Lecture: Research Methods & Ethics
Lecture script · about 15–20 minutes spoken · Week 2
Opening
Think about the last time someone told you "studies show." Maybe it was a headline about screen time, or an ad for a sleep app, or a friend who swears a supplement helps them focus. A nurse weighing a new hospital routine hears claims like that. So does a manager reading a report about remote work, and so does a parent reading about phones. Each of them needs the same three answers. What design produced the finding? What is that design allowed to conclude? And were the people in the study treated fairly?
In Week 1, you met psychology as the science of behavior and mental processes. You saw that common sense explains everything after the fact, which is hindsight bias, and that a science has to test its ideas against evidence. In this lecture I'll show you how that testing works. We'll go through the research loop, the three main designs, how to read a correlation, what makes an experiment different, and the ethics rules that protect participants. Next week, in Week 3, the course turns to the brain, and you'll read every brain finding with these questions in hand.
Most of my examples come from a few studies you'll meet all week: a memory study published in 1885, a study of eyewitness memory, and a small set of sleep and quiz numbers from ten students. Let's start with the loop.
1. The research loop and the operational definition
Imagine your roommate says, "I remember things better in the morning." That might be true. But what does "remember better" mean? Better by how much, and measured how? Until someone answers that, nobody can check the claim.
Research runs as a loop. A theory produces a hypothesis, which is a prediction the data could prove wrong. The hypothesis needs a measurement. Data come in. Then someone repeats the study, and the result has to show up again. That repeat is called replication: repeating a study to see whether its result appears again.
The step that makes a fuzzy idea measurable has a name. An operational definition is the exact, observable procedure a researcher uses to measure a construct. A construct is an idea you can't see directly, like memory or stress.
Here's the classic example. Hermann Ebbinghaus, who published his memory experiments in 1885, wanted to study forgetting. He memorized lists of nonsense syllables, waited, and relearned them. He measured memory as percent savings: how much less time or effort relearning takes than learning the first time. His savings fell from 58.2% at 19 minutes to 21.1% at 31 days. Because he wrote down his exact procedure, anyone could repeat it.
Pause and try this. Savings fell from 58.2% at 19 minutes to 44.2% at 63 minutes, and from 25.4% at 6 days to 21.1% at 31 days. Which stretch lost more?
Compare the two drops before I go on.
Here's the answer. The first hour lost far more. 58.2 minus 44.2 is 14.0 points, while 25.4 minus 21.1 is only 4.3 points. Most forgetting happened fast, and then it slowed down.
2. Descriptive research: describing without explaining
Picture a researcher with a clipboard near a busy four-way stop. She doesn't stop anyone. She counts how many drivers come to a full stop and how many roll through. By the end, she knows what drivers do. She doesn't know why.
That's descriptive research: it records what people do, say, or report as it happens, without testing what causes it. There are three main kinds. A case study looks at one person or a small group in great depth. Naturalistic observation watches behavior in its everyday setting without changing it, like the researcher at the stop sign. A survey asks many people to report their own attitudes, beliefs, or behavior.
Each one gives you something real. A case study can follow a rare experience in rich detail. Observation sees behavior outside a lab. A survey can reach thousands of people cheaply. But each has a limit. One person may not be like anyone else. People who know they're being watched may change. And people answering a survey may misremember or try to look good.
Pause and try this. A survey of office workers finds that people who eat lunch away from their desks report less stress. Can that survey show that leaving your desk lowers stress?
Think about what the survey actually measured.
Here's the answer. No. The survey only recorded two answers from each person. Maybe less-stressed workers feel free to leave their desks, so the design can describe the link but not explain it.
3. Reading a correlation
Now picture ten students in a study group. Each wrote down how many hours they slept and their score on the next day's 20-point quiz. Nobody changed anyone's sleep. The researcher measured two things that were already happening. That's a correlational study: it measures variables as they naturally occur and reports how strongly they move together.
