Research strategies and ethics

You would want to know how the claim was studied. Did researchers simply notice that cereal eaters were healthier? Or did they assign some people to eat the cereal and compare them with people who did not? Those two kinds of study can sound alike in a headline, but they support very different conclusions (Spielman et al., 2020).

This lesson walks through the main ways psychologists gather evidence, what each can and cannot tell us, and the protections every study owes the people who take part in it. Good research is careful about both: the truth of its conclusions and the dignity of its participants.

Sometimes they can. But behavior is visible and the mind is not. If someone is crying, we can see the tears, but not whether the person is sad, in pain, or happy. People may be unwilling to answer honestly, or unable to answer at all, as with an infant (Spielman et al., 2020). So psychologists need a toolbox of methods, each suited to a different kind of question.

Case studies examine one person or a few people in great depth. They are often used for rare conditions, and the richness of the information is unmatched. The catch is the same rarity: a person who is unlike most people is hard to generalize from, meaning it is hard to apply what we learn to everyone else (Spielman et al., 2020).

Krista and Tatiana Hogan, Canadian twins joined at the head, show why case studies still matter. Their brains appear to be connected at the thalamus, a relay center for sensory information, so researchers have followed them closely to learn whether one twin can experience what the other senses. Over time it has become clear that they share some sensory experiences and motor control, yet remain two distinct people (Spielman et al., 2020). No large survey could have taught that.

Naturalistic observation means watching behavior where it normally happens, as unobtrusively as possible, because people act differently when they know they are being watched. Ask a class who always washes their hands after using the restroom and nearly everyone will raise a hand; quietly watching the sinks tells a different story. Jane Goodall spent nearly five decades observing chimpanzees this way. In a study of peer exclusion, researchers watched preschool children on a playground while a few children wore wireless microphones, which they quickly forgot about (Spielman et al., 2020).

The payoff is ecological validity, or realism: people behave as they normally would, so the findings carry over to real life. The cost is control. The researcher cannot make the behavior happen, and may wait days for it. Sometimes researchers use structured observation instead, watching people during a set task. Mary Ainsworth’s Strange Situation, in which a stranger enters a playroom and a caregiver leaves and returns, is a classic example used to study infants’ attachment (Spielman et al., 2020).

Observers can also see what they expect to see. Researchers guard against that observer bias with clear rules for what counts, and by having several observers record the same events and checking how well they agree, which is called inter-rater reliability (Spielman et al., 2020).

Surveys ask many people the same questions. They can reach a large sample, a subset of the population the researcher cares about. But survey answers are only as good as people’s honesty and memory. People may misremember or answer in ways that make them look good (Spielman et al., 2020).

Careful wording can help. In one survey study, participants would not report prejudice toward any ethnic group when asked directly. But when asked indirect questions, such as how likely they would be to introduce themselves to someone, they showed less willingness to interact with Arab Americans than with other groups, a sign of subtle prejudice they did not admit (Spielman et al., 2020).

To summarize a pile of answers, researchers often start with central tendency. The mode is the most common response, the median is the middle value, and the mean is the arithmetic average. The mean is the most useful for later analyses, but a few extreme scores, called outliers, can pull it far from what is typical (Spielman et al., 2020).

Archival research uses records that already exist, such as school or hospital records. It is inexpensive, but the researcher has no control over what was collected or how (Spielman et al., 2020).

Some questions are about change over time. Longitudinal research follows the same people for years. Cross-sectional research compares different age groups at one moment (Spielman et al., 2020).

Cross-sectional studies are faster, but each age group, or cohort, grew up in a different era. If 20-year-olds and 60-year-olds answer differently, the cause may be the times they grew up in rather than their age. Longitudinal studies avoid that problem, but they take years and money, and they lose participants who move, fall ill, or simply stop, a loss called attrition. They can be powerful. Long-term studies sponsored by the American Cancer Society, which followed very large numbers of people, gave some of the first scientific evidence linking smoking with cancer (Spielman et al., 2020).

All of these methods are correlational: they can reveal relationships, but not cause and effect. Only an experiment can do that (Spielman et al., 2020). Figure 2.1 compares the methods side by side.

A correlation means two variables change together. Researchers summarize it with the correlation coefficient, written r, a number from −1 to +1 (Spielman et al., 2020).

• The sign gives the direction. In a positive correlation, both variables rise together, like height and weight. In a negative correlation, one rises as the other falls, like hours of sleep and daytime tiredness.
• The size gives the strength. The closer r is to +1 or −1, the more tightly the points hug a straight line and the better one variable predicts the other. An r near 0 means little or no linear relationship, like shoe size and hours of sleep (Spielman et al., 2020).

