When Researchers Assign the Conditions
In Observational Studies and Their Limits, you learned that researchers can record existing conditions and describe associations, but that an observational study alone does not establish cause. A different kind of investigation deliberately gives participants different conditions. That act of assigning conditions is the key feature of an experiment.
An experiment is a study in which researchers impose one or more treatments on individuals or other observational units and measure the responses. A treatment is a specific condition applied to an experimental unit. Treatments might be different amounts of a product, different instructional methods, or a treatment compared with no treatment or a standard condition.
The individuals or other units that receive treatments are called experimental units. When the units are people, they are often called subjects. The explanatory variable is the factor whose effect the experiment is designed to investigate; its assigned values or categories define the treatment conditions. The response variable is the outcome measured to assess the effect of those conditions.
A study’s label depends on what investigators do, not on whether they call it a “trial,” “test,” or “study.” If people choose whether to use an app and researchers record their scores, the researchers are observing existing choices. If researchers assign some people to use the app and others not to use it, they are imposing treatments.
Why an Experiment Can Support a Cause-and-Effect Conclusion
Imposing a treatment creates a direct comparison: researchers can compare the responses of units assigned to different conditions. However, merely imposing different treatments does not automatically make every comparison convincing. If units are placed into groups in a way related to their characteristics, those characteristics may be tangled with the treatment.
For instance, if the most experienced participants receive a new training program and the least experienced participants receive the usual program, a later performance difference could reflect experience, the program, or both. This is a potential confounding problem: the effect of the training program cannot be cleanly separated from the effect of experience.
Random assignment uses chance to place experimental units into treatment groups. Because the assignment is not chosen according to participants’ preferences or characteristics, it helps make the groups comparable at the start of the experiment. Random assignment does not guarantee that groups will be identical. Instead, it tends to distribute both known and unknown factors among the groups, reducing the chance that a systematic pre-existing difference explains the outcome.
When an experiment uses random assignment and is carried out appropriately, a difference in responses can reasonably be attributed to the imposed treatments rather than to a systematic difference that determined who received each treatment. This is why an experiment can provide evidence about cause and effect in a way an observational study alone cannot.
That conclusion has limits. The experiment supports a causal conclusion about the units studied and the conditions tested. Generalizing the result to a broader population depends on how those units were selected. Random assignment and random selection answer different questions: assignment concerns how units enter treatment groups; selection concerns how people or units enter the study.
A Quick Design Check
When a scenario is not immediately clear, focus on the sequence of decisions. Ask whether investigators set the conditions or simply recorded choices and circumstances that already existed. Then identify what was assigned and what was measured.
Name the people, objects, or other observational units involved.
Ask whether investigators deliberately assigned a treatment, or merely recorded an existing behavior, exposure, or category.
State the conditions being compared and the outcome measured for each unit.
Random assignment strengthens a causal interpretation. If units chose their own groups or were assigned by a nonrandom rule, possible differences between groups can complicate the conclusion.
Describe an observed association for an observational study. For a well-conducted experiment with random assignment, a cause-and-effect conclusion about the treatments is supported.
Worked Example: A New Study Routine
Worked Example: A New Study Routine
A teacher wants to compare a brief daily retrieval-practice routine with the class’s usual independent review. In one plan, students choose which routine to use for two weeks, and the teacher compares their quiz results. In a second plan, the teacher uses chance to assign students to one of the two routines, then gives everyone the same quiz. For each plan, classify the study, identify the treatment and response, and explain what conclusion is supported.
Plan 1: Students choose. The experimental units would be the students, the possible routines are the conditions of interest, and quiz score is the response variable. But the teacher did not assign students to routines. Students chose their own condition, so the teacher is recording existing choices. This plan is observational, even though the routines happen before the quiz.
Students who choose retrieval practice could differ from students who choose usual review in motivation, prior achievement, or available study time. Those differences could be related to quiz scores, so the observed score difference would not establish that the routine caused a change.
Plan 2: The teacher assigns by chance. The teacher imposes the two routines and measures the same response, so this is an experiment. Random assignment helps make the groups comparable before the routines begin. If the retrieval-practice group has a higher average quiz score, the results would support the conclusion that the assigned routine caused higher scores for these students under these conditions.
This conclusion does not automatically apply to all students. The scenario does not say that the class was randomly selected from a wider population. Random assignment supports the causal comparison between treatments; it does not establish broad representation.
