Lecture: Inductive Generalization and Analogy
Lecture script · about 15–20 minutes spoken · Week 9
Opening
Picture a new taco truck parked outside the student center. Your friend goes once, orders 1 taco, waits 20 minutes, and gets a cold tortilla. That night the group chat says, "That truck is terrible. Don't bother." By morning 4 people agree, and none of them has eaten there.
One meal stood in for every meal. A nursing student does the same check when weighing one person's story against a study of thousands. A club officer does it when polling only the members who showed up. In each case, you ask who was in the sample and how they got there, because that tells you how far a result can travel.
Here's where we are. Week 1 taught you to find an argument, and Week 2 to put it in standard form. Week 3 gave you validity for deductive arguments, and strength and cogency for inductive ones. Weeks 4 and 5 tested deductive forms. Weeks 6, 7, and 8 showed how words, irrelevant reasons, and hidden assumptions lead an argument astray.
Now the course turns from the passage on the page to the evidence in the world. We'll cover seven ideas. First, the two groups in every generalization. Then three ways a sample misleads, a famous poll from 1936, and the running question's two surveys. Then the hasty generalization, and finally two ideas about analogies.
1. The pot and the spoonful
Let's start in a kitchen. You stir a pot of soup, take one spoonful, and decide the whole pot needs salt. You never drank the whole pot. You reasoned from a small part to the whole, and you trusted that spoonful because you stirred first.
That move has names. A population is the whole group a claim is about, such as every student enrolled on a campus, whether or not anyone has observed all of its members. A sample is the part of a population that is actually observed or asked, such as the 400 students a survey reaches, used as evidence about the whole group.
And the step between them has a name too. An inductive generalization is an inductive argument that reasons from what is observed in a sample to a conclusion about the whole population, so it can be strong or weak but never valid.
Remember Week 3's two standards for inductive arguments. An argument is strong when its premises, if true, make the conclusion likely. It's cogent when it's strong and every premise is true. Even a cogent argument can have a false conclusion. A generalization always leaves part of the population unchecked, so it can be very strong, but it's never valid.
Here's the running question's version. Picture a campus of 12,000 students. A survey asks 400 of them. Since 12,000 divided by 400 is 30, the survey hears from 1 student in 30, and then it reasons to all 12,000.
Pause and try this. In that survey, which group is the population, and which is the sample?
Say each group out loud before you go on.
Here's the answer. The population is all 12,000 students, and the sample is the 400 who were asked. The survey is an inductive generalization from the 400 to the 12,000.
Stir the pot first; then trust the spoonful.
2. Three traps
Now, a spoonful can mislead you. Suppose your gym wants to know whether members like the new class schedule. It has three easy ways to ask, and each one can go wrong in its own way.
It could ask 5 members. That sample is too small. Five people can all happen to be morning regulars.
It could ask whoever is in the weight room at 6 a.m. That gives you a biased sample: a sample chosen in a way that favors some members of the population over others, such as asking only the people already standing where the question matters most to them. Early risers are the members most likely to like an early class. The night-shift worker who trains late never gets asked.
Or it could email every member and count the replies. That one is open to nonresponse. That's the problem that arises when the people who choose to reply to a survey differ from the people who do not, so the replies misstate the population even if the people asked were well chosen. Upset members may be the most eager to write back.
The cure for the first two traps has a name. A random sample is a sample drawn so that every member of the population has the same chance of being picked, which is different from asking whoever happens to be nearby. Random takes a list and a fair draw. Standing in a hallway and asking whoever walks by isn't random in this sense. It favors the people who use that hallway at that hour.
A random sample is the usual way to get a representative sample. That's a sample whose members resemble the population in the ways that matter to the question, so that what holds in the sample is likely to hold in the population.
One honest limit. Even a random draw can suffer from nonresponse, because a fair draw fixes who gets asked, not who answers.
Pause and try this. A friend says, "I asked random people outside the dining hall." Is that a random sample in this course's sense?
Think about who had a chance of being asked.
Here's the answer. No, it isn't. Only people near the dining hall at that hour had any chance, so the sample favors them, and a random sample gives every member the same chance.
Too few, tilted, or self-selected: check all three.
