Transcription
Another another point to consider is that an experiment on a group of subjects must be designed so that all subjects receive the same experience and can be compared. So, this necessity can result in a design that's not the best one for all your subjects.
So, during an experiment on behavior modification, an experimenter may discover that a subject does not respond to a reinforcer that has worked on previous subjects. So, if the design is a single-subject one, the experiment can be modified on the spot by switching reinforcers or by alternating the instructions.
And another problem that can be solved by a design modification is when a large change occurs in the participant's behavior that the experimenter suspects is caused by an outside event rather than the experimental manipulation. I mean, think about it. If you were conducting a study and then coronavirus took place, that might affect the stress levels of participants, the behaviors that they take, the accessibility to treatment conditions, etc. So, the experimenter can immediately switch the conditions to see if the behavior changes correspondingly. A group experiment, however, would call for continuing all participants in the same procedure and hoping that the outside events would essentially cancel each other out. So, this would be a design limitation of a group study.
So, we should also mention some disadvantages of the single-subject approach. If there were none, after all, no multiple-subject experiments would ever be done because it's a lot easier to do single-subject research than it is to do groups. Group research. So, first, some effects are small relative to the amount of variability in the situation. It may be impossible to control the other sources of variability sufficiently to observe the experimental effect in one subject. This is why modern statistical methods were developed in the first place. The only way to decide if having been abused as a child causes adults to be child abusers themselves is to compare large groups of individuals. So, it's no surprise that group experiments and statistical analysis are discussed together so often. Statistical procedures for analyzing single-subject data are not as well developed.
Second, some experimental effects are, by definition, between-subjects effects. So, you're looking at different subjects. It's impossible to have a participant who simultaneously receives two opposite sets of instructions in a social psychology experiment or who is taught the same material by two different methods without them retaining the first method. So, you need to have different participants to compare. So, this is where it becomes more advantageous to have different groups versus different individuals. So, when you're using a group design, you compare one group of subjects against another or a group of subjects in one condition against the same subjects at another condition. So, the assumption that the groups were equal before the treatment is the basis of your attributing the effect to the manipulation rather than to something else. So, this assumption can be tested by statistically analyzing the differences between groups.
So, when you have only one subject, you have to use a different strategy for the outcomes between conditions. And that strategy is to compare the behavior that occurs before and after the introduction of the experimental manipulation. That the behavior before the manipulation must be measured over a long enough time span to obtain a stable baseline against which the later behavior will be measured. So, the baseline is, it really serves two purposes. It's the, it measures the current level of behavior and it predicts what the behavior would be in the future if no treatment were administered. The evaluation of that fact can only occur if baseline measurements show the behavior to either be remaining at the same level or changing in the direction that is opposite to the predicted treatment effect.
So, suppose you want to measure the effectiveness of a treatment for anorexia nervosa, a disorder characterized by voluntary starvation. You would need to measure the patient's weight and food intake for a period of time before initiating treatment so that you could show that the weight was stable and that the patient had not begun gaining weight spontaneously. How long this baseline measure should be continued is to say, this is one of the areas where you'd run into potential problems. The judgment of stability is subjective. However, the experiment would be useless unless it was evident that the patient had not begun gaining weight before treatment. So, a declining baseline may be acceptable because the treatment in this example is expected to, is expected to cause an increase in behavior. So, a decrease in weight in the absence of treatment is not unusual for patients with anorexia nervosa because they're not getting their required caloric intake. So, a weight loss would constitute an acceptable baseline for comparison with treatment.
So, this example illustrates another consideration for obtaining baseline measures. Sometimes the existing condition is harmful to the subject or even life-threatening, as in this example. So, the goal of a stable baseline may be overridden by other factors. So, that's important to consider as well.
