Lecture: Public Opinion, Political Behavior & the Media
Lecture script · about 15–20 minutes spoken · Week 12
Opening
Picture a dining hall that wants to know whether students want it open later on weekends. Every number in this story is illustrative. Dining services asks two ways. It draws 400 students at random from the full enrollment list. It also posts a click poll on its app. The random sample says 58 percent favor later hours. The click poll says 70 percent, from 2,000 answers. Which number would you trust?
You'll meet that question far beyond a dining hall. Maybe you're a nursing student reading a hospital satisfaction survey. Maybe you're a business major judging a customer poll before a launch. Maybe you're a neighbor wondering whether a call-in vote on a new park speaks for the whole town. Every one of those asks when a sample can speak for everyone.
Here's where we are. Week 1 gave you the empirical and normative sort. Week 8 gave you the three-layer read of design, practice, and verdict. Week 11 opened this unit on participation and walked you through your first real data, the 2024 UK election, with the data scaffold. In this lecture we turn from votes to opinions: how they're measured, how they form, and what the media do to them.
I'll walk you through eight moves. We'll sample a population, check a poll's four facts, size the margin of error, and separate precision from bias. Then we'll read a real poll, trace how opinions form and who votes, weigh three media effects, and end on a debate with no verdict. Week 13 then opens the comparative unit.
1. Stir the pot before you taste it
Think about tasting soup. You don't drink the whole pot. You stir it and taste one spoonful. The spoonful works because you stirred, so every part of the pot had a fair chance to land on the spoon. Skim the top without stirring and you taste only what floated up.
A poll works the same way. The population is the whole group a poll wants to describe, such as all U.S. adults or every student enrolled at one school. The sample is the smaller group of people who are actually asked a poll's questions. And public opinion is the distribution of attitudes across a whole population on a political question, usually measured by asking a sample rather than everyone. Notice that word distribution. Opinion is a spread, not one number.
A random sample is a sample drawn so that every member of the population has a known, non-zero chance of being selected, which is what lets a poll carry a margin of error. That's the stirred spoonful. A self-selected sample is a sample whose members choose to take part, as in a click poll or a call-in poll; because no one has a known chance of selection, it carries no margin of error at any size. That's the skimmed top.
Back to the dining hall. The random 400 split three ways: 232 in favor, 120 opposed, and 48 unsure. That's 58 percent, 30 percent, and 12 percent, and they add to 100. The click poll's 1,400 yes votes out of 2,000 make 70 percent. The click poll has 5 times the answers.
Pause and try this. The click poll has 5 times as many answers as the random sample. Can dining services attach a margin of error to its 70 percent?
Think about how each person got into the sample.
Here's the answer. No, because the app users chose themselves, so no one had a known chance of selection, and size can't fix that.
2. Ask for the four facts
Before you buy a used car, you check the mileage, the year, the accident record, and who's selling it. A seller who hides all four has already told you something. Polls work like that too.
A trustworthy poll always reports four facts. The first is the sample size. The second is the field dates, the days on which a poll's interviews were carried out, which tell a reader when the snapshot of opinion was taken. The third is the sampling method, the procedure a pollster used to choose who was asked, such as a random draw from a list or a standing survey panel. The fourth is the stated margin, the margin of error a pollster reports for its own survey, usually a little larger than the rule-of-thumb approximation because of weighting and design; it's the figure a reader cites.
Each fact answers a question you'd otherwise have to guess. Size tells you how precise the poll can be. Dates tell you when the snapshot was taken. The method tells you whether the sample was random. The margin tells you how much room to leave around each number.
Pause and try this. The dining app's click poll reports its size, 2,000, and its method, anyone who opened the app. Which of the four facts does that method rule out?
Think about what a self-chosen sample can't carry.
Here's the answer. It rules out a stated margin, because a self-selected sample has no known chance of selection, so no margin can be attached.
So the four-fact check doesn't only find missing facts. Sometimes one fact tells you the poll can't be trusted for the whole population.