Worked Example: Plant Growth and Light
Worked Example: Plant Growth and Light
A greenhouse team is interested in whether a longer daily light period affects the growth of seedlings. In Scenario A, the team measures the light exposure that seedlings happen to receive in different parts of the greenhouse and records their growth. In Scenario B, the team assigns similar seedlings at random to receive either 8 hours or 12 hours of light each day, then measures each seedling’s growth over the same period. Which scenario is an experiment, and what can each show?
Classify Scenario A. The team observes light exposure that already occurs in different greenhouse locations. It does not assign the exposure, so Scenario A is observational. It could show an association between light exposure and growth among the seedlings measured. But locations might differ in temperature, moisture, or other growing conditions. The association alone would not show that the light period caused a growth difference.
Classify Scenario B. The team deliberately assigns each seedling to one of two light-period treatments. The seedlings are the experimental units; 8 hours and 12 hours are the treatments; and growth is the response variable. Because the team assigns treatments at random, Scenario B is an experiment that can support a cause-and-effect conclusion about light period and growth under the conditions of the experiment.
The team should keep other procedures, such as watering and measurement timing, consistent across treatment groups. That helps avoid introducing additional differences that could cloud the comparison. Random assignment is a strength, but careful implementation and a clearly defined response also matter.
Worked Example: An App and Walking
Worked Example: An App and Walking
A community health team compares two ways of investigating whether a walking-reminder app affects daily steps. In one investigation, volunteers who already use the app report their average daily steps, and the team compares them with people who do not use it. In another, the team assigns volunteers at random either to use the app for four weeks or to continue without it, then records each person’s steps during the final week. Explain the design difference and the conclusion each investigation can support.
Existing app use. In the first investigation, people already chose whether to use the app. The team measures app-use status and steps but assigns no treatment. This is an observational study. It can show whether app users and nonusers in the data differ in their reported average steps, but it cannot establish that the app caused a difference. For example, people who use the app might already be more interested in walking.
Assigned app use. In the second investigation, the team imposes the app-use and no-app conditions. The volunteers are the experimental units, app assignment is the treatment, and steps during the final week are the response. Since assignment is random, a difference in steps between the groups supports a causal conclusion about the app for the volunteers studied, assuming the investigation is conducted as planned.
Neither scenario describes random selection from all community members. Therefore, even the randomized experiment does not by itself show that the same effect would occur for every person in the community. The causal question and the generalization question must be considered separately.
Common Mistakes and AP Exam Tips
- Calling a study an experiment because it compares groups. Observational studies can compare groups too. The defining question is whether investigators imposed treatments.
- Treating “happened before” as equivalent to “caused.” A behavior may precede an outcome in an observational study, but other differences between groups may still explain the association.
- Assuming any assigned treatment guarantees a strong causal conclusion. Random assignment is important because it helps make groups comparable. A nonrandom assignment rule may leave groups different in ways that affect the response.
- Confusing assignment with selection. Random assignment concerns who receives which treatment; random selection concerns who enters the study. Explain which one the scenario actually uses.
- Claiming that random assignment makes groups identical. It does not. It helps distribute differences by chance, but chance can still produce some imbalance.
- Generalizing farther than the design allows. An experiment with volunteers can support a causal conclusion for the volunteers under the tested conditions, but without random selection it does not automatically represent a wider population.
For full credit, state the design and connect it to the conclusion. For example: “Researchers randomly assigned the students to the two study routines, so this is an experiment. The difference in quiz results supports a cause-and-effect conclusion about the routines for these students. Because they were not described as a random sample, the result should not automatically be generalized to all students.”
Check Your Understanding
For each situation, decide whether researchers impose treatments, identify the response when possible, and state what conclusion the design can support.
- A researcher records the type of breakfast students choose and their alertness ratings later that morning. Is this an experiment or an observational study? Why?
- A coach assigns athletes at random to use one of two warm-up routines and compares their sprint times. Identify the experimental units, treatments, and response variable.
- People who already own air purifiers report their indoor air-quality readings, which are compared with readings from people who do not own one. Why can this comparison not establish that owning a purifier caused a difference?
- A school randomly assigns students to two homework formats, but does not randomly select students from a larger population. What kind of conclusion does random assignment support, and what does it not establish?
- In one sentence, explain the difference between random assignment and random selection.