The number that reports it is the correlation coefficient, written as the letter r. It runs from minus 1 to plus 1. The sign gives the direction. Positive means the two rise together. Negative means one falls as the other rises. The size gives the strength. Close to 1 in either direction is strong, and close to 0 means little linear link.
For these ten students, mean sleep was 7.25 hours and the mean score was 14.7 points. And r came out to 0.8457.
Pause and try this. With r at 0.8457, what's the direction, and is the link strong or weak?
Look at the sign first, then the size.
Here's the answer. The sign is positive, so more sleep tends to go with higher scores. And 0.8457 is close to 1, so the link is strong. In this sample, students who slept more tended to score higher.
Notice one thing. r is not a percentage, and it isn't a share of anything. It's a measure of how tightly two variables move together.
4. Why correlation is not causation
So should we write "sleep raises quiz scores"? Not yet. Picture a hardware store that sells more umbrellas on the same days the town reports more fender-benders. Umbrellas don't cause car crashes. Rainy weather drives both.
That's the third-variable problem: an outside factor drives both measures, so they move together even though neither causes the other. There's a second trap, the directionality problem: the causal arrow might run the other way from the one you assume.
Let's apply both to the sleep data. A student who slept more may also have finished studying earlier. Finishing early could make sleep easier and also raise the score. That's a third variable. Or maybe a good day of studying made sleep easier, so the score's cause came first. That's directionality.
Pause and try this. Name one third variable, other than studying early, that could link more sleep with higher scores.
Think about the rest of a student's day.
Here's the answer. A lighter work schedule would work, and so would feeling less anxious about the quiz. Either one could raise both sleep and scores, which is why this design earns "associated with" and nothing stronger.
5. The experiment
To claim a cause, you need a different design. In an experiment, the researcher changes one thing on purpose, measures what happens, and decides by chance who gets which condition.
The classic example is Loftus and Palmer, 1974. Students watched films of car accidents. Then they were asked how fast the cars were going when they hit each other. But different students heard different verbs: smashed, collided, bumped, hit, or contacted. The verb is the independent variable, the variable the researcher manipulates. The speed estimate is the dependent variable, the variable the researcher measures. With "smashed," the average estimate was 40.8 miles per hour. With "contacted," it was 31.8. That's a gap of 9.0 miles per hour from one word.
In a second experiment, 150 students came back a week later. They were asked whether they'd seen broken glass. There wasn't any. The control group, which got no speed question at all, gives the baseline. The experimental group is any group that gets the manipulation.
Pause and try this. In that second experiment, 32% of the "smashed" group reported glass, against 14% for "hit." Which variable is the IV here?
Ask yourself what the researchers changed.
Here's the answer. The IV is still the verb, because that's what the researchers changed. Whether students reported glass is the DV. And 32 minus 14 is 18 points more for "smashed," so about one in three who heard "smashed" remembered glass that was never there.
6. Random sampling versus random assignment
What lets the verb study claim a cause? Random assignment: placing each participant into a condition by chance. Careful viewers and distracted ones end up spread across every group. So the verb is the only planned difference, and you can say the wording changed the estimate.
There's a second random step that people mix up with this one. Random sampling means choosing participants so that every member of the population, the whole group you care about, has an equal chance of being picked. The people you end up with are your sample.
Here's a way to keep them apart. Picture a company with 2,000 employees testing a new afternoon break. A manager draws 100 names from the full staff list. That's random sampling. It decides who is studied, and it lets her generalize to the whole staff. Then she flips a coin for each person to decide who gets the new break. That's random assignment. It decides who gets which condition, and it lets her claim a cause.
Pause and try this. Suppose the manager draws names fairly but lets people choose their own break. Can she generalize, claim a cause, both, or neither?
Ask which random step she kept.
Here's the answer. She can generalize, because the sampling was random. She can't claim a cause, because people chose their own condition, and those who chose the new break may differ in other ways.
7. Reliability versus validity
Step on a bathroom scale three times. It reads the same number each time. Good. Now suppose that number is always five pounds lighter than your real weight. The scale is consistent and also wrong.
