Week 12 Political Analysis Workshop: How Much Can 5,119 People Tell Us?

Introduction to Political Science Free sample edition

Part 1 — The Big Picture

This week you learned that a poll can speak for millions only when its sample is random, that a trustworthy poll reports four facts, and that the margin of error, about one over the square root of the sample size at the 95 percent level, covers sampling error and nothing else. In this workshop you read one real poll the way a political scientist does. You check Pew Research Center's survey of 5,119 U.S. adults against its own methodology page, rebuild the margin approximation for four sample sizes in a spreadsheet, and compare it with the margin Pew states. Then you run the five data questions on the poll's headline share, sort three claims about it by kind, and weigh whether a lawmaker should follow such a poll. Last, you check a chatbot's version of the same analysis for a misattributed source, an invented statistic, a made-up quotation, an invented court case, a slant presented as fact, an oversimplified media claim, and a blurred empirical and normative claim.

Time budget: Plan for about 90 minutes.

Part 2 — The Task

Guiding question: What can Pew Research Center's survey of 5,119 U.S. adults tell us about all U.S. adults, how large is its margin of error and why, and what does its headline share of 44 percent show and not show?

Workshop data: the four facts and headline share of Pew Research Center's Americans and AI 2026 survey

Source: The sample size, field dates, panel, stated margin of error, and headline share of Pew Research Center's survey Americans and AI 2026: Chatbots, Smart Devices and Views on Impact, as its report and methodology page print them; you check each one against the methodology page yourself.

Rows: One survey of 5,119 U.S. adults, summarized in 4 rows of facts in Table 1, beside 4 sample sizes for the margin arithmetic in Table 2
Facts reported: sample size; field dates; sampling method (the American Trends Panel); stated margin of error at the 95 percent level
Headline share: 44 percent of U.S. adults reported they have used ChatGPT

Part 3 — Materials & Procedure

Tools and materials

Authoritative archives and data sources (the National Archives, the Avalon Project, Project Gutenberg, the Legal Information Institute, un.org, the House of Commons Library, Pew Research Center, Freedom House, V-Dem, the World Bank, Our World in Data) plus a spreadsheet for the seat-allocation and margin-of-error arithmetic, and one approved chatbot used as an analysis partner whose political-science slips the student catches
A document or notebook for your written answers