So, if you simply measure the baseline behavior, usually referred to as or denoted by A in single-subject designs, and then a treatment, usually called B, it's called a comparison design or an A-B design. So, in 1996, there was a paper that reported cases of phantom pain that were treated effectively with the A-B design. So, in this paper, the 63-year-old woman in this study had lost her leg below the knee, yet was still experiencing pain sensations from the part of the leg that had been amputated. So, that was the baseline condition that they were looking at. During a single session of the proposed treatment, the woman was taught a strategy for deferring her attention from the pain. So, that was the treatment condition. After the treatment, the woman reported that the pain had substantially decreased. But can we be sure that the treatment really led to pain relief? Well, no. But the difficulty with an A-B design is that you will not know whether other variables that may have coincidentally changed at the same time that the treatment was administered actually produced the change in behavior. So, here's the potential confounding variables have a big deal. In fact, you would have the single participant equivalent of the quasi-experimental design called the one-group pretest-post-test design. So, you know that that could be tricky. But if you're interested in this paper, it's a 1996 paper by Marchian, Sten R. John Johnson, and Karen Fisher, and it is an interesting read, but it also introduces a lot of questions as to what, you know, whether the actual treatment effect was what they were seeing, which is one of the issues with this.
So, although an A-B design may be appropriate for some clinical situations, the argument that the treatment is the cause of the change is considerably strengthened if the treatment is withdrawn after a period of time and the behavior shows a return towards the baseline. So, this use of treatment withdrawal as part of an experimental condition is often referred to as an A-B-A design.
So, two principal problems are associated with the A-B-A design. First, the effect of the manipulation may not be fully reversible. Right? Just because you remove a treatment, it may not show the effect that you're looking for. So, if the treatment were a lesion of the brain that causes obesity, clearly it would be impossible to reverse that. And if a learning procedure causes a more or less permanent change in a participant's behavior, that too really wouldn't be reversible.
The second problem with A-B-A designs is that you may want to leave the participants in the new condition rather than returning them to their original state. So, treatments involving weight control, phobias, also behaviors are typical examples in which it would really be unethical and definitely undesirable to withhold treatments until the patient returned to their original state. In such cases, the experimenter may withdraw treatment temporarily before the behavior change has reached the desired level, but after the behavior shows some reversal of the trend towards improvement, he or she would reinstate the treatment. In any case, experimenters in behavior modification seldom end an experiment with the baseline or withdrawal condition. Rather, they reintroduce the treatment to produce the maximum benefit for the client. So, you really, you know, have to think about the effects of these single-subject designs, especially if you're, if you're talking about introducing a new form of a treatment to treat an issue that a person's really suffering from.
So, then we also have an A-B-A-B design. In the examples I just discussed, the treatment was repeated after the withdrawal phase to leave the participant with the full benefit of the training. This repetition of treatment also has the advantage of providing another opportunity to evaluate the effect of the treatment and its reliability. So, this kind of A-B-A design experiment in which treatment is repeated is actually called an A-B-A-B or replication design. Repeated presentation with all of a variable, confused, strong evidence for the validity of the independent variable's effect. So, anybody who watches a pigeon who's been trained in a Skinner box respond to the presence of a light and not respond to its absence would be impressed by the control the light exerts over the animal's behavior. The light appears to turn the behavior on and off by a switch. So, this would be an example of this design.
Another important rule of a single participant or single participant research in general is to vary only one thing at a time. If two variables are changed simultaneously, it's impossible to decide whether the change in behavior was caused by one or the other or by two together.
On occasion, it may be important to evaluate the effect of two or more different treatments to assess which one would be the most effective. So, for a study done in 2000 by Michael Mazzoni and Stephanie R. Tennety, they were interested in ways that therapeutic rehabilitation sessions could be more effective for people with brain injuries. Many of the therapeutic activities were frustrating or difficult for participants, and brain injury patients often have difficulty in remaining on task. And so, the researchers wondered whether, or wondered which of the three different treatments would be most effective in helping patients stay on task: allowing the patient to earn short breaks, giving them praise for cooperation, or displaying the progress of rehabilitation.
So, to meet the requirement that only one variable at a time be changed in a single-participant design, they alternated the treatments across sessions with a 15-year-old male patient in a speech therapy session. This method is called an alternating treatments design. Behavior was observed during the first baseline phase, then three treatments were administered. The patient had randomly chosen one of the three alternating treatments in each session. And if a strict alternation had been followed, then the type of treatment would have been confounded with the time of day or day of week. So, it was important that the treatments were randomly allocated. It's possible that attention might be lower in the afternoon than in the morning, for example. So, the patient was told which of the three treatments or consequences would be in effect for each session before it began. The results showed that allowing the patient to earn short breaks was the most effective of the three treatments. So, you can see how this would be a useful design for certain types of single-subject studies.
