By Rajiv S. Jhangiani, I-Chant A. Chiang, Carrie Cuttler and Dana C. Leighton, adapted by Marc Chao and Muhamad Alif Bin Ibrahim

In the late 1960s, social psychologists John Darley and Bibb Latané introduced a surprising idea: the more witnesses there are to an accident or crime, the less likely any one of them is to help the victim. This phenomenon, known as the “bystander effect”, occurs because each witness feels less personal responsibility to take action, a process called “diffusion of responsibility”.

Darley and Latané referenced real-world cases to support their theory, including the tragic murder of Catherine “Kitty” Genovese in New York. Despite multiple witnesses, no one intervened to stop the attack. However, the researchers recognised that single events like this could not conclusively prove their hypothesis. It was impossible to know whether fewer witnesses would have led to different outcomes.

To test their theory, Darley and Latané conducted a controlled experiment in a laboratory setting. Participants were isolated in separate rooms and told they would discuss university life over an intercom system. During the conversation, one of the supposed students (actually a pre-recorded voice) began to simulate an epileptic seizure, pleading for help with increasingly desperate and choking sounds.

Participants were placed in one of three conditions:

1. They believed they were the only witness to the emergency.
2. They thought one other person was also listening.
3. They assumed four other people were part of the discussion.

The results were striking: the more witnesses participants thought were present, the less likely they were to help. In the one-witness condition, 85% of participants sought help. This dropped to 62% with two witnesses and 31% with five.

This experiment is a classic example of how causal relationships between variables are tested in psychology. By carefully controlling conditions and isolating the variable of “number of witnesses”, Darley and Latané demonstrated a clear cause-and-effect relationship. This study remains one of the most influential pieces of research in social psychology, illustrating the power of experiments to uncover how human behaviour is influenced by social situations.

What Is an Experiment?

An experiment is a research method designed to determine if there is a causal relationship between two variables. In other words, whether one variable directly causes changes in another. These two variables are referred to as the independent variable (the one that is manipulated) and the dependent variable (the one that is measured).

Experiments have two key characteristics.

First, researchers manipulate the independent variable by systematically changing its levels, which are known as conditions. For example, in Darley and Latané’s study on helping behaviour, the independent variable was the number of witnesses participants believed were present. The researchers created three conditions by telling participants there were either one, two, or five other students in the discussion. It is important to note that while there are three conditions, there is still only one independent variable, namely the number of witnesses, with three levels (one, two, or five). Beginners sometimes mistakenly think that each condition represents a separate independent variable, but this is not the case.

Second, researchers control extraneous variables, meaning factors other than the independent or dependent variable that might influence the results. In their study, Darley and Latané ensured consistency by testing all participants in the same room and exposing them to the same emergency scenario. They also randomly assigned participants to each condition to ensure the groups were similar at the start of the experiment.

While the words “manipulation” and “control” might seem similar, they have distinct meanings in research. Researchers manipulate the independent variable by changing its levels, while they control extraneous variables by keeping them constant across conditions. Together, these two features allow experiments to provide clear, reliable evidence of cause-and-effect relationships.

Manipulating the Independent Variable

Manipulating an independent variable means systematically changing its levels so that participants are exposed to different conditions. This can happen either by assigning different groups of participants to different levels of the variable or by exposing the same group to different levels at different times.

For example, if a researcher wants to study whether expressive writing affects health, they might ask one group of participants to write about traumatic experiences and another group to write about neutral experiences. These two scenarios represent different levels of the independent variable, often referred to as conditions. Researchers typically name these conditions for clarity, such as the “traumatic condition” and the “neutral condition”.

A key point is that manipulation requires active intervention by the researcher. Simply comparing groups that already differ on the independent variable is not manipulation and, therefore, is not an experiment. For instance, if a researcher compares people who already keep journals with those who do not, they are not manipulating anything. Pre-existing groups like these might differ in other meaningful ways, such as personality traits (e.g., being more conscientious or introverted) or stress levels. As a result, any observed health differences could stem from these factors rather than from journaling itself. Active manipulation is essential for ruling out alternative explanations and establishing a clear cause-and-effect relationship.

However, there are situations where manipulating an independent variable is not possible due to practical or ethical reasons. For example, a researcher cannot randomly assign people to have or not have early childhood illnesses to study their effects on hypochondriasis. While experiments are not possible in such cases, researchers can still study these relationships using non-experimental methods, which will be discussed later.