Real correlations are often modest. Student researchers at the University of Minnesota found a weak negative correlation, r = −0.29, between how many nights a week students slept less than five hours and their grade point average (Spielman et al., 2020). Even modest correlations are useful for prediction. An admissions office could check how well test scores have tracked grades among its current students and use that to predict how applicants might do (Spielman et al., 2020). Try Figure 2.2 to see what different values of r look like.

As ice cream sales rise, so does crime. No one thinks a cone causes a crime spree. Both rise in hot weather, when more people are outside and more people want something cold. Temperature is a confounding variable: a third factor that produces a relationship between two others (Spielman et al., 2020). In Figure 2.2, choose the ice cream example and reveal the hidden variable to watch it happen. Figure 2.3 then gives the hidden variable its own dimension, so you can turn the data and see it from the side.

There is a second trap: direction. Even when two variables really are connected, a correlation cannot tell you which one drives the other. And sometimes the relationship is not there at all. An illusory correlation is a relationship people believe in that does not exist. Many people are sure the full moon makes people act strangely, but a review of nearly 40 studies found no such relationship (Rotton & Kelly, 1985). We notice odd behavior on full-moon nights and forget the ordinary ones, a habit called confirmation bias (Spielman et al., 2020). Illusory correlations are not harmless. Research suggests they help form prejudice, when certain behaviors are wrongly linked to certain groups (Spielman et al., 2020).

This is why the cereal headline deserves caution. Research did find that people who eat a certain breakfast cereal may have a lower risk of heart disease, and cereal makers have every reason to present that finding in the best light (Spielman et al., 2020). But people who eat that cereal may differ in many ways from people who do not. A correlation is a clue worth following, not a verdict.

An experiment takes control of the variable in question. The researcher changes one thing, the independent variable, and measures its effect on another, the dependent variable. The dependent variable “depends” on the independent one (Spielman et al., 2020).

The simplest design has two groups. The experimental group receives the manipulation. The control group is treated the same in every way except that one. Each variable also needs an operational definition, a precise description of how it will be measured, so that anyone could repeat the study (Spielman et al., 2020).

Two different kinds of randomness make the logic work:

• Random sampling decides who is in the study. In a random sample, every member of the population has an equal chance of being chosen, so a large enough sample tends to resemble the population (Spielman et al., 2020). This protects generalization.
• Random assignment decides who goes into which group. Every participant has an equal chance of landing in either group, so the groups are unlikely to differ in any systematic way before the study begins. Then a difference at the end can be credited to the independent variable (Spielman et al., 2020). This protects cause and effect.

Expectations can also sneak in. If participants believe a pill will help, they may feel better even when it is a sugar pill, which is called the placebo effect. And researchers who know who got the real treatment may read ambiguous results in their favor, which is experimenter bias. When only the participants are kept unaware of their group, the study is single-blind. In a double-blind study, neither participants nor researchers know who is in which group, which guards against both (Spielman et al., 2020).

Who takes part matters, too. Psychology has long relied on college students, who tend to be younger, more educated, and less diverse than the general population, so findings from them can be hard to generalize (Spielman et al., 2020). Build a study step by step in Figure 2.4.

Some variables cannot be assigned. A researcher who wants to know whether sex affects spatial memory can compare men and women, but cannot randomly assign anyone’s sex. Studies like this are quasi-experimental, and they do not support cause-and-effect claims (Spielman et al., 2020).

Other variables must not be assigned. No one could ethically assign children to experience abuse to see whether it lowers self-esteem in adulthood, so questions like that have to be studied with correlational methods (Spielman et al., 2020). And because experiments often take place in artificial settings, it is fair to ask whether a lab result holds up in everyday life (Spielman et al., 2020). The strongest conclusions usually come from several methods pointing the same way.

Even random groups differ a little by chance. A statistical test estimates how likely a difference that large would be if the groups did not really differ. Psychologists usually call a result statistically significant when that chance is less than 5 percent (Spielman et al., 2020).

Then the work goes through peer review. Other experts, usually anonymous, check the reasoning, the methods, the statistics, and the ethics before a journal publishes it. Peer review also requires the methods to be described clearly enough for others to replicate the study, that is, repeat it with new participants (Spielman et al., 2020).

Replication is where science corrects itself. In a large project, researchers repeated 100 published psychology studies. About 97 percent of the originals had reported significant results, but only about 36 percent of the replications did (Open Science Collaboration, 2015). Findings like this sparked what is often called the replication crisis. It has pushed many scientists to share their data and plans more openly, and some argue it is improving how science is done (Spielman et al., 2020).