The first part of the alternating treatments design answers the question of which is the more effective treatment. It doesn't, however, necessarily show that the behavior change was the result of that treatment, as the baseline may have been changing anyway. So, evaluating whether the treatment caused the behavior change would have required some sort of repetition of the design. One way would have been to remove all three treatments, then measuring baseline performance for a period of time. When the baseline was again stable, the more effective treatment could be reintroduced. So, essentially, this design is similar to the A-B-A-B design with the one added piece that the more, the more than one type of treatment is administered during the first non-baseline condition. However, the researchers didn't want to withhold treatment from the patient because they were afraid therapeutic gains may be reversed. Instead, they continued to implement the most effective treatment, earning short breaks from speech therapy, somewhat like changing criterion designs. They gradually increased the amount of time the patient wouldn't have to concentrate in order to earn a break, from two minutes all the way up to thirty minutes.
So, another effective way to demonstrate that the manipulation caused the behavior change is to reintroduce the manipulation at different times for each of several different behaviors to see if the onset of the behavior change coincides with the manipulation. For example, suppose a researcher is trying to determine if rewarding a child who has an intellectual disability for doing certain personal tasks is effective. So, if the researcher begins to reward tooth brushing, facial washing, and washing and air calming all at the same time, it's possible that the presence of the experimenter, the attention received, or a spontaneous decision to turn over a new leaf was responsible for the change. The researcher, however, could begin rewarding only tooth brushing the first week, tooth brushing and face washing the second week, and so forth, until after four weeks, all behaviors were being rewarded. This sequence would make it implement, would make it possible to see whether the increase in behavior coincides with reward. So, this design is known as a multiple baseline design. The separate experimental baselines may be different behaviors in the same individual, as the example I just gave, or the same behaviors in different individuals. A third possibility is to test the same behavior in the same individual but in different behavior settings. So, an example of this last approach will give it a second, but multiple baseline designs are especially useful if the expected behavior change is irreversible.
So, in a study done by Singh, Dawson, and Gregory in 1980, they used a multiple baseline design with withdrawal of treatment to stop an 18-year-old female with profound intellectual disability from hyperventilating. This problem, in which a person breathes too deeply, can have serious medical consequences. So, several treatments, including reprimand and medication, had been attempted on the participant, really without any success. So, the researcher has decided to punish episodes of hyperventilation by briefly presenting ammonia, so smelling salts, if you've ever experienced that, if you've ever fainted or or you know, been feeling woozy, it is a very powerful, noxious experience. Yes, I wouldn't call it torturous, but it definitely will distract and deflect from one's reaction if they're having one. So, smelling salts and ammonia were presented every time the participant hyperventilated. So, the design was a multiple baseline design with the withdrawal of treatment as a test probe. Baseline recordings were made for five days in each of four different settings: classroom, the dining room, bathroom, or in experimental sessions. Lasted two hours per day, 30 minutes each in each of the four settings. During this time, all instances of hyperventilation were recorded but not punished. Then, the baseline recordings were continued in three of the settings, and punishment was administered to any episodes of hyperventilation that occurred in the classroom. After five more days, punishment was instituted for episodes that occurred in the dining room as well as in the classroom, and baseline reporting continued in the bathroom and living room. After five more days, punishment was extended to the bathroom, and after another five days, to the living room.
The data from this experiment, it really indicated that the frequency of hyperventilation did not decrease in any of the baseline conditions. In each situation, when punishment was initiated, however, hyperventilation decreased dramatically. Again, and this kind of punishment, I mean, it's extremely powerful, extremely distracting. So, it kind of took her out of the state that they were trying to take her out of. So, after punishment had been continued for five days in the last setting, the punishment was withdrawn, but the episodes of hyperventilation were still recorded up until the participant began hyperventilating again in all situations, but ceased abruptly when punishment was reinstated. After 15 more days of punishment for any instance of hyperventilation in the four settings, the procedure was generalized. Instead of having one experimenter administer punishment in the four settings during a two-hour session, all nurses in the ward were to administer punishment whenever they observed hyperventilation in any setting during an eight-hour day. So, this practice was instituted to consolidate and maintain the previous gains.