Ways of working

primary-text-analysis
read-the-data
argument-map-and-steel-man

Procedure

Open Pew Research Center's methodology page for Americans and AI 2026 in your browser. Use your browser's find command to search for 5,119, for margin, and for American Trends Panel, and fill in the middle column of Table 1 with each fact as the page reports it, including the field dates. Tool: Authoritative archives and data sources (the National Archives, the Avalon Project, Project Gutenberg, the Legal Information Institute, un.org, the House of Commons Library, Pew Research Center, Freedom House, V-Dem, the World Bank, Our World in Data) plus a spreadsheet for the seat-allocation and margin-of-error arithmetic, and one approved chatbot used as an analysis partner whose political-science slips the student catches. Check: Table 1 shows 5,119 U.S. adults, the American Trends Panel, a stated margin of plus or minus 1.6 percentage points at the 95 percent level, and field dates in February 2026.
Open a blank spreadsheet. In row 1 type the headers Sample size, Square root, and Approximate margin in columns A to C. In cells A2 to A5 type 100, 625, 3600, and 5119. Tool: Authoritative archives and data sources (the National Archives, the Avalon Project, Project Gutenberg, the Legal Information Institute, un.org, the House of Commons Library, Pew Research Center, Freedom House, V-Dem, the World Bank, Our World in Data) plus a spreadsheet for the seat-allocation and margin-of-error arithmetic, and one approved chatbot used as an analysis partner whose political-science slips the student catches. Check: Column A holds the four sample sizes from Table 2, with Pew's 5,119 in row 5.
In cell B2 type =SQRT(A2) and in cell C2 type =100/B2, then copy both down to row 5. Copy each square root and approximate margin into Table 2, rounded to two decimal places. Tool: Authoritative archives and data sources (the National Archives, the Avalon Project, Project Gutenberg, the Legal Information Institute, un.org, the House of Commons Library, Pew Research Center, Freedom House, V-Dem, the World Bank, Our World in Data) plus a spreadsheet for the seat-allocation and margin-of-error arithmetic, and one approved chatbot used as an analysis partner whose political-science slips the student catches. Check: Row 2 shows a square root of 10 and an approximate margin of 10, and every margin in column C is smaller than the one above it.
In cell D5 type =1.6-C5 to find how far Pew's stated margin sits above the approximation for 5,119 people, and write the result in the last column of Table 2. Then answer analysis questions 1 and 2. Tool: Authoritative archives and data sources (the National Archives, the Avalon Project, Project Gutenberg, the Legal Information Institute, un.org, the House of Commons Library, Pew Research Center, Freedom House, V-Dem, the World Bank, Our World in Data) plus a spreadsheet for the seat-allocation and margin-of-error arithmetic, and one approved chatbot used as an analysis partner whose political-science slips the student catches. Check: D5 shows a small positive number, so the stated margin is a little larger than the approximation, as real margins usually are.
Complete Table 3 by answering the five data questions about the headline share of 44 percent in your own words, and write the range the stated margin gives around it. Tool: Authoritative archives and data sources (the National Archives, the Avalon Project, Project Gutenberg, the Legal Information Institute, un.org, the House of Commons Library, Pew Research Center, Freedom House, V-Dem, the World Bank, Our World in Data) plus a spreadsheet for the seat-allocation and margin-of-error arithmetic, and one approved chatbot used as an analysis partner whose political-science slips the student catches. Check: Every cell in Table 3 is filled, and the correlation-or-cause row explains why a single share shows no cause.
In Table 4, label each of the three claims empirical or normative. Then state the strongest case for the delegate view and the strongest case for the trustee view of how a lawmaker should treat this poll, before you say which persuades you more. Tool: Authoritative archives and data sources (the National Archives, the Avalon Project, Project Gutenberg, the Legal Information Institute, un.org, the House of Commons Library, Pew Research Center, Freedom House, V-Dem, the World Bank, Our World in Data) plus a spreadsheet for the seat-allocation and margin-of-error arithmetic, and one approved chatbot used as an analysis partner whose political-science slips the student catches. Check: Table 4 has no blank cells, and each view's case is written in a form its defenders would accept as fair.
Paste your completed tables into one approved chatbot, then paste the prompt from Part 6 below them. Record each of its answers in Table 5 and check every one against the methodology page, your tables, and this week's definitions; or, on the equivalent route, mark up the printed flawed analysis instead. Tool: Authoritative archives and data sources (the National Archives, the Avalon Project, Project Gutenberg, the Legal Information Institute, un.org, the House of Commons Library, Pew Research Center, Freedom House, V-Dem, the World Bank, Our World in Data) plus a spreadsheet for the seat-allocation and margin-of-error arithmetic, and one approved chatbot used as an analysis partner whose political-science slips the student catches. Check: Table 5 has one row for each of the chatbot's seven answers, each marked as checked against a source.

Fallback route

When to use it: Use this route if no spreadsheet is available or Pew's methodology page cannot be reached.

Work the same steps by hand. Find each square root in Table 2 by asking which number multiplied by itself gives the sample size (for 5,119, use about 71.55), divide 100 by it, and round to two decimal places. Table 3's facts are those printed in Part 2 of this workshop. Write each calculation so it can be checked, complete every table, answer every analysis question, do the critique step, and submit the same items for equal credit.

This route earns the same credit as the primary route.

Part 4 — The Scaffold

Table 1 holds the four facts a trustworthy poll reports; fill in each one from Pew's methodology page before you compute anything.

Student scaffold table 1
Fact	What the methodology page reports	Why a reader needs it
Sample size
Field dates
Sampling method
Stated margin of error

Table 2 rebuilds the margin approximation in your spreadsheet so you can see why a larger sample buys less and less precision.

Student scaffold table 2
Sample size	Square root	Approximate margin in points (100 ÷ square root)	Stated margin minus the approximation (Pew only)
100			not needed
625			not needed
3,600			not needed
5,119 (Pew's survey)

Table 3 walks the five data questions so that what the headline share shows stays apart from what it does not.

Student scaffold table 3
Data question	Your answer
What is measured?
Over whom, and when?
What does it show? Give the range the stated margin puts around 44 percent.
What does it not show? Name one source of error the margin leaves out.
Correlation or cause?