Sometimes the published record has to be corrected outright. A retraction withdraws a published article because its data were falsified or fabricated or its design had serious problems. Several studies claiming that childhood vaccines cause autism were retracted; in the best-known case, the lead researcher had a financial interest in establishing that link. Large-scale research has since shown that vaccines do not cause autism, but the original reports had already frightened many parents. Between January and October 2019, the United States saw 22 measles outbreaks and more than a thousand cases (Spielman et al., 2020).

Two more words matter for any measurement. Reliability is consistency: does the tool give the same result each time? Validity is accuracy: does it measure what it claims to? A kitchen scale that always reads 50 grams too heavy is reliable but not valid. A valid measure must be reliable, but a reliable measure is not automatically valid (Spielman et al., 2020).

Each comes in several kinds. Reliability includes inter-rater reliability, internal consistency (whether survey items that measure the same thing agree with each other), and test-retest reliability (whether a measure gives similar results when it is repeated). Validity includes ecological validity, construct validity (whether a variable captures what it is meant to), and face validity (whether it looks right on the surface) (Spielman et al., 2020).

Research protections were not handed down; they were learned, often painfully. After the Second World War, the Nuremberg Code was drafted to judge doctors who had experimented on concentration camp prisoners. It became the prototype for later research codes (National Commission, 1979).

In the United States, the most notorious case is the U.S. Public Health Service syphilis study at Tuskegee, Alabama. From 1932 to 1972, researchers followed 600 Black men, 399 with syphilis and 201 without. There is no evidence the men gave informed consent, and they were not offered treatment even after penicillin became widely available (Centers for Disease Control and Prevention, 2024). Many of the men were harmed, and some of their wives and children were infected. The study ended in 1972, after news reports brought it to public attention (Centers for Disease Control and Prevention, 2024; Spielman et al., 2020).

The outrage led to the National Research Act of 1974. The commission it created wrote the Belmont Report in 1979, which names three basic principles for research with people (National Commission, 1979):

• Respect for persons: treat people as able to make their own choices, and protect those who cannot fully decide for themselves. It is applied through informed consent.
• Beneficence: do not harm, and maximize possible benefits while minimizing possible harms. It is applied through a careful weighing of risks and benefits.
• Justice: share the burdens and benefits of research fairly, so that no group is chosen simply because it is easy to use. It is applied through fair selection of participants.

Choose a pillar in Figure 2.5 to see how each principle becomes a working rule.

Any institution that receives federal support for research with people must have an institutional review board (IRB), a committee of scientists, administrators, and community members that reviews each study before it begins (Spielman et al., 2020).

In the United States, federal rules often called the Common Rule govern how IRBs work (U.S. Department of Health and Human Services, 2018). Review is scaled to risk. Some studies with minimal risk, such as certain anonymous surveys, can be exempt from ongoing review, and other minimal-risk studies can get expedited review by an experienced board member. Studies with more than minimal risk go to the full board. At most institutions, the IRB, not the researcher, decides which level applies.

Psychologists also follow the APA Ethics Code. It opens with five general principles: beneficence and nonmaleficence (benefit people and do no harm), fidelity and responsibility, integrity, justice, and respect for people’s rights and dignity. Its enforceable standards then spell out specific duties (American Psychological Association, 2017).

The APA Ethics Code spells out what participants must be told in informed consent: the purpose, how long it will take, and what will happen; their right to decline or withdraw at any time; any foreseeable risks or discomfort; possible benefits; the limits of confidentiality; any incentives; and whom to contact with questions (American Psychological Association, 2017). For participants under 18, a parent or guardian must also give permission (Spielman et al., 2020). Paying participants is allowed, but the code warns against offers so large or inappropriate that they pressure people into taking part (American Psychological Association, 2017).

Sometimes telling participants the exact hypothesis would change how they act. Deception is allowed only when the study has significant value and no honest alternative would work, and never about anything expected to cause physical pain or severe emotional distress. Afterward, participants must receive a debriefing: a full, honest explanation of what happened and why, with the chance to withdraw their data. Any deception must be explained as early as feasible, preferably at the end of a person’s participation and no later than the end of data collection (American Psychological Association, 2017).

The duties continue after the data are in. The code forbids fabricating data, and it expects researchers to share their data with qualified colleagues who want to check the results (American Psychological Association, 2017). Figure 2.6 follows one participant through a study, from the IRB’s approval to the moment their data are stored.

Animals are protected too. The APA estimates that about 90 percent of animal research in psychology uses rodents or birds. Researchers must design studies to minimize pain and distress, and every animal study is reviewed by an Institutional Animal Care and Use Committee (IACUC) of administrators, scientists, veterinarians, and community members. The committee requires humane treatment and inspects animal facilities twice a year (Spielman et al., 2020).

Put together, this is what trustworthy psychology looks like: a clear question, a design that can actually answer it, honest reporting that others can check, and genuine care for every person who makes the knowledge possible.