Whilst it is easy to say that a confounding variable is an extraneous variable that could provide an alternative explanation for our results, what do we actually mean by this? An extraneous variable becomes a confounding variable when the extraneous variable changes systematically along with the independent variable(s) that you are studying. There are two components to this: COMPONENT #1: There must be three or more variables involved and COMPONENT #2: These variables must change systematically with each other. Each is discussed in turn:
If there are only two variables involved in a study (i.e., one independent variable and one dependent variable), there cannot be any confounding variables. For a confounding variable to exist, there must be at least one additional variable (i.e., three variables in total, or more) involved. After all, it is this suspect third extraneous variable that may be a confounding variable. We illustrate this with an example shortly.
For this suspect third extraneous variable to be a confounding variable, it must change systematically with at least one of the other variables you are measuring (usually the independent variable). We talk about the third variable changing systematically because it must behave in a way that is similar to the variable(s) that you are intentionally studying. This makes it difficult to know whether the change in the dependent variable is the result of the independent variable that we are intentionally measuring, or the third, suspect extraneous variable. Let's look at an example.
The relationship between background music and task performance amongst employees at a packing facility
The initial results from our example experiment suggested that the use of background music improves task performance in the packing facility. However, let's imagine that we change the way that the original experiment was conducted. Previously, we suggested that the control group and treatment group were both measured at the same time, once every hour from the beginning of their shift to the end of their shift (i.e., a period of 8 hours). However, let's imagine that since all the employees in the packing facility work in one giant room, this makes it impossible to provide the treatment group with background music without the control group hearing the music. Since this would be a clear threat to internal validity, we change the experimental design. Instead of both groups being measured at once, we turn the music on for the first 4 hours of the shift, and then turn it off for the second 4 hours of the shift. We record the number of tasks performed correctly from the treatment group during these first 4 hours, and then record the number of tasks for the control group during the second 4 hours. However, in doing this, we have invited an extraneous variable, time of shift, into our experimental design.
Now imagine that the results from this revised experimental design are the same: the use of background music improves task performance, since the treatment group perform far more tasks in their 4 hour shift than the control group did in their 4 hour shift. The problem is that unlike before, we do not know whether task performance was improved by the introduction of background music, or the third, suspect extraneous variable: time of shift. The question arises: Why is this third, suspect extraneous variable, time of shift, a confounding variable?
The answer is that the variable, time of shift, changed systematically with the independent variable that we were measuring (i.e., background music). It changed systematically because as the time of shift changed from (a) the first 4 hours of the shift to (b) the second 4 hours of the shift, so the independent variable changed from (a) the introduction of music (i.e., the treatment group) to (b) the removal of music (i.e., the control group). Maybe the relationship that we are observing is not between background music and task performance, but time of shift and task performance. After all, we would expect that in a physically demanding job where employees are constantly 'on the move', tiredness would play a role in task performance. Furthermore, we would expect task performance would drop during the second 4 hour shift when the control group (i.e., no music) were being measured, compared with the first 4 hours when the treatment group (i.e., with music) were being measured. So did the control group simply perform worse in terms of the number of tasks performed because they were more tired than those employees in the treatment group? The answer is: We don't know. This is the problem, and it threatens the internal validity of the experiment; that is, it provides an alternative explanation for the relationship between the independent variable (i.e., background music) and the dependent variable (i.e., task performance).
When you perform quantitative research, the ideal is to be able to control all extraneous variables, with the exception of the independent variable that you are trying to manipulate. Whilst some of these extraneous variables can be controlled for (e.g., individual differences that can be accounted for through the division of employees into groups by random assignment during the sampling process), others are more difficult to control (e.g., certain environmental conditions that may be particularly difficult to plan for during the experimental design process). You can learn more about threats to internal validity in the article: Internal validity.