Table 4 sorts three claims about the poll by kind and then states both views of how a lawmaker should treat it.

Student scaffold table 4
Claim	Kind (empirical or normative)
Pew surveyed 5,119 U.S. adults.
44 percent of U.S. adults reported they have used ChatGPT.
Lawmakers ought to follow this poll when they write rules about AI.
Strongest case for the delegate view
Strongest case for the trustee view

Table 5 logs the chatbot's answers, or the flawed analysis's claims, and the check you ran on each one.

Student scaffold table 5
The chatbot's answer or the flawed claim	Right or wrong?	Your correction and its evidence

Part 5 — Analysis Questions

Report the square root and the approximate margin for each sample size in Table 2, and show the spreadsheet formula or hand calculation you used for one of them. How many times larger is 3,600 than 100, and how many times smaller is its margin?
Compare Pew's stated margin with the approximation for 5,119 people. Which figure do you cite, and why is the stated one usually a little larger?
Using Table 3, explain what the range around 44 percent means at the 95 percent level, and why question wording and non-response could still move the share without the margin showing it.
Using Table 4, explain why the claim about what lawmakers ought to do is normative even though the 44 percent is a fact. Then state the strongest case for the delegate view and for the trustee view before you say which persuades you more and why.
Name two limits of this analysis, such as that a single share is a snapshot from one week, that self-reported use may differ from actual use, that the approximation ignores weighting and design, or that a margin says nothing about a slanted question, and say how each one limits what you can conclude.
Choose one reflection option and answer it in 3–4 sentences: (a) Which fact in Table 1 would you most want to check before trusting any poll you see in the news, and why? (b) After this workshop, would you rather a lawmaker act as a delegate or a trustee on AI rules, and what is the strongest case against your choice?

Part 6 — AI-Critique Moment

Choose one equal-credit critique route. Route A: ask an approved AI tool to critique your completed work, then check its response against your own evidence. Route B is the no-AI equivalent printed below.

Approved tools: Claude, Gemini, ChatGPT, Grok, Microsoft Copilot.

Prompt to paste:

I've pasted my completed workshop tables above. They read Pew Research Center's survey Americans and AI 2026. Please do seven things. First, tell me who conducted this survey and on which panel. Second, give me the survey's sample size and its margin of error. Third, quote, word for word, the sentence on Pew's methodology page that states the margin of error. Fourth, name a court case that requires polls to report their margin of error. Fifth, tell me whether the 44 percent who reported using ChatGPT shows that chatbots are good or bad for the country, and which kind of voter uses them more. Sixth, tell me whether news coverage of AI tells people what to think about chatbots. Seventh, tell me whether lawmakers ought to follow this poll on AI rules, and label each of the three claims in my Table 4 as empirical or normative.

Test the response for these likely failures:

Chatbots often credit a well-known survey to the wrong organization, such as another polling firm or a government agency, or describe the sample as a phone poll of voters.
Chatbots often give a rounded or invented sample size, or report the course approximation, about 1.4, as the poll's margin instead of the stated figure.
Chatbots often put a smooth sentence in quotation marks and present it as Pew's own wording, though no such sentence appears on the page.
Chatbots often name a court case, with a confident citation, that supposedly requires pollsters to report their margin of error.
Chatbots often turn the 44 percent into a verdict that chatbots are good or bad, or claim one party's voters use them more, presenting a slant as something the data show.
Chatbots often answer flatly that the media tell people what to think, treating a contested media-effects claim as settled.
Chatbots often say the poll proves lawmakers should follow it, blending what people reported with what officials ought to do.

In 3–5 sentences, say what the chatbot got right, quote at least two sentences it got wrong, and correct each one with evidence from the Pew methodology page, your tables, or this week's definitions.