Two-Level vs. Multi-Level Designs

Independent variables can be manipulated to create either two conditions or multiple conditions:

• Single-Factor Two-Level Design: The independent variable has two conditions (e.g., a group with one witness vs. a group with five witnesses).
• Single-Factor Multi-Level Design: The independent variable has more than two conditions (e.g., one witness, two witnesses, and five witnesses).

In Darley and Latané’s bystander study, they used a single-factor multi-level design with three conditions: one witness, two witnesses, and five witnesses. This allowed them to observe patterns across multiple levels of the independent variable, offering richer insights than a simple two-level comparison would have provided.

Control of Extraneous Variables

Extraneous variables are any factors in a study that are not the independent or dependent variables but could still influence the results. In an experiment testing whether expressive writing affects health, examples of extraneous variables include:

• Participant variables: writing ability, diet, gender
• Situational variables: time of day participants write, whether they use a computer or write by hand, and even the weather

These variables are a problem because they can affect the dependent variable (e.g., participants’ health) in ways unrelated to the independent variable (expressive writing). For instance, someone’s health might improve because of their healthy diet rather than their participation in expressive writing exercises.

If extraneous variables are not controlled, they can make it hard to tell if changes in the dependent variable were caused by the independent variable or by something else.

To address this, researchers must control extraneous variables by holding them constant across all participants and conditions. For example:

• All participants could be asked to write at the same time of day.
• Everyone could use the same method of writing (e.g., all on a computer).

By keeping these variables consistent, researchers can reduce their influence and ensure that any observed effects on the dependent variable are more likely to be caused by the independent variable.

Extraneous Variables as “Noise”

Extraneous variables can make it harder to see the effect of an independent variable in two key ways. One major way is by adding “noise” or extra variability to the data, which can obscure meaningful results.

Imagine an experiment testing how mood (happy vs. sad) affects people’s ability to recall happy childhood events. Participants are shown a happy or sad video clip to set their mood, then asked to recall as many happy memories as possible.

In a perfect world with no extraneous variables, the data would be clear. Every participant in the happy mood condition might recall exactly four happy memories, and every participant in the sad mood condition might recall exactly three memories. The difference is obvious.

In reality, extraneous variables create variability in the data. For example:

• Some participants in a happy mood might recall fewer happy memories because they have fewer to draw on or are less motivated.
• Some participants in a sad mood might recall more memories because they naturally have better recall strategies.

Even if the average difference between the groups stays the same, the increased variability (or noise) makes the effect of mood harder to detect.

How to Control Extraneous Variables

One way to reduce noise is to hold extraneous variables constant:

• Situational variables: Test all participants in the same room, give identical instructions, and treat everyone exactly the same.
• Participant variables: Some studies control for specific participant traits. For example, language studies often limit participants to right-handed people because brain areas related to language are usually more consistent in right-handed individuals.

Balancing Control and External Validity

In theory, researchers could control extraneous variables by selecting only a very specific group of participants, for example, 20-year-old, right-handed, female psychology majors. This would reduce variability but also limit external validity, which is the ability to apply the results to a broader population.

For instance:

• Results from young, female participants might not generalise to older male participants.

In most cases, researchers aim for a balance: they reduce variability enough to detect meaningful effects while keeping the sample diverse enough to ensure the findings can be applied to a wider group of people.

Extraneous Variables as Confounding Variables

Extraneous variables can interfere with an experiment by becoming confounding variables. A confounding variable is an extraneous variable that systematically varies along with the independent variable, making it difficult to determine which one is actually causing changes in the dependent variable.

For example, in most experiments, participants’ IQ (intelligence quotient) is considered an extraneous variable because people naturally have different IQ levels. This variability is usually acceptable if IQ is evenly distributed across all experimental groups, meaning participants with lower and higher IQs are equally present in each condition. Figure 9.1.1 illustrates a hypothetical example: if participants in one group, such as a happy mood condition, have much higher IQs on average than those in another group, such as a sad mood condition, IQ becomes a confounding variable.

The word “confound” means to confuse, and that is precisely what confounding variables do. If participants in the happy mood condition perform better on a memory task, it becomes unclear whether their improved performance is due to their mood or their higher IQs. When two variables, such as mood and IQ, vary together, it becomes impossible to know which one is responsible for the observed outcome.

