Research Designs

Return to Behavioral Research Methods

Before you can run experiments, you have to design them. A strong design will provide meaningful results. Behavioral research is scientific research, like physics and biology. So it relies on empirical methods (which means using a combination of observations and measures to create data which describes a scenario). As such, behavioral research must be: The approximate design of research can be described by these steps: This section will address parts 1, 2, & 5. You should know the other 2 parts from learning about statistics.
 * 1) Systematic - events during the research must be planned and organized.
 * 2) Objective - facts only! The researcher's opinion should not affect the data (think about ethics).
 * 3) Precise - think about your variables in terms of how you will measure them (see Operational Definitions).
 * 1) Formulate a hypothesis
 * 2) Operationalize the variables and their relationships
 * 3) Collect data
 * 4) Analyze data
 * 5) Draw conclusions and interpret results

Hypothesis/Hypotheses
Statement of your predictions of how events or behaviors might be related. The wording of your hypothesis(es) depends on the type of study: Experimental or Non-Experimental.

Experimental
Involves manipulation of at least one variable. Only experimental studies can be used to determine cause and effect.


 * An example of an experimental hypothesis would be:


 * "Medicine A causes more deaths than Medicine B when treating for cancer."

This is an experimental hypothesis because you can directly manipulate which medicine the participants receive. Since there is a variable that is manipulated, you have an experimental hypothesis.

Non-Experimental
Involves the observation or measurement of variables (no direct manipulation). Cannot determine cause and effect (but can support the possbility).


 * An example of an experimental hypothesis would be:


 * "Eating healthy food is related to a person's body-mass index (BMI)"

This is a non-experimental hypothesis because you cannot directly manipulate what food a person eats (not ethically). Instead, you can find people with different diets and categorize them by their diet: Healthy, average, unhealthy. Since there is no direct manipulation of a variable, you have a non-experimental hypothesis.

Operationalizing
Operational definitions describe what you want to analyze in terms of how you will measure it.


 * For example, if you want to use a person's intelligence as a variable, you may use the IQ test. This means that you are not necessarily measureing intelligence, you are measuring the person's score on the IQ test. So, your variable is technically not a person's intelligence, rather it is their score on the IQ test. This means the operational definition of a person's intelligence is provided by the IQ test.

The hope is that the measure (i.e., the IQ test) is actually measuring your desired variable (i.e., a person's intelligence).

Collecting a Sample
There are many ways to collect a sample.

Random Sampling
Individuals from a population are randomly selected to participate in the research. Each person in the population has an equal chance of being selected.

Although behavior researchers like to call this 'random sampling' it is actually more of a pseudo-random sampling method. In a true random sample, each individual would have an equal chance of being selected. In behavioral research, we do not like to use the same person more than once in any given experiment. This means that once a person has done the experiment, they are no longer included in the subject pool. In turn, the subject pool becomes smaller and individuals remaining in the subject pool gain an increasing chance to be selected.

Stratified Sampling
A population is divided by characteristics that are important to the researcher. Then a sample is collected that has the same proportion of individuals with each characteristic as the population.


 * For example, if 20% of the people in a population of interest do not like hotdogs and your research has to do with hotdogs, then 20% of your participants should not like hotdogs.

Quota Sampling
Similar to 'stratified sampling', this is when the researcher decides how many people s/he wants in the experiment with certain characteristics.


 * For example, even though 80% of males play video games and 50% of females play video games, you may be interested in comparing people who play video games to people who do not. So, you want 50% of your participants to be video game players and 50% to not be video game players. It doesn't match the population but it matches your research.

Purposive Sampling
When you sample with a purpose in mind! This sample may not be representative of the population but it is a targeted group that fits your research.


 * For example, if your research hypothesis has to do with men who wear pink shirts, you would go find men who wore pink shirts instead of wasting time with those who do not.

Convenience Sampling
This is a sample of convenience. Maybe you're under time constraints or maybe you have a limited pool of participants or maybe you're low on resources. Whatever the reason, convenience sampling is when you just grab any joe shmoe you can find and use them as participants for your research.


 * If your research is about the average Joe, then this is not necessarily a bad thing.


 * If your research is about fighter pilots and you grab some average Joes, then this is a really bad thing.

Conclusions and Interpreting Results
When you run your analyses and find your statistical results, you will know if your variables are related the way you hypothesized (see the appropriate statistical analysis for how to determine significance).

If they are related the way you hypothesized, then congratulations! But don't get too excited. It is good that they were related in your research but it may not be so in other situations. You should suggest that further research be done to determine under what conditions that relationship exists.

If they were not related the way you hypothesized, then do not worry. It may be that those variables are not related in that way or maybe they are not related in the situation of your experiment. You could take a closer look at your experiment and determine what could change to produce that relationship OR you could revise your hypothesis to see if those variables are related in a different way.

Whatever you do, do NOT extend your results further than your study. You can suggest and address the possibility but do not claim your research applies to more than the confines of your experiment.


 * For example, if I did a study that found rats are highly motivated by cheese, I cannot say that people will also be highly motivated by cheese. That's an extreme example.

Ok, more realistically, pretend I did a study that found people can see light with wavelengths between 390 and 750 nano-meters (that's actually the range of the visual spectrum). I cannot now say that people can see that range in any condition. People can only see that range when they are outside, the sky is clear, and the sun is in the middle of the sky. To say that people can see that range in any condition (fog, artificial lighting, etc.) would be misleading and inaccurate (or more bluntly, lying).