Student FAQs in Psychology : Lesson # 1 - Random Assignment versus Matched Groups in PsychologicalExperiments

I, Jyotika Varmani, tutor students of Psychology at all levels. I reside in Mumbai and tutor students online. You can contact me personally on my e-mail id jyotikapsychology@gmail.com or call/message me on 9892507784 for enquiries.

---

The myriads of questions that students pose to their tutors are more than satisfying for them, for they reflect students' curiosity and their desire for knowledge. Among these countless questions that are impossible to keep track of, are a few that the tutor cannot help but notice as being distinctive, for they come to her so often that they enter her personal collection of FAQs that she then anticipates and incorporates into her lessons. My own lessons have benefited from quite a few such questions themselves and in this series, I present a selection of those questions and answers that have made their way into my personal repository of tutoring.

In this first post of the FAQ series, I address a query on psychological research, full of misconceptions and half-understood ideas - something that's very typical in student FAQs.

---

Q. Matching participants on experiments assures that any differences measured in them with respect to the dependent variable is solely due to the manipulation of the independent variable….because they were equal before the experiment began. But the teacher says that random assignment is superior. I don't see how...

My Response.

To begin with, this question is asked to me by both, high school students belonging to international boards and students pursuing their Masters in psychology. Since students of all age groups fail to gauge the purpose of using random assignment in experiments and that of matching participants in the same, it gives me the impression that these purposes are never made explicit in lectures. Moreover, the foundation on which these concepts lie - random assignment and matching as methods of creating comparable groups in independent groups design - is also not set out firmly.

Let's begin by understanding the need for creating comparable groups in independent measures design. Here, I begin with the assumption that you already know what an experimental design is and particularly, what the independent groups design is. 

Once an experimenter chooses to employ an independent groups design, that is, to subject different groups of participants to different levels of an independent variable, he needs to assure that the groups are equivalent to begin with, with respect to what are called individual differences or participant characteristics. Individual differences or participant characteristics refer to those characteristics of participants that they possess, which can influence their outcome on the dependent variable of an experiment. Take for example, an experiment to study the impact of the rate of the reading out of a string of letters on subsequent memory for them. For this experiment, if participants to be employed already differ with respect to their memory capacities, then this individual difference between them can severely affect their outcome on their recall for the letters, if the experimenter happens to create two groups in a way that one of them has more participants who have a relatively higher memory capacity than the other. Likewise, if the participants differ with respect to how motivated they are to perform well on the experiment, then this individual difference between them can also severely affect their outcome on their recall for the letters, if the experimenter happens to create two groups in a way that one of them has more participants who have relatively higher motivation than the other. Such significant impact of variables within participants that directly or indirectly influnce their outcome on the dependent variable of an experiment necessitates the creation of comparable groups by the experimenter when he chooses to employ an independent groups design.

Before we look into what random assignment and matching are and how they work, it is critical to note that these techniques are employed to create comparable groups, meaning equivalent and not equal groups. That is, they are used to create groups that are more or less equal with respect to certain characteristics but not exactly identical with respect to them. Keep in mind that psychological variables can never be found equally among individuals - no two individuals can have exactly the same standing with respect to a psychological variable. Hence, equating groups exactly on a participant-by-participant basis is simply not possible. Besides, when we compare groups in experiments, we compare group performances and not the performances of individual participants against each other in those groups. Therefore, as long as the groups created are roughly equal on an average, with respect to individual differences, then our purpose of comparison is served.

With this background in mind, let's understand what random assignment is. Random assignment is the allocation of participants employed in an experiment into groups, in an unpredictable fashion, that is, on the basis of probability. Simply put, this is the assignment of participants to different conditions of the independent variable on a purely 'chance' basis, with every participant as likely as another to be subject to either of the conditions of the independent variable. This makes the assignment of any one, particular participant, totally unpredictable. 

Contrast this with systematic assignment of participants - say on the basis of their order of entry in an experiment, or on the basis of their gender, or in an odd-even pattern. In such cases, the level of the independent variable that a given participant ends up in is attributable to some condition, which opens up the possibility of a relationship between the condition of assignment and differential allocation of participants to conditions. For example, assigning the first half of participants to enter the experiment to one condition and the second half to another opens up the possibility that the first half participants are more motivated to perform well on the experimental task than the second half. Their motivation might be the very reason for their entering into the experiment before the other participants. Motivation is one extraneous variable that can confound any experimental task which is why such assignment is bound to harm the results of any experiment. As another example, take the possibility that a significant number of participants entering in even positions in an experiment do so on the basis of wanting to participate in whatever activity the individual before them is participating in, that is, for wanting to pair up with that person. Such a characteristic of having a strong desire for associating with someone might influence performance on an experimental task, particularly if the task is related to personality or relationships. 

