Chapter 4 Causality

4.1 Experimental, Quasi-Experimental, and Non-Experimental Studies

Research psychology is a process of identifying constructs and describing how they relate to other constructs. We can classify research designs as experiments, quasi-experiments, and non-experiments.

Experiments are the only kind of research that shows causal relationships (that is, that construct A causes a change in construct B). So an experiment could show if smoking causes lung cancer. To do this, experiments need two things (or they are not experiments)

All experiments have a manipulation. This means that the experimenter changes something within the environment of the experiment (called an independent variable) to see if it causes a change in the outcome (called a dependent variable). For our smoking example, a manipulation would be assigning one group of participants to a lifetime of smoking and another group of participants to a lifetime of no smoking.

Experiments require random assignment. The experimenter decides when to vary the levels of the manipulation (change the manipulation) based on random assignment. Random assignment means that every participant has the same chance as being in one condition as another. For our smoking example, random assignment means each participant has a 50% chance of being in the smoking group.

As may be clear from the smoking example, we cannot always do experiments because of ethical (it would be wrong to assign people to smoke) or practical reasons (you cannot randomly assign people to genders, for example). The solution is a quasi- or non-experimental study.

In summary: experiments are powerful because they uniquely demonstrate causality (causal relationships). However, experiments require a manipulation and random assignment, which are not always possible.

In a quasi-experimental study, there is a manipulation but no random assignment. Whenever participants are assigned to levels of a manipulation non-randomly, the research is quasi-experimental. In a quasi-experimental smoking study, we could ask people if they had smoked before and assign them to smoking or non-smoking groups based on that answer.

In summary: quasi-experiments do not require random assignment, but they do not show casual relationships.

In a non-experimental study, no manipulation is done. If you want to look at the effects of gender on lung cancer, you would simply observe (collect data on) the genders of patients. By only observing, you would not be manipulating gender.

The differences between quasi- and non-experimental studies are sometimes slight (Pedhauzer & Schmelkin, 1991); if the researcher is manipulating an IV, then the work is quasi-experimental.

In summary: non-experimental studies are observational. Like quasi-experimental studies, they do not show causal relationships.

It’s worth repeating that only experiments demonstrate causality. Quasi- and non-experiments can show that a relationship exists but do not say whether one variable causes the other. Any non-causal relationship has three possible explanations:

  1. A \(\rightarrow\) B one variable causes another; in an experiment, this is the only explanation
  2. B \(\leftarrow\) A the relationship is reversed; the first variable is actually the outcome
  3. C \(\rightarrow\) A; C \(\rightarrow\) B a third variable exists that was not measured in the study; the third variable causes a change in both A and B. There are many ‘C’ variables, potentially.

In a non-experimental smoking study, you could not say whether smoking causes lung cancer or people who are predisposed to lung cancer are more likely to smoke. A third possibility is that a separate, third variable causes both lung cancer and a desire to smoke.

4.2 Demonstrating Causality

In the 19th century, John Stewart Mill said that we could be satisfied that a relationship is causal if the following three things could be demonstrated:

  1. The cause preceded the effect
  2. The cause was related to the effect
  3. We can find no plausible alternative explanation for the effect other than the cause

Experiments aim to identify causal relationships by manipulating something, observing the outcome, seeing a relationship, and using various methods to reduce other explanations.

4.3 Statistics and Causality

Statistics are an important tool for establishing causality, but it’s important to know that the choice of statistical technique does not affect the level of causal evidence; demonstrating causality is the job of the research design, not the statistics.

A common misconception arises from the term correlational research design, which people use as a label for quasi-experimental and non-experimental research. It is easy to confuse this term with correlation which is a statistical technique.

Recall that statistics has two branches: Descriptive stats provides tools to summarize variability. Inferential stats provides tools for generalizing samples to populations.

To demonstrate causality, we need to satisfy Mill’s second requirement. Inferential statistics can help us do that. Two techniques are particularly useful: correlation (and its statistic r) and the t¬-test (and its statistic, t). Next, we will see how these techniques work.

4.4 Correlation

In a correlation analysis, you measure two variables. Each variable needs to be continuous (fractional values allowed) or dichotomous (only two possible values). The correlation analysis will tell you if you have evidence that the variables are related. Related means that you could make a prediction of one value if you knew the other value.

Correlation answers the research question: How are two variables related (strength and direction)? Actually, it only looks for linear relationships. There are four possible relationships between two variables: Positively related variables: As one variable increases, the other variable also increases. Higher scores of variable x are associated with higher scores of variable y.

Negatively related variables: As one variable increases, the other variable decreases. Higher scores of variable x are associated with lower scores of variable y.

No relationship: As one variable increases, the other variable might increase or might decrease.

Non-linear relationship: As one variable increases, the other will change according to a function.

When you graph the two variables, they follow one of these patterns:

  • No Relationship
  • Positive Linear Relationship
  • Negative Linear Relationship
  • Nonlinear relationship

4.5 Two-Sample t-Tests

In a two-sample t-test, a researcher has two groups. Another way of saying this is that the t-test requires a discrete, dichotomous IV. Additionally, the t-test requires a continuous DV.

This analysis will provide evidence that two groups have different values of the DV. For example, if you are comparing an experimental drug with a sugar pill and measure patient outcomes on a 0-100 point scale (with 100 being a perfect outcome and 0 being the worst outcome), running a t-test can demonstrate that the sugar pill group and experimental drug patients had different scores.

4.6 Choosing the Right Statistic

Both correlation and the t-test can show us if two variables are related. Thus, which one is the right statistic depends on the type of measurement used in the study, NOT whether the study is an experiment, quasi-experiment, or non-experiment. If you have two continuous variables, you should use a correlation. If you have a discrete IV and a continuous DV, you could use either correlation or a t-test. You can use a correlation to analyze experiments, quasi-experiments, or non-experiments. You can use a t-test to analyze experiments, quasi-experiments, or non-experiments. A correlation analysis is not the same thing as a “correlational research design.” For this reason, “experiment, quasi-experiment, and non-experiment” are much clearer labels.

4.7 Ethics

It’s easy to develop research ideas in a vacuum. When your research questions involve thought or behavior, you may need to collect evidence from living people to answer your research question. Doing so requires consideration of research participants as human beings and not taking advantage of them. Although this may seem obvious, our discipline’s history is full of examples of researchers advancing science at the cost of health, wellbeing, and safety of research participants.

In 1979, the US Congress got involved, and the Belmont Report was released. The Belmont report describes the ethical framework for conducting research with human participants. From this origin, the following “big 3” of research ethics have evolved as fundamental principles for human-subjects research:

  • Respect for persons: Treat participants as autonomous individuals. Participants are free to choose to participate in a study. Participants are not subject to coercion.
  • Informed consent is an important part of the principle of respect for persons. Tell participants what the study is about and ask their permission to participate. Allow them to withdraw their consent at any time.
  • Beneficence: Do no harm. Balance risks to participants with benefits to participants. Note that benefits are not the same thing as compensation. Being cured of a disease is a benefit, being paid to participate in a study is compensation. Note, also, that a quest for knowledge in a study is a benefit to society and the researchers, it is not a benefit to the participants in the study. When considering risks and benefits, this principle applies to the individual participants.
  • Justice: Justice means fairness. The people who bear the burden of research should receive its benefits, and vice-versa.

Other ethical issues that can arise in research settings include:

  • Conflicts of interest
  • Fraud and accurate reporting of research results
  • Duplicate publication
  • Intellectual property and plagiarism
  • Use of nonhuman organisms in research

We will explore these issues in more detail later in our course.