I was recently asked by a student whether it is possible to do a research project without a control group, and I am copying my answer here, in case it is of interest to readers of the blog. The TL;DR is that you can, but if you need a more detailed explanation, you can find it in the paragraphs that follow.
In this article you will learn:
- What a control group is;
- Whether a control group is always necessary;
- Why omitting the control group could be a problem;
- When you might consider avoiding a control group.
Disclaimer: The primary audience to whom this article is addressed is students doing coursework in applied linguistics, TESOL/ELT and education and similar fields. Chances are that the content applies equally well to other research areas, but it is possible that there are disciplinary differences that you need to be aware of. If you are working in different fields, you will probably need to discuss this matter further with your research supervisor.
What is a control group?
When we do experimental research, it is common practice to divide our sample in two groups. One group (the ‘intervention‘ or ‘experimental’ group) is the people on whom we test whatever new idea or method we want to measure. The other group, or control group, usually goes about doing whatever they did, except they get tested in the same ways as the intervention group. This provides us with some ‘baseline’ information about what ‘normal’ outcomes are, which we can then compare with those in the actual intervention group.
You can read more about experimental design, control and intervention groups by following this link.
Is a control group necessary?
Although a two-group design is the simplest and probably most common experimental format, your research questions might suggest a different approach. This was why one of my students came up to me with the following question:
When can I omit or not use the control group in my research, and can I use more than one experimental groups instead? Sometimes we have to do this in Education. When and why?
Such an approach, which uses two experimental groups and no control group is indeed possible, and sometimes appropriate. For example, it is a cost-effective way to measure the effectiveness of two activities, coursebooks or teaching approaches. I will use the term ‘variables‘ to describe these things that we are interested in comparing.
Provided that our sample is large enough, and we group membership randomly, we can assume that all the experimental groups are more-or-less similar in most respects, except for the variables that we are comparing. If we then apply a different intervention to each group, we can reasonably assume that it is the different variables that are causing any differences in learning outcomes.
An example of a study with no control group
Suppose we want to improve our students’ fluency in a foreign language. It seems plausible that the more input learners are exposed to, the more fluent they become. However, if we only have budget for either a new library or new audiovisual equipment, we should find out whether it’s more effective for students to spend more time reading or listening.
To find out more, we could assign students to two intervention groups, one of which would spend an hour a week watching TV programming in a foreign language, while the other would spend an hour a week reading books. By measuring the students’ fluency before and after the interventions, we’d be able to make an evidence-based claim about which method is more effective.
Are there problems with omitting a control group?
Research designs without a control group are useful for answering many simple questions about language teaching and learning, but they come with an important limitation: We cannot be certain that it is our interventions that are producing whatever outcomes we measure.
In the example above, it is possible that students became more fluent because they had more practice speaking at school; or perhaps they became more confident as they grew older; or maybe they just practiced more, because they knew that were being tested (a phenomenon known as the Hawthorne effect). So, while we might claim that watching television proved more (or less) strongly associated with developing fluency compared to reading, we cannot claim with equal confidence that “if you want to become more fluent you need to watch more television”.
What’s the verdict?
So then, when should we use more than one experimental groups without a control? Ideally, you would use this design only when there is enough evidence in the literature that the intervention you are making is associated with some outcomes, and you are therefore only interested in comparisons. For example, you might know that method A works, and you also know that method B works, so you just want to find out which of the two is more effective. This is something that you will have to establish in the literature review.
If you cannot find enough evidence in the literature to establish this connection, but there are also insurmountable pragmatic reasons why you cannot use a control group (e.g., because you don’t have access to enough participants), then you would have to acknowledge the lack of a control as a limitation of the study when reporting your findings.
If you landed on this page looking for information that will help you with your research project, I hope that the content of this article was of some use. I am happy to answer any other questions you might have, if you’d like to share them as a comment under this post or by contacting me directly. Please feel free to use the social sharing buttons at the bottom of the page to forward this article to anyone who might find this information useful.
- Seliger, H. W., & Shohamy, E. (1989). Second Language Research Methods. Oxford: Oxford University Press. pp. 140-141.
- Brown, J. D., & Rodgers, T. S. (2002). Doing Second Language Research. Oxford: Oxford University Press. pp. 211-213
- Dörnyei, Z. (2007). Research Methods in Applied Linguistics: Quantitative, Qualitative, and Mixed Methodologies. Oxford: Oxford University Press. pp. 116-117
About this post: This post was originally published in December 2013, in response to a student query. It was revised, for reader-friendliness, in February 2020.