1.4 Types of study
Having dealt with the main types of variable that we can have, it is time to think about the main types of study we can perform. The most important distinction is between observational studies and experimental ones. There is also an intermediate type called quasi-experimental.
1.4.1 Observational studies
Observational studies are studies that contain only measured variables. They are sometimes also known as correlational studies. The researcher collects data in some study population about phenomena or constructs that interest them, and establishes how those phenomena or constructs relate to one another.
Observational studies are very important. They are the only practical way of studying some topics. However, you have to be very careful about how you interpret the results, because there are many threats to causal interpretation (see section 1.5).
In our observational study about physical activity and depression, say we find that people who do more activity have fewer depressive symptoms. One reason for this could be that physical activity reduces depressive symptoms. But there are many other possible reasons. For example, people with higher incomes both find it easier to schedule physical activity (perhaps they can afford a personal trainer), and feel less depressed. In this case, income is the real reason there is an association between the other two variables (we call this a ‘third variable’ explanation). Or, having more depressive symptoms might make it harder to keep up doing physical activity; that is, the causal arrow runs from depression to physical activity, not the other way. In short, there are many ways that inferring ‘physical activity reduces depressive symptoms’ from an observational study could be a wrong conclusion.
For this reason, when we report the results of observational studies, we avoid the language of cause and effect. In an observational study, we would talk about physical activity being associated with lower depressive symptoms, not of the effect of physical activity on depressive symptoms. Typically, we are really interested in causal questions, and our findings will be interpreted as claims about cause and effect (Grosz et al., 2020). The cautious language of an association is a convention to signal the tentative nature of causal conclusions from observational data.
In observational studies, we designate the variable we wish to explain as the outcome variable, and the things that might affect it as the predictors. Other variables, that are neither the outcome nor the predictors specified in our research question, but which we nonetheless might need to take into account, are referred to as covariates.
Observational studies can be roughly divided into cross-sectional ones and longitudinal ones. In a cross-sectional study, all the variables are collected simultaneously. For example, we survey people on their current level of physical activity, and their current depressive symptoms, on a particular date in July 2025. The association of interest is whether a person’s level of physical activity in July 2025 is statistically related to their depressive symptoms in July 2025, i.e. at the same point in time.
Longitudinal studies involve measuring things in the same people (or animals) at multiple points in time. Their strength is that you can examine how a change in one variable relates to the change in the other (or, how a predictor variable at one time is related to an outcome variable at a later time). For example, you could look at whether people who increased their level of physical activity in 2025 also experienced a decreased in depressive symptoms.
Longitudinal studies are often considered to be more informative about causality than cross-sectional ones. Many important third variables - childhood background or socioeconomic position - will probably not change much from one year to the next. So if a person starts doing more physical activity at some point in time, and around that time their depressive symptoms reduce compared to what they were, that is suggestive of a causal impact of activity on depression. Even so, we would still use the language of association rather than effect when describing the findings.
1.4.2 Experimental studies
An experimental study in psychology or behavioural science is a study which contains at least one manipulated variable. In fact, the condition is slightly more specific than this: an experiment is a study that: (i) contains at least one manipulated variable; and (ii) the manipulated variable(s) are the variable(s) of causal interest in the research question.
To make this clearer, if we manipulate physical activity by getting half of the participants to exercise four times a week, and the other never to do so, and the research question is whether physical activity causes a reduction in depressive symptoms, then we are conducting an experiment, also known as a randomized control trial (RCT).
Sometimes, however, we can have a manipulated variable but our study is still not an experiment. Let’s say we are interested on the relationship between childhood trauma and depressive symptoms in adulthood. Childhood trauma is measured by a retrospective questionnaire. Depressive symptoms are measured by a symptom checklist. Worried that it might make a difference which order the participants complete this measure in (talking about your depressive symptoms beforehand could colour the way you remember your childhood), we assign half of our participants to do the childhood trauma questionnaire first, and half to do the depressive symptoms checklist first. In this study, then, there is a manipulated variable: order with levels {childhood trauma questionnaire first, depressive symptoms questionnaire first}. But, the study is not an experiment, because the thing whose causal influence our research question concerns, childhood trauma, is measured rather than manipulated. In fact, the very same study would be an experiment to address the research question ‘does filling in a depressive symptom checklist affect the way you recall chldhood trauma?’. But it is not an experiment with respect to the question ‘does experiencing childhood trauma increase adult depression?’. There is probably no ethical or practical way of doing an experimental study on that question in humans.
The term experiment is used in a much looser sense in many other scientific disciplines (molecular biology, chemistry, physics), to refer to any set of measurements that takes place under highly controlled or standardized conditions. Look up Thomas Young’s famous ‘double slit’ experiment, for example: this would be considered an observational study rather than an experiment in psychology.
I think it is really important to maintain the strict sense of experiment in psychology and behavioural science, and never to use the term loosely. One of the main things that psychology and behavioural science have going for them, amongst the disciplines that deal with humans, is that they are strong and sophisticated in thinking about causation (see section 1.5). In large part this is due to their commitment to conducting experiments (in the sense used here), and to distinguishing between inferences that arise from manipulating variables, and inferences that arise from measuring them. Therefore, we should reserve the term experimental for the cases where it is justified.
In an experiment, we call the variable(s) that we manipulate the independent variable(s) or IV(s). We often call different levels of an IV experimental treatments even if they are not actually medical treatments. We usually use random assignment to determine which treatment a participant receives on a particular occasion. This is because we don’t want the set of people receiving one treatment to differ in any systematic way from the set of people receiving the other treatment(s). The best way of protecting ourselves from any systematic difference is to entrust the assignment to chance.
The outcome variable measured in an experiment is called the dependent variable(s) or DV(s). As with observational studies, we refer to variables that are neither IVs nor DVs as covariates.
1.4.3 Quasi-experimental studies
Sometimes a variable varies in a way that was not actually manipulated by the experimenter, but nonetheless, different groups experience different levels of it somewhat by chance. A study of such a situation is called a quasi-experimental study.
In 2008, the US state of Oregon had more people that wanted to join Medicaid than it had capacity to accept. (Medicaid is a government-run health insurance programme for people on low incomes.) It therefore conducted a lottery. About 90,000 people applied and 10,000 of them, chosen at random, were given Medicaid places. Finkelstein et al. (Finkelstein et al., 2012) compared the health, health care utilization and financial strain of people who had been successful in the Medicaid lottery, and people who had entered but been unsuccessful. The researchers had not manipulated the IV (the lottery was happening anyway), but, on the other hand, assignment was as good as random. (In this case, it was actually random. In many other cases of quasi-experimental studies, it is ‘as good as’ random; that is, it involves some chancy processes that have little to do with the properties of the people who end up in the different groups.)
In a quasi-experiment, the critical feature is that the groups receiving different levels of the IV do not differ systematically, even though they were not created by the experimenter. It is important that they do not differ even in having volunteered (versus not) to receive the treatment. In the Medicaid example, it is important that both groups consisted of people who had come forward to enter the lottery. Comparing people who won a Medicaid place through the lottery to people who had not entered the lottery would not do, even if the groups were matched for income and other demographic factors. Entering the lottery is already a systematic difference between the groups; it might reflect all kinds of things, such as greater motivation, greater need, or better information.