To avoid all such, often unfathomable possibilities of confounding, random assignment is employed. Random assignment works on the principle of breaking down patterns or systems in the allocation of participants. The nature of random assignment is such that it simply does not allow for a pattern to influence or to develop in the allocation of participants. Since by this technique, every participant has an equal chance of belonging to any condition of the independent variable, then it is simply impossible for say, every odd numbered participant to enter a particular condition or all beginning participants to enter a single condition. Extended to the systematic assignment of participants with different characteristics to different conditions of the independent variable, this manner of assignment breaks any potential pattern of differential assignment.

What is very helpful in the case of random assignment, is that it deals with not any one particular possibility of systematic confounding coming from within participant, but all such possibilities - those imagined by and those unimaginable to the researcher. Participants in any experiment vary greatly with respect to factors that have the potential of influencing their performance on whichsoever task might be allocated to them - motivation, mood, current state of health, alertness, etc. - to name a few. Further, they also differ with respect to factors that can influence their performance on a given experimental task. For example, in an experiment measuring memory, participants would differ with respect to innate characteristics such as attention, speed of processing, use of memory in everyday life, etc., all of which would influence their performance on a memory task in an experiment. Imagine a researcher having to assure that participants were similar to each other on all these counts! That is a definite impossibility. Random assignment ensures that participants in all conditions of an experimental task are equated, on an average, with respect to all possible participant variables that could influence their performance - the logic being that in exactly the same way that random assignment breaks down any systematic pattern of allotment of participants to conditions in the case of one innate variable, it does so for the case of any other innate variable.

While the above discussion would make it appear like random assignment is a miraculous solution to control all possible individual differences among participants, there are two primary barriers to its effectiveness. One is the way in which participants are selected for the experiment in the first place - a topic that is out of scope of the present discussion; the other is the strength of participants in an experiment, delving into a discussion of which will lead us to the topic of matching groups. 

Random assignment works in the way discussed above only when there are a sufficiently large number of participants to be alloted to various conditions of the independent variable. What number justifies this 'sufficiently large' condition is debatable, with several rules of thumb - ranging from as less as 10 to as many as 50 available for the same. There are, of course, sophisticated ways of calculating the numbers required, too. Keeping this point of discussion aside, why a high strength and not a low one of participants works for random assignment is because random assignment averages out the indivual characteristics in any group. And for averages to be unbiased, and truly representative of the data they are generated on, there is a need for a fairly large amount of data in calcualting them.

Following this logic, in the case of small samples, random assignment would not prove to be useful - for it would be biased according to extreme cases of individual differences present among participants. Therefore, in the case of small samples, what is often done instead, is the adoption of the technique of matching. Besides the inefficiency of random assignment with small sample sizes, matching is actually only feasible in the case of small samples. In the case of large samples, matching is often too laborious, given that it requires thorough pre-testing of some sort; and more importantly, unnecessary in most cases, given that random assignment is by all means, the more desirable alternative for exerting control. 

Matching is principally different from random assignment. While random assgnment seeks to balance out all possible innate characteristics of the participants of an experiment, matching attempts to hold such characteristics constant. Then, any differences in the performance of groups can be assigned solely to the difference in their treatment. What is immediately obvious here, given our prior discussion for the need of random assignment, is that there are far too many participant characteristics that have the potential of influencing the performance of participants and matching would be practically limited to a very few of them. What researchers typically do, is that they match participants on the very dependent variable that their experiment is designed to assess. For example, if intelligence is the dependent variable of interest in an experiment, then a pre-test measuring the intelligence of participants would form the basis of matching them. If the experimenter attempts to go beyond this and match participants with regard to related variables such as educational background and socioeconomic status,  both of which are known to affect intelligence, then matching would become increasingly complex because having two participants equated within some range of one variable is still possible but having them equated on two, and then three, and then four…..you can imagine how increasingly impossible this would shape up to be!