3. Size the wobble
Scoop two spoonfuls from the same well-stirred pot and they'll taste almost the same, but not exactly. Two random samples land a little apart by chance alone. The margin of error puts a size on that wobble.
The margin of error is the plus-or-minus range around a poll's figure that reflects sampling error alone. At the 95 percent level, it's about one over the square root of the sample size. The 95 percent level is a statement that the sampling procedure, repeated many times, would land within the margin about 95 times in 100. This course teaches the margin as an approximation, and I'll label it that way every time.
Here's how it works in points. You divide 100 by the square root of the sample size. The square root of 400 is 20, so the margin for 400 people is about 5 percentage points. The dining poll's 58 percent becomes a range from 53 to 63 percent. For 1,000 people, the approximation gives about 3.2, which we call about 3 percentage points. For 1,500 people it's about 2.6, and for 2,500 it's 2.
Now notice the pattern. Quadruple 400 and you get 1,600 people, whose square root is 40, for a margin of 2.5 percentage points. That's half of 5. The extra 1,200 people bought only 2.5 percentage points. That's diminishing returns.
Real polls report their own stated margin, usually a little larger. Pew Research Center's short read gives about 3 percentage points for 1,067 cases, and the Pew survey you'll read in Section 5 states a margin a little above the approximation.
Pause and try this. Using the approximation, what's the margin for 400 people, and what's the margin for 1,000?
Picture the ladder from the bottom rung up.
Here's the answer. About 5 percentage points for 400 and about 3 for 1,000, because the margin is about one over the square root of the sample size.
4. Precision is not truth
Imagine a bathroom scale that always reads two pounds light. Weigh yourself again and again and you'll get a very steady number. It'll still be two pounds wrong. More readings made it steadier, not truer.
Polls have the same two problems. Precision is how tightly a poll's estimate is pinned down by its sample size; a larger random sample gives a smaller margin of error and more precision. Bias is a lean built into a poll by who answered or how the question was asked, which pushes results away from the population's true views and which a bigger sample does not remove.
Two sources of bias matter most. Question wording is the exact words a poll uses to ask its question, and different wordings can move answers. Non-response is the failure of some selected people to answer, and when those who don't answer differ from those who do, the results lean. Neither one is in the margin of error. Put everything together and you get total error: all the ways a poll's figure can miss the population's true value, of which the margin covers only the sampling part.
So here's the trap. People read a margin of about 3 percentage points as "the poll is accurate to 3 percent." It isn't. The margin is a 95-percent statement about the sampling procedure, and it covers sampling error only.
Pause and try this. If dining services kept its click poll open until it had far more answers, would its lean go away?
Think about the scale that reads light.
Here's the answer. No, because the lean comes from who chose to answer, and adding more self-chosen answers adds more of the same lean.
Carry this line with you: bigger fixes precision, not bias.
5. Read a real poll
Now a real one. In June 2026, Pew Research Center published Americans and AI 2026: Chatbots, Smart Devices and Views on Impact. The topic isn't a hot-button political fight, so you can practice reading without a side to take.
Run the four facts. The sample size is 5,119 U.S. adults. The field dates run from the 17th to the 23rd of February 2026, which is 7 days. The sampling method is the American Trends Panel, which Pew describes on its methodology page. The stated margin is plus or minus 1.6 percentage points at the 95 percent level. That's 4 of 4.
The approximation gives about 1.4 percentage points for 5,119 people. The stated 1.6 is a little larger, as real margins usually are. You cite the 1.6.
Now the headline: 44 percent of U.S. adults reported they have used ChatGPT. With the stated margin, the range runs from 42.4 to 45.6 percent. Let's run the scaffold. What's measured? Self-reported use. Over whom and when? U.S. adults, over one week in February 2026. What does it show? A share at one point in time. What doesn't it show? Why people use the tool, or whether the wording or non-response moved the share.
Pause and try this. Does this poll show correlation or cause?
Think about how many things the poll measured at once.
Here's the answer. Neither, as it stands, because a single share is a snapshot, and it shows no cause.
That's the habit: a poll is a snapshot, not a reason.
