Classic Reversal

Simplest Single-Case Design

Alternating-Treatments Design

Multiple-Baseline Design

The essence of a single-case design is the appraisal of a condition with continued assessment of the condition through the treatment period. In school psychology and counseling applications, the "condition" is most often (though not always) a problem, either academic or behavioral.

The classic single-case research design is typically identified as an A-B-A or a reversal design. First, a condition to investigate is identified. For example, a student is experiencing symptoms of anxiety. Then:

A- The severity of the problem is quantified with measurement of the anxiety in a baseline period before treatment is introduced.

B- Treatment is then introduced with continuing measurement of the anxiety.

A- Treatment stops but the anxiety measurement continues for a period of time with expectation that the treatment effect will go away.

The logic of the A-B-A design is straightforward. If the treatment (independent variable) is effective, there will be a positive change in the condition being measured (dependent variable) after treatment is introduced, and there will be a return to the baseline level when the treatment is withdrawn.

Elaborations of the design are easy to create. For example, the researcher could determine the impact if treatment is reintroduced (A-B-A-B); the researcher could investigate the impact of another treatment (A-B-A-C), and so forth.

There is an elegant simplicity in the A-B-A design. But, there is also an inherent problem which may explain why single-case research has not already been embraced by school psychologists and counselors. The classic A-B-A or reversal design assumes that there will in fact be a reversal. When treatment is withdrawn, the condition should return to at least nearly what it was before the treatment began.

In our practices, we believe (or would certainly like to believe) that the interventions we provide usually have a more lasting effect. When a school counselor terminates a counseling relationship with a student, the idea is not that the problems which led to the counseling will immediately recur. The goal, of course, is just the opposite.

There are some instances, particularly involving consulting, when the classic A-B-A design may be an appropriate choice for the school psychologist or counselor. The focus, however, in this program will be on the single-case designs in which the treatment is expected to have some lasting effect. When a reversal effect is not reasonable, design alternatives include a simplified A-B time series, an alternating-treatments design, and a multiple-baseline design.

An A-B design is the least complicated of the single-case design alternatives. It is a simplified time-series which could be designated as a B(aseline)-T(reatment) model.

To use this design, you need only to have repeated measurements of the problem or condition before your treatment (A) and then repeated measurements during the treatment period (B). The analysis is simply a comparison of the two sets of measurements.

The A-B model has limitations as a "research" design, but it certainly can be your starting place for work with the single-case model. It fits easily into the workplace, and it could be argued that the data required should always be gathered as a part of best professional practice.

As a research design, the problem with the A-B model is the lack of control of extraneous variables. This is an internal validity problem of the design. A student has elevated anxiety. You recommend and/or provide treatment. The level of anxiety is significantly reduced. What you don't know is whether it was your treatment that led to the reduction in anxiety or instead it was one or some combination of a variety of other possibilities.

It would, however, be a mistake to dismiss this design too quickly. First, for whatever reason, in the example above there was improvement after the treatment began. That is important information, and, too often, we have not taken the time to gather such data. Knowing the "cause" would be helpful but knowing that there was improvement may be the more crucial piece of information.

The statistical tool kit includes three procedures, described in the statistical analysis tools section of this program, which can be used with the A-B design: the binomial expansion, a chi square analysis, and a simplified time series using the C-statistic. With either of these analysis procedures, your interpretations from the A-B design are enhanced by considering whether any change is likely to have been the result of chance alone.

Obviously crucial to the single-case