That scale separates two qualities of any measure. Reliability is consistency: a reliable measure gives the same or a very similar result when you repeat it. Validity is accuracy: a valid measure captures what it claims to measure. A measure can be reliable without being valid, and the five-pounds-light scale proves it.
This matters every time a study uses an operational definition. Say a study app measures your focus by counting how often you open it each day. The count may be steady from day to day, so it's reliable. But opening an app a lot might mean you're focused, or it might mean you're distracted. So it may not be valid.
Pause and try this. A measure gives a very different reading every time you repeat it. Can it be valid?
Think about what a valid measure has to do first.
Here's the answer. Not really. If the readings jump around, you can't trust any one of them to capture the real thing. Reliability doesn't guarantee validity, but you need it first.
8. Protecting the people in the study
Imagine you sign up for a memory study. Halfway through, you feel upset and want to stop. Can you leave? Who checked the study before it began? Research ethics answers those questions, and psychology learned many of its rules the hard way.
In Milgram's study, published in 1963, participants were told to give shocks to a learner for wrong answers, up to a switch marked 450 volts. No real shocks were given, but participants believed they were real. Milgram reported that 65% of participants, 26 of 40, continued to the 450-volt switch. Many showed strong stress. The obedience finding belongs to Week 13. Here, the study matters because of how its participants were treated.
A harsher piece of history: from 1932 to 1972, the U.S. Public Health Service followed Black men with syphilis without their informed consent and without offering available treatment. The outcry after this study at Tuskegee led to the National Research Act of 1974 and to the modern review system.
Today, the American Psychological Association's ethics code, Standard 8, sets the rules. Informed consent means people agree after being told what the study involves and its risks. They have the right to withdraw at any time, and protection from harm. Deception is allowed only when there's no other way to study the question and no harm is expected. It must be followed by debriefing, which tells people the truth afterward. Confidentiality keeps individual answers private. And an Institutional Review Board, or IRB, must approve the study before it begins.
Pause and try this. Milgram's participants believed the shocks were real. Which protection must follow any deception like that?
Think about what happens after the study ends.
Here's the answer. Debriefing. Participants must be told the true purpose and the deception afterward, and have their questions answered.
9. The three-question move
Let's put the whole lecture into one habit. Here's a headline: "Study shows people who eat breakfast do better in school." Before you share it, ask three questions. What design produced it? What can that design conclude? And what third variable could explain it?
If researchers measured breakfast and school results as they happened, the design is correlational, and it earns "associated with." A family with a steady morning routine might also make time for homework. That's a third variable. Only an experiment with random assignment could earn "causes."
Try the habit on the two studies from this lecture. The sleep data are correlational, so they earn "associated with," with an r of 0.8457, and studying early is a third variable. The verb study is an experiment, so it earns "causes." Random assignment spread the other differences across the groups.
Pause and try this. A headline says a randomized study found a new reading method changed scores. Which of the three questions does random assignment help answer?
Think about what random assignment rules out.
Here's the answer. The third question. Random assignment spreads other factors across the groups by chance, so a third variable is much less likely to explain the difference.
Closing
Let me pull it together. Research runs as a loop, and an operational definition makes a fuzzy idea measurable, the way Ebbinghaus measured memory as percent savings. Descriptive research describes without explaining. A correlation gives direction and strength with r, and earns only "associated with." A third variable or a reversed arrow can explain any correlation. An experiment changes the independent variable, measures the dependent variable, and uses random assignment to earn "causes." Random sampling lets you generalize; random assignment lets you claim a cause. A good measure is both reliable and valid. And every study with people needs consent, the right to withdraw, protection from harm, debriefing after deception, confidentiality, and IRB approval.
Next in this module come the slide deck, the Readings & Resources page, the lecture tutorial, the practice set, and the quiz. Then the discussion, the assignment, and the Analysis Workshop let you use these questions on real claims. Next week, in Week 3, we'll look at the biological bases of behavior: the neuron, the synapse, and the brain.
Built with The Course Maker · samples.thecoursemaker.com