6. Where opinions come from, and who votes
Think back to your first political opinion. Maybe it came from a dinner-table argument, a teacher, a friend, a job, or a video that stuck with you. You didn't invent your views alone.
Political socialization is the lifelong process by which people acquire their political attitudes, values, and habits from the people and institutions around them. The agents of socialization are the sources that shape a person's political attitudes: family, school, peers and life events, and the media. That's why opinion is a spread. People start in different places.
Then comes the question of who takes part. Turnout is the share of eligible people who vote in an election. It tends to rise with age, education, and income. It runs higher in presidential years than in midterm years. It varies with registration rules. Those are documented patterns, stated plainly.
Each pattern is a correlation, an association in which two things tend to rise or fall together; a correlation alone does not show that one causes the other. Turnout tends to rise with education. But age, income, and other factors travel with education, so the pattern alone can't show that schooling itself makes people vote.
Pause and try this. Is "turnout runs higher in presidential years than in midterm years" an empirical claim or a normative one?
Ask whether you could check it against evidence.
Here's the answer. It's empirical, because it describes a documented pattern you can check against the record.
"Turnout ought to be higher" would be normative, and it belongs in a separate sentence.
7. About, not what
Imagine a week when every local news site leads with a broken bridge. By the end of it, people at the bus stop are talking about the bridge. They don't agree about who's to blame, but they're all thinking about it.
That's the first of three claims about media effects, the ways news and other media may shape what the public knows, thinks about, and believes. Agenda-setting is the finding that media coverage shapes which issues the public thinks are important, rather than what people conclude about them. Framing is the way a story is presented, such as which cause or consequence it stresses, which can shape how people judge the issue. An echo chamber is a setting where people mostly meet views that match their own.
Each claim has critics, and you should hear them at full strength. On agenda-setting, critics respond that people also choose what to follow, so the arrow can run both ways. On framing, critics respond that its effects are often smaller in competitive real-world settings than in controlled studies. On echo chambers, proponents argue they harden opinion, and critics respond that many people meet more varied views than the claim assumes.
Pause and try this. A friend says, "The media tells us what to think." Which media effect does that blur, and how?
Think about the bridge at the bus stop.
Here's the answer. It blurs agenda-setting, because agenda-setting concerns what we think about, not what we conclude.
8. Delegate or trustee?
Imagine you're elected class representative, and the class wants a group project swapped for an exam. You think the project teaches more. Do you vote the way your classmates asked, or the way you judge is best?
A delegate is a representative who votes as the people they represent want, as polls, messages, or meetings show it. A trustee is a representative who votes on their own judgment of the public's interest, even when that runs against current opinion.
This is this week's open question. Proponents of the delegate view argue that representation means following the represented, and that a well-run poll is a direct way to hear them. Proponents of the trustee view argue that representatives can study a question in depth that most voters don't have time to study. They add that a snapshot poll can be badly worded, can miss the people who didn't answer, and can shift next month. Each side names a real cost in the other.
Pause and try this. A poll finds most residents oppose a new bus depot. Is "the council member ought to vote with the poll" empirical or normative?
Ask whether any poll could settle it.
Here's the answer. It's normative, on the verdict layer, because it says what the member ought to do, and the poll only reports what residents said.
The verdict is yours, and your reasons are what count.
Closing
Let me pull it together. Public opinion is a distribution, and only a random sample can speak for the whole population. A trustworthy poll reports its size, dates, method, and stated margin. The margin is about one over the square root of the sample size, so about 5 percentage points for 400 and about 3 for 1,000, and quadrupling the sample halves it. The margin leaves out wording and non-response, so bigger fixes precision, not bias. Pew's 2026 poll of 5,119 adults shows a share, not a cause. Socialization shapes views, and turnout patterns are correlations. Agenda-setting is about what we think about. And delegate or trustee is yours to argue.
Next in this module come the slide deck, the Readings and Resources page, the lecture tutorial, the practice set, and the quiz. Then Week 13, Comparative Politics, asks how you'd test a cause across countries.
Built with The Course Maker · samples.thecoursemaker.com