So, this study illustrates how to test the effectiveness of an experimental manipulation on a single participant in an unambiguous manner. Because the manipulation was instituted at four different times in four different settings, the researchers would otherwise have had to use four different alternative hypotheses to account for the decrease in hyperventilation. After the increase in hyperventilation that occurred followed the removal of punishment and the abrupt decrease when punishment was reinstated, to give further evidence of the effectiveness of treatment.
So, another way of showing the manipulation caused the behavior change is to change the criteria for reward over time. After a baseline measurement, a reward can be given for meeting a lax criterion of the behavior. So, after the behavior stabilizes at that level, the criterion can be raised until the behavior stabilizes again, and so forth. If the behavior begins to change after each change in the criterion, then the conclusion that the reward is the cause of the improvement is rather convincing.
So, suppose that a child is unable to sit still in class. A teacher may reward the child for sitting still for five minutes, or for five minutes at a time, I should say, until the performance becomes stable. Then, the criterion may be set at ten minutes, later 15 minutes, and so forth. The behavior at each criterion becomes the baseline against which to evaluate the effect of the manipulation of the next criterion. So, like the multiple baseline design, a what's called changing criterion design is useful when behavior change is irreversible or when the return to the initial baseline is not desirable, as in a case involving the reading skills of a man with schizophrenia named Craig. And this is in a study done by Skinner, Skinner and Armstrong in 2000. And note, it's not the same Skinner. So, Craig was able to read at the beginning of the experiment, but only read on average one page per day. So, that's not that great. They wanted to increase the amount of pages read per day. And so, the increase of the amount of pages read per day, what, or the number of pages read per day, was the dependent variable, and the opportunity to earn a soft drink was the treatment. So, he was being rewarded for reading more pages per day. Craig had demonstrated during a baseline that he could read one page per day. So, this was set at the initial criterion level, and after Craig achieved the criterion level three days in a row, the criterion would be raised by one page per day. So, at the end of the program, Craig was reading on average eight pages per day, and the behavior change persisted for about seven weeks after the program ended. So, it was effective, work, but it was effective for at least a period of time.
So, that's all we have for today's lecture on single-subject research design. As always, make sure to answer the discussion questions posted for this week. Please be mindful of the part two prompt for the second part of the research paper. A few notes on that. When you are submitting your research paper, the second part should be submitted as one continuous document with the part one research paper. That is, you're not just submitting the second half, just like you submitted just the first half for the part one assignment. It's going to be one big paper. So, the idea for this assignment was that you would have a large research paper where the first component would be the literature review, the research questions, and the hypotheses, then the references you have so far. The second part is going to be your proposed methods, your anticipated results, and then a discussion section that ties everything together. And so, in that discussion section, you're going to be following the same order of the introduction of information you had in your literature review. In that introduction section, with all of the information that you know, when you talked about research that still needs to be done, research that kind of led you to where you are, seeing those data points as things to address. Your discussion, how your proposed study would address the information you've organized in the literature review. So, if you have any questions about how to do this, please feel free to reach out, email me. I'd be happy to guide you through the process. Discussion sections are often a lot more fun to do than the introduction because you've already done the background research, and essentially, you're addressing point by point what had to be done and how your study did it. And if your study didn't do it, how your study contributed to answering that question. And then at the end of the discussion, and this is important, you should have suggestions for future research based on the research that you're proposing doing. Because nobody is going to be proposing a topic that addresses everything. Usually, when we answer some questions, it has the potential to bring up more questions in the future. And so, you're keeping the sign to the cross by suggesting future research ideas. So, be sure to organize it that way. Also, do the best you can to resolve my comments that I have made on your papers. So, just be sure to integrate that in whatever way possible. If you have any questions about the comments that I've made, feel free to reach out to me and and to email me, and we can discuss that further. Alright guys, take care. Have a great week, and I'll look forward to reading your discussion posts.