3. Bigger is not the same as better
If a small sample is risky, you might think the cure is always more. A real poll from 1936 shows why that guess fails, and its electoral result comes from the National Archives.
That year a popular magazine, the Literary Digest, mailed straw-poll ballots to about 10,000,000 people. The names came largely from telephone directories and automobile-registration lists. About 2,400,000 ballots came back. The magazine predicted that Landon would win. Instead, Roosevelt won in a landslide, 523 electoral votes to 8.
Hold that sample size next to the running question's survey of 400. The Digest's 2,400,000 replies were about 6,000 times as many. And size still didn't rescue the poll.
Why did it miss? Two explanations are taught, and accounts differ on how much each one did. One is the lists. Directories of telephone and car owners may have favored some kinds of voters over others, which is the biased-sample trap. The other is the replies. The people who mailed their ballots back may have differed from the people who didn't, which is the nonresponse trap. This course teaches both and grades neither as the cause.
Here's the lesson to carry. A small sample drawn fairly can tell you more than a giant sample drawn badly. A bigger tilted sample is still tilted.
Pause and try this. Which would you trust more as a guide to a whole population: about 2,400,000 replies from tilted lists, or 400 people drawn fairly from a full list?
Ask how each sample was chosen before you ask how big it is.
Here's the answer. The 400 drawn fairly. A fair draw gives every member the same chance, while a huge sample from tilted lists keeps the tilt however large it grows.
More of a tilted sample is still tilted; check the method.
4. Same size, different door
Now let's run the full sampling check on the running question. Two illustrative surveys ask whether the campus library should stay open around the clock during finals week.
The first is the doorway sample. Picture someone with a clipboard at the library entrance at 11 p.m. They ask 400 students, and 300 say yes. That's 300 out of 400, which is 75 percent.
Now run the bias check. Who's walking into the library at 11 p.m.? The students already there late at night, the very students most likely to want it open late. So the sample was chosen in a way that favors some members of the population. It's a biased sample.
The second is the random sample. This time 400 students are drawn at random from the enrollment list, so every enrolled student had the same chance. 240 say yes. That's 240 out of 400, which is 60 percent.
Here's the key move. Both samples hold 400 students. So size can't explain why one reads 75 percent and the other reads 60 percent. The method of choosing does. The random sample is the better guide to the 12,000.
Let me connect that to Week 3. Take the doorway argument: 300 of 400 doorway students favor overnight hours, so about 75 percent of the campus does. Its premise can be true. Those 300 really did say yes. The argument is still weak, because the sample is biased. Now take the random-sample argument: 240 of 400 randomly chosen students favor overnight hours, so about 60 percent of the campus does. That argument is strong. If its premise is true, it's cogent, and it can still miss.
Notice one more thing. Nobody had to lie for the doorway survey to mislead you. The flaw is in the step from those 400 people to all 12,000, which is a flaw in the reasoning, not in the replies.
And this check tells you which number is the better guide to student opinion. It doesn't tell you whether the library should stay open. This course takes no side on that question.
Pause and try this. Both surveys asked 400 students. Why can't size explain the difference between 75 percent and 60 percent?
Look at what the two surveys share and what they don't.
Here's the answer. The samples are the same size, so the only real difference is how they were chosen. The doorway sample asked only late-night library users, and that tilt explains the higher number.
Same size, different door; the door decides.
5. One story is a sample of 1
Back to the taco truck. One visit, one cold tortilla, and a verdict about every meal the truck will ever serve. You now have a name for that move.
A hasty generalization is the fallacy of drawing a conclusion about a whole group from too few cases, or from cases that do not represent the group. Its most common form leans on a story. A generalization from one vivid anecdote is a hasty generalization that rests a claim about a whole group on a single memorable story, letting how striking the case is stand in for how typical it is.
Here's the passage from the course's fallacy set: "My cousin caught the flu even though she had a flu shot, so flu shots don't work." Find the population first. The conclusion is about flu shots in general, given to many people. Now find the sample. It's 1 person, the cousin. So the passage commits the hasty generalization, here a generalization from one vivid anecdote.
Notice what that verdict doesn't say. It doesn't say the cousin's story is false. She may well have caught the flu. The flaw is the step from 1 case to all of them. Whether flu shots work is a question about many people, and one case can't settle it either way.