One way to avoid confounding variables is to hold them constant. For example, researchers could control IQ by only including participants with an IQ of exactly 100. While this approach prevents IQ from becoming a confounding variable, it significantly reduces diversity in the sample and limits how well the results can apply to a broader population.

A more practical and widely used approach is random assignment, where participants are randomly placed into experimental groups. Random assignment helps ensure that extraneous variables, such as IQ, are evenly distributed across conditions, reducing the likelihood that they will confound the results.

Treatment and Control Conditions

In psychological research, a treatment refers to any intervention designed to improve behaviour, mental health, or well-being. This can include therapies for mental health disorders, medical treatments, educational strategies, or programs aimed at reducing prejudice or promoting conservation. To test if a treatment works, participants are randomly assigned to either a treatment condition (where they receive the treatment) or a control condition (where they do not). If participants in the treatment condition show better outcomes, such as reduced depression, faster learning, or improved behaviour, researchers can conclude that the treatment is effective. Studies like these, especially in medical or therapeutic contexts, are often called randomised clinical trials.

Control conditions help researchers ensure that improvements in the treatment group are due to the treatment itself and not other factors. In a no-treatment control condition, participants receive no treatment at all. However, this approach has a limitation: placebo effects. People often experience improvement simply because they expect to feel better. This expectation can reduce stress, anxiety, and even improve immune responses, creating real changes despite the absence of an actual treatment.

Placebo effects are fascinating but problematic for research. Imagine a study where participants in a treatment group improve more than those in a no-treatment control group. Figure 9.1.2 illustrates hypothetical results where participants in a treatment condition show greater improvement, on average, compared to those in a no-treatment control condition. However, if these two conditions (represented by the two leftmost bars in Figure 9.1.2) were the only comparison points, it would be impossible to confidently conclude that the observed improvement was due to the treatment itself. It remains possible that participants in the treatment group improved primarily because they expected to do so, while those in the no-treatment control group lacked this expectation.

One way to address this is to include a placebo control condition. In this setup, participants receive something that looks and feels like the real treatment but lacks its active component. For example, if the treatment involves taking a pill, participants in the placebo group would take an identical pill without the active ingredient. In psychotherapy research, the placebo might involve unstructured talk therapy, where participants meet a therapist but without any structured therapeutic techniques. If both groups expect improvement, but the treatment group still shows greater improvement, researchers can confidently attribute the effect to the treatment itself.

Ethical guidelines require participants to be informed about whether they might receive a treatment or a placebo, although they will not know which one until the study ends. Often, participants in the placebo group are offered the real treatment once the study is over.

Another option is a wait-list control condition. Participants in this group know they will eventually receive the treatment but must wait until others have completed it first. This setup allows researchers to compare people currently receiving the treatment with those still waiting, but who also expect improvement in the future.

Finally, researchers might skip a control condition entirely and compare a new treatment with the best existing alternative treatment. For example, a new therapy for phobias could be compared with standard exposure therapy. Since both groups receive active treatment, their expectations for improvement are similar. This approach helps answer a more meaningful question: “Is the new treatment better than what’s already available?” instead of just “Does it work at all?”

The Power of the Placebo Effect

Many people expect placebos to help with psychological conditions like depression, anxiety, or insomnia. However, placebos have also been shown to improve conditions generally considered purely physical, such as asthma, ulcers, and even warts (Shapiro & Shapiro, 1999). Remarkably, there is evidence that placebo surgery, which is also known as “sham surgery”, can sometimes be as effective as actual surgical procedures.

In a notable study, medical researcher J. Bruce Moseley and his colleagues investigated the effectiveness of two arthroscopic surgery procedures for osteoarthritis of the knee (Moseley et al., 2002). In the control group, participants underwent the full pre-surgical process: they were prepped, given a tranquilliser, and even had three small incisions made on their knees. However, they did not actually receive the arthroscopic procedure.

It is important to note that the study’s use of deception would have undergone a thorough review by an Institutional Review Board (IRB). The IRB determined that the potential benefits of the study outweighed the risks and that no other method could have answered the research question about placebo procedures effectively.

The results were striking. All participants, including those in the placebo (sham surgery) group, showed improvements in both knee pain and function. In fact, the placebo group improved just as much as those who received the actual surgical procedure. As the researchers concluded, “This study provides strong evidence that arthroscopic lavage with or without débridement [the surgical procedures used] is not better than and appears to be equivalent to a placebo procedure in improving knee pain and self-reported function” (p. 85).

References