Main Difference
The two are distinguished based on whether there is a direct manipulation or control of variables. In a nonexperimental approach (e.g., observation or survey studies), variables are observed or measured as they occur naturally. In the experimental approach, one variable (independent variable-IV) is manipulated, and then, the other variable (dependent variable-DV) is interested in will be measured.
Although the relationship between variables observed from an experimental study is often interpreted as a cause-and-effect relation (i.e., IV cased the DV), a relationship observed in a non-experimental study cannot be interpreted as a cause-and-effect relation (see below).
Limitations of Nonexperimental Research
- When researchers use a nonexperimental approach, even if they find two variables that are co-varying, they cannot determine the direction or casality of the two variables. In other words, you cannot determine whether there is a clear cause and effect relation (more critical issue).
- For example, let’s say you measured how much exercise people do and how anxious they are, and found that there seems to be a strong relationship between the two (the more you exercise, the less anxious you are).
- However, there could be at least possible directions of causality:
- Variable X could cause variable Y (e.g., more exercise could cause less anxiety)
2. Variable Y could cause variable X (e.g., more anxiety could cause less exercise)
3. There is no direct relation between X and Y. But there is a third variable that could be causing both X and Y to behave in a certain way (e.g., Socioeconomic status, especially income, of a person make the degree of exercise and depression behave in the observed specific manner).
Case Study: Nonexperimental Research
Camacho et al. (1991) performed a longitudinal study and measured physical activity and depression of a sample from Alameda County in California. Subjected reported their exercise level through a self-report measure and their depression was measured in a follow-up.
The researchers found that “Associations between 1965–1974 changes in activity level and depression in the 1983 follow-up suggest that the risk of depression can be altered by changes in exercise habits, although these associations were not statistically significant after adjustment for covariates.”
This research case suggests that a relationship observed in nonexperimental research could have been due to third variables.
Experimental Research Approach
In an experimental approach, one variable (IV) is manipulated. For example, a group of participants is asked to do some exercises (the top panel of the figure below) while the other group of participants is not asked to do or prevented from exercise (the bottom panel).
After the manipulation of the IV (i.e., exercise – with exercise vs. no exercise), the other variable (DV, anxiety) that the researcher is interested in will be measured.
Most importantly, any differences between groups in terms of the DV (anxiety) will be attributed to the manipulation of the IV (exercise).
Control of Extraneous / Confounding Variables in Experimental Research
In the experimental approach, as the DV difference will be attributed to the IV, it is critical to control other possible variables that could also explain the DV differences. For example, what if the exercise group was not only just an exercise group but also a rich group (see below)?
Then, now the researcher cannot tell whether the anxiety score difference between the groups is due to the exercise itself or their socioeconomic status difference. In this case, the extraneous variable that varied along with the IV is called a confounding variable and it ruins the underlying logic of the experiment.
Randomization
To avoid confounding variables, it is critical to randomly assign subjects to different experimental groups so that the extraneous variables vary in a non-systematic manner. This process is called randomization. As a result of the randomization, participants with various characteristics will be (assumed to be) distributed equally in the two groups.
Theoretically, the individual characteristic composition of the two groups will be virtually identical in every way. Then, the infinite extraneous variables are just as likely to affect one group as they are to affect the other group. Therefore, extraneous variables cannot be the cause of any systematic differences between conditions and the researcher would conclude that the difference in DV was due to the IV.