For most practical purposes, matching is confined to a single variable - typically the dependent variable of interest or some variable so strongly related to it that its matching between participants is a necessary precondition to testing the dependent variable. For example, take reading comprehension as the dependent variable of some experiment. Here, it is absolutely essential for researchers to match participants with respect to educational background, for a sixth-grader simply cannot be expected to be comparable in reading to a tenth-grader. If the researcher is to conduct an intervention that enhances reading comprehension, he has to ensure that there is a matched composition of sixth, seventh, eighth-graders...and so forth, exposed to all conditions of the independent variable.

Matching, given its limitation, is used only in certain cases where it does serve to offset the need for random assignment to at least, some extent. One is the case of small samples. As noted earlier, random assignment fails to work with small samples, given the extremities of participant characteristics present in them. In such cases, it is not advisable to use random assignment and take a chance of it working out favourably for the experiment. Matching is the superior alternative. As a foundational example, if there are four participants in total in an experiment, with two having high intelligence and two having lower intelligence, then there is strong chance that random assignment would place the former two in a separate condition of the independent variable and the latter two in a separate one. It is advisable for the experimenter in such a case to go ahead and himself place one participant of each type in each condition of the independent variable. Of course, which among the pair of participants having high intelligence and which among the pair having lower intelligence would be assigned to which condition of the independent variable could be determined on a random basis. Using random assignment as a second step of allocation in the manner just described has the potential of preventing confounding by other participant characteristics. Keep in mind though, and this is something that students overlook, that matching is one type of systematic assignment of participants. The control of any variables correlated with the dependent variable being measured through random assignment then, is contingent on the strength of participants available per level of matching. For example, in an experiment having four participants with high intelligent and four low on the same, random assignment is less likely to balance out the effects of other potentially condounding variables than an experiment having twenty participants of each type rather than four.

Another misconception that needs to be cleared here is that of students implicit assumption that matching means having an exactly identical participant composition in all conditions of the independent variable. As students should know, psychological variables are never found to manifest perfectly identically in any two individuals. This means that when participants are matched for an experiment, they can only be matched with respect to some range of their standing on a variable and not on exact standing. For example, if participants are to be matched with respect to attention span, some cutoff can be decided to demarcate high, moderate and low attention spans and then, participants can be matched with regard to the category they fall in. Of course, the narrower the ranges of the categories decided, the more accurate that the matching would be.

The last misconception of students regarding matching that I would like to clear here is that selection of participants on some predetermined criteria implies matching. When students read sentences in research papers such as, "All participants were matched with respect to age, sex and socioeconomic status," they assume that the researcher has employed a matched groups design in the study. In reality, sentences like these imply that during the selection of the sample of the study, care was taken that all participants fell in the same ranges of age, sex and socioeconomic status. As an example, assume that we have an experiment in which, the effects of a neurotransmitter on behaviour are being measured. Biological processes vary greatly with age and hence, the experimenter here would assure that all participants come from the same age range. This does not amount to creating a matched group design for it has to do with the selection of participants and not the creation of comparable groups through assignment.

To summarize the above discussion, I'll address the FAQ directly and succinctly now -

• Firstly, matching does not assure that any differences between participants are due to their differential exposure to the independent variable only. There are a host of variables innate to individuals that can influence the dependent variable besides the  independent variable. Matching accounts for only one or two of them, but not all;

• Secondly, matching participants does not mean that they are found to be equal with respect to their standing on some variable. It means that they have found to be roughly equal or belonging to the same range of standing on some given variable;

• In a line, why random assignment is superior to matching it balances out all participant conditions likely to affect the dependent variable; and not a selected few, like matching does.

You might have a few questions in your mind after reading this post such as, "How is random assignment different from random selection?" or, "What is the procedure of random assignment of variables?" or, "Is random assignment guaranteed to balance out all participant characteristics in an experiment?" or, "Why not simply use a repeated measures design instead of a matched groups one if the sample size is small?" You can drop such and similar questions in the comments below. If several of you ask the same question, I'll feature it on this FAQ series.

---

(Search Terms - random assignment, matched groups design, creating comparable groups, research in psychology, psychological research, research designs, IB Psychology research, A-levels psychology research, experiments in psychology, experimental psychology, masters in psychology, bachelors in psychology, independent groups design, student FAQs in psychology, excelling psychology, excellingpsychology.blogspot.com, Jyotika Varmani)