The neighbor that isn't a fallacy is a generalization from a large, representative sample, such as a health agency tracking thousands of people chosen to represent the population. That kind of evidence is what one story can't replace.
Pause and try this. A neighbor says, "My coworker's electric scooter broke down twice in one month, so electric scooters are unreliable." What's the fallacy?
Count the cases behind the claim.
Here's the answer. It's a hasty generalization from one vivid anecdote. The sample is 1 scooter, and the conclusion is about electric scooters in general.
One story is a sample of 1, not a trend.
6. Alike in the ways that matter
Let's change shape. You're choosing a new phone. Your last phone from the same maker had a battery that lasted all day. The new model uses the same battery, the same screen size, and the same kind of chip. So you expect its battery to last all day too.
That's an argument from analogy: an inductive argument that reasons that two things alike in several known ways are also alike in a further way. It is stronger when the similarities bear on the conclusion and no relevant difference outweighs them. A disanalogy is a relevant difference between the two things an analogy compares, one that bears on the conclusion and so weakens the argument.
So the test has three steps. List the similarities that bear on the conclusion. Look for the differences that bear on it too. Then judge whether the similarities survive the differences. For the phone, the battery, the screen, and the chip all bear on battery life, so the analogy is fairly strong.
Now try a famous comparison. People sometimes say memory works like a computer's hard drive, storing each event as a file. Both take in information, keep some, and use it later. But the comparison says nothing about how memories are actually kept, and people often remember an event differently each time they recall it. The comparison can help you picture memory. It fails as proof that memory works like a hard drive.
And here's an honest limit. Counting similarities isn't the whole test. Two things can share ten features that don't matter and differ in the one that does. Relevance comes first.
Pause and try this. Suppose the new phone adds a brighter screen that stays on all the time. Is that a disanalogy for the battery argument?
Ask whether the difference bears on battery life.
Here's the answer. Yes, it is. A brighter screen that's always on bears directly on how long the battery lasts, so it weakens the analogy.
Alike in the ways that matter, not merely alike.
7. The college and the store
Some analogies aren't weak by degree. They miss. Their similarities are real, but none of them bears on the conclusion.
A weak analogy is the fallacy of resting a conclusion on a comparison whose similarities do not bear on that conclusion, or are outweighed by relevant differences.
Here's the passage from the fallacy set: "A college is like a store, and in a store the customer is always right, so students should decide their own grades." Run the three steps. The similarity is that students pay, as customers do. Now the disanalogy. A grade is a judgment of whether work meets a standard, not a product the buyer chooses. In a store you pick the shirt you like. A grade reports how your work measured up. Paying is a real similarity, but it doesn't bear on who should judge whether work meets a standard. So the passage commits the weak analogy.
That makes this week's tally from the fallacy set. One hasty generalization and one weak analogy make 2 Week 9 passages. You had keyed 15 through Week 8, so now it's 17. The set holds 20, which leaves 3 still to come.
Keep structure apart from agreement, the way you have all along. Calling the store analogy weak doesn't settle how grading should work. Someone could make a better argument on other grounds.
Pause and try this. Does spotting the weak analogy show that students should never have any say in their grades?
Separate the argument's link from its topic.
Here's the answer. No, it doesn't. The verdict says only that this comparison fails to support its conclusion, and a different argument about grading would need its own test.
A real likeness can still be the wrong one; ask if it bears.
Closing
Let me pull it together. A population is the whole group, a sample is the part you check, and a generalization reasons from one to the other, strong or weak, never valid. A sample misleads when it's too small, biased, or skewed by nonresponse, and a random sample gives every member the same chance. The 1936 poll shows that a huge tilted sample is still tilted. The doorway survey's 75 percent comes from a biased sample, while the random sample's 60 percent is the better guide. A hasty generalization rests on too few cases, often one vivid story. An analogy is tested by its relevant similarities and its disanalogies. And a weak analogy leans on a likeness that doesn't bear on its conclusion.
Next in this module come the slide deck, the Readings and Resources page, the lecture tutorial, the practice set, and the quiz. Then, in Week 10, Causal Reasoning turns the sampling check into a test for causes.
Built with The Course Maker · samples.thecoursemaker.com