Part 2: Research Methods (continued)

Research Designs

As stated previously, whether we use self-report or obserational methods to obtain data, both can be used in the research design of our choice. There are four different types of research designs.

• True Experimental
• Quasi-Experimental
• Correlational
• Descriptive

The textbook does a great job of explaining each of these individually, with Table 3.1 giving a nice overview. An important take-home point from these methods is that there are trade-offs associated with each approach. While true experiments give us the greatest ability to establish causal relationships (i.e., say for certain that X variable causes a change in Y variable), they are often the most difficult to conduct due to practical concerns, especially in social psychology. True experiments rely on random assignment, that is, participants must be able to be assigned to the independent variables of interest. While that can be done in some situations (i.e., if we want to study the impact that violent video games have on aggression, we can randomly assign participants to either play violent video games for 10 hours per week or non-violent video games for 10 hours per week, then test their level of aggression), there are other situations where this isn't possible. For instance, if we want to study gender differences in aggressiveness, well, one problem you should be able to quickly identify...how do we randomly assign someone to a gender?! Well, we can't. This is where we have to rely on quasi-experimental designs. While this is as good as we can get in establishing causality on variables for which we can't randomly assign, it comes with limitations. We are more prone to internal validity violations (i.e., confounds) that may explain the results. Variables that may correlate with gender could be causing the differences we find in our dependent variable.

Sometimes, we are unable to establish causality, and all we can hope for is to determine if a relationship occurs between variables. For instance, if we're interested in how one's diet is related to their mental health, it can be very difficult (and expensive) to run a large-scale experiment and manipulate participants' diets, so researchers often rely on correlational methods to obtain this data. They may simply administer a survey that assesses one's diet and their mental health status, and then look for a relationship between the two. One thing to keep in mind, and you've probably heard this over and over, is that correlation does not equal causation. That is, we cannot tell the direction of the relationship here. Does poor eating lead to poor mental health? Does poor mental health lead to poor eating? Is there some third variable involved that influences both (e.g., experiences of trauma). This cannot be repeated enough. Correlation does not equal causation! Correlational research helps set the foundation for future research to examine, preferably in true and quasi experiments, but it alone is insufficient for establishing causal relationships.

Last, we have descriptive methods, which are as they sound, they simply describe a phenonenon of interest. These methods are often used as an exploratory first step in a research process. Rather than utilize inferential statistics, like t-tests and F-tests, which can allow us to make inferences beyond the sample we test, these methods rely on descriptive statistics, like means, frequencies, and percentages. Thus, these methods are not sufficient for making any claims beyond the sample that was used. These methods are not widely used, especially in the published literature, and are often reserved for pilot testing, or seeing if we're even on the right track before we invest time in a full-fledged study. For instance, if we were interested in studying the impact that seeing angry and happy facial expressions have on mood, before we even decide to run the study, we should proabably pilot test our facial expressions to make sure the angry ones actually look angry and the happy ones actually look happy before running a whole study using them. So we can show those faces to 20 participants and have them rate how angry/happy they all look. Then we examine the means to make sure that indeed, the happy faces look happy and the angry faces look angry. Then we can proceed with our actual study.

Research Settings

In addition to the considerations above, there are trade-offs in the setting in which research is conducted. Research can be conducted in the laboratory, where greater experimental control can be utilized. This can often limit the threats to internal validity (i.e., control other variables that may explain the results), however, the laboratory environment is artificial and the results may lack external validity (i.e., how generalizable the study results are). Field studies, where the research is conducted in a more realistic environment, will often lead to greater external validity, but more threats to internal validity. There's always a tradeoff to consider when conducting research. There is no such thing as the perfect research study. Using various methods in various settings is as good as we can get.