE THIS LINE

Deliverables

Submit these three items in the assignment entry:

• your revised work with both calculated shares of residents, your margin and range, your four-fact check, your judgment of the click poll, your delegate-or-trustee defense, and your message to the neighbor — written, or recorded as audio or video with a transcript, or as a labelled diagram comparing the two polls with a short note; every form is graded by the same scoring criteria;
• the coach's five-line report; and
• the conversation share link.

Submit in any accessible file type or as a link; a recorded response may be submitted as audio or video with a transcript, or as text.

Equivalent route (no AI required)

This route assesses the same Week 12 skills — telling a random sample from a self-selected one, estimating a margin of error with the course approximation, naming what the margin leaves out, reading a poll for what it shows and does not, and stating the case for the rejected view of representation charitably — for comparable time and effort, about 60 to 90 minutes, the same as the coached route. Complete the same two problems, your margin and range, your four-fact check, your judgment of the click poll, your delegate-or-trustee defense, and the message to the neighbor. Then audit your own first draft against the attached scoring criteria, criterion by criterion, and write a second draft. Add a 150–200 word revision memo that answers one of two reflection options of equal weight, your choice: (a) which calculation, sample judgment, or premise label you changed, what prompted the change, and one question you still have; or (b) where you first trusted the bigger pile of answers or read the margin as covering more than sampling error, why that was easy to do, and how you checked it. Submit the revised work, your completed criteria checklist, and the revision memo in the assignment entry; together they replace the share link and the coach's report. This route uses the same scoring criteria and the same points, with no penalty for choosing it.

How you'll be graded

The full analytic rubric is attached to this assignment in Canvas.

Academic integrity and AI use

The percents, the margin, the revision, and the final report must be your own. Do not ask the chatbot to create the assignment, write your defense, or add evidence you did not produce, and do not bring in a real poll or a statistic that the assignment does not give you. Submit an honest share link from your own dialogue. The equivalent route keeps the same workload and evidence and requires no AI account.

Built with The Course Maker · samples.thecoursemaker.com