Dealing with extraneous and confounding variables in research can be achieved by identifying whether an extraneous variable might become a confounding variable and designing your study to include potential confounding variables. Each is discussed in turn:
Since studies can have hundreds or thousands of extraneous variables, it is important to learn how to distinguish between those extraneous variables that can become confounding variables, and those that play a relatively neutral role. When thinking about your own study, it may help to think about different types of extraneous variables; those that relate to:
Individual differences between participants (e.g., age, gender, salary, etc.).
The environment in which the study is conducted (e.g., the weather, the physical surroundings, etc.).
The independent variable (in the example experiment: type of music, loudness of music, time of day, etc.).
The dependent variable (in the example experiment: employee tiredness, employee motivation, job satisfaction, etc.).
Once you have identified these extraneous variables, you need to ask yourself the two questions posed in the previous section:
What academic evidence is that that the extraneous variables that you can identify affect the dependent variable?
Are they any logical or practical reasons to assume that an extraneous variable might become a confounding variable?
Click here if you missed this section.
When setting up your experiment, if you think that an extraneous variable (i.e., a variable that you did not want to study; or think about studying) might become a confounding variable, you have two options: OPTION #1: Design potentially confounding variables out of the experiment and OPTION #2: Include the potentially confounding variables within your experimental design. Each option is discussed in turn:
Design potentially confounding variables out of the experiment
Some extraneous variables can be controlled for by designing them out of the experiment. For example, individual differences between participants (e.g., age, gender, salary, etc.) may be confounding variables, but it is possible to account for these differences during the sampling process by assigning participants into different groups (e.g., the treatment and control group) according to such individual differences. For example, you could put an equal number of male and female participants into the treatment and control groups. Similarly, you could make sure that the two groups are similar in terms of the salary earned by participants.
When designing potentially confounding variables out of the experiment, you need to ask yourself:
What extraneous variables might become confounding variables in my study (i.e., the two questions you asked yourself previously: I've forgotten what these were).
Which of these potentially confounding variables is it possible to design out?
Since it is not possible to design out all potentially confounding variables, you may need to include these potentially confounding variables in your experimental design.
Include the potentially confounding variables within your experimental design
When you (a) manage to identify an extraneous variable that might become a confounding variable, and (b) are not able to design it out of your experiment; or you actually want to know its effect, you need to try and include this extraneous variable in your experimental design. Let's go back to our example experiment where we identified the time of shift as a confounding variable [remind me of this example in full].
The time of shift (i.e., first 4 hours or last 4 hours) changed systematically with the independent variable that we were measuring (i.e., background music) because as the time of shift changed from (a) the first 4 hours of the shift to (b) the last 4 hours of the shift, so the independent variable changed from (a) the introduction of music (i.e., the treatment group) to (b) no music being played (i.e., the control group). Therefore, we did not know whether the control group simply performed worse in terms of the number of tasks performed because they were more tired than those employees in the treatment group. After all, it makes sense that employees in physically demanding jobs get tired as the day goes on, which affects their physical performance (i.e., in this case, task performance).
To account for this, we could have chosen to measure employee tiredness for both the control group and treatment group throughout their 8 hour shift. By (a) examining how much employees tired during the day, not only in total, but between the two groups, and (b) assessing this in terms of the changes in task performance for the two groups, we would know whether the extraneous variable, time of shift, was a confounding variable or not. It may be that despite the fact that employees became more tired during the day, which meant that employees in the control group without the music were more tired when we measured the number of tasks they performed, this tiredness was not sufficient to alter the conclusion that the introduction of background music improved task performance. In other words, employee tiredness was not such a large problem that it provided an alternative explanation for our finding that the introduction of background music improved task performance.
Whilst this is just an example, it aims to highlight that by including (i.e., measuring) potentially confounding variables within your experimental design, you can examine whether they are actually confounding variables or not. You may even be able to examine what impact that they had on the dependent variable (e.g., how much tiredness decreased task performance compared to how much background music improved task performance).
To learn more about possible threats to internal validity in your quantitative dissertation, see the article: Internal validity.