Equivalent route (no AI required)

No AI account is required for this workshop. This route assesses the same objectives, Objective 6 (explain American government and political participation — federalism, separation of powers, and the U.S. Constitution as one case of the survey's concepts; parties, elections, and voting systems and their effects; public opinion, polling, political behavior, and the media — reading election and polling data for what it documents and what it does not settle) and Objective 1 (describe the subfields and methods of political science — political theory and philosophy, comparative politics, international relations, American government, and political methodology — and analyze a political text or dataset with the discipline's core tools: concept application, argument analysis, evidence evaluation, the comparative method, and the empirical/normative distinction), takes comparable effort of about the same 90 minutes, uses the same rubric, and earns the same credit. Instead of the chatbot, critique this flawed analysis, printed here in full: "A national polling firm surveyed about 10,000 voters by phone for this study, and its margin of error is about 1.4, so the 44 percent is accurate to within 1.4 percent. Pew's methodology page says, 'this poll is accurate for every American.' A Supreme Court ruling requires every poll to report its margin of error, so this one must be trustworthy. The result shows these tools are good for the country and that younger voters of one party use them most. Because so much news covers the topic, the media have told Americans what to think about these tools. Since 44 percent reported using the tool, lawmakers must follow the poll, and the data prove it." Mark every error you can find, quote at least two, and correct each one in 3–5 sentences with evidence from the Pew methodology page, your tables, or this week's definitions. Submit the marked-up analysis and your correction in place of the chatbot report and share link.

Part 7 — What to Submit

All five completed scaffold tables
Written answers to all six analysis questions
The chatbot critique report with its share link, or the equivalent route's marked-up flawed analysis

Deliverable:

A short Political Analysis Workshop report on Pew Research Center's Americans and AI 2026 survey: five completed tables (the four facts, the margin approximation for four sample sizes, the five data questions, three claims sorted by kind with both views of representation, and a critique log), written answers to six analysis questions, and a critique of a chatbot's version of the same analysis, or, on the equivalent route, a marked-up correction of the printed flawed analysis. 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. Choose how you show it: written answers, a recorded explanation with a transcript, or a labelled chart of the margin against sample size with short notes; every form is graded by the same scoring criteria.

Rubric

The rubric below shows how each submitted part is evaluated; bands are full, partial, or none.

Workshop evaluation criteria
Criterion	Full	Partial	None
Checking and recording the poll's four facts	All four facts in Table 1 are checked against the methodology page and recorded accurately, each with a reason a reader needs it, and the stated margin is reported in percentage points at the 95 percent level.	Most facts are checked, but one is copied wrongly, a field date or the panel is missing, or a reason is left out.	The facts are not checked, or they are replaced with numbers from memory or a chatbot.
Computing the margin approximation	Every square root and approximate margin in Table 2 is computed correctly, the stated margin is compared with the approximation for 5,119 people, and the diminishing-returns pattern is described.	Most margins are right, but one or two contain a slip, or the comparison with the stated margin is missing.	No margins are computed, or the approximation is reported as the poll's margin.
Reading the data with the five questions	All five data questions are answered; the range around 44 percent is stated with its meaning; one error source outside the margin is named and explained; the share is read as a snapshot that shows no cause; and two real limits are stated with their effect on the conclusion.	Some questions are answered well, but the range is misread as accuracy, the error source is missing, or only one limit is given.	The data questions are unanswered, or the reading claims the poll shows a cause or a verdict.
Critique and correction	The chatbot critique report or the marked-up flawed analysis quotes at least two errors and corrects each with evidence from the methodology page, the tables, or this week's definitions; both routes earn this band equally.	The chatbot critique report or the marked-up flawed analysis names an error but corrects it without evidence, or corrects only one; both routes earn this band equally.	No chatbot critique report or marked-up flawed analysis is submitted, or the output is pasted without any check; both routes are scored the same way.
Kind, both views, and reflection	All three claims in Table 4 are labelled correctly by kind, both views of representation are stated fairly, and the chosen reflection option is answered with a reason.	One claim is mislabelled, one view is weakened, or the reflection gives no reason.	No claims are labelled, no view is stated, and no reflection is given.

Quality Gate

Every fact in Table 1 matches Pew's methodology page, and no sentence is quoted unless it can be found there word for word.
Every square root and margin in Table 2 comes from your own formula or hand calculation, never from memory or a chatbot.
The poll's stated margin, not the approximation, is the figure cited for the survey, and every margin is written in percentage points.
Every claim in Table 4 is labelled by kind, and both views of representation are stated as their own defenders would put them.
No court case, statistic, study, poll, or source is invented anywhere in the submission, and no living politician is named.
No verdict on whether chatbots are good or bad, or on which view of representation is right, is presented as something the data show.
The critique quotes the chatbot or the flawed analysis and corrects at least two errors with evidence.

Built with The Course Maker · samples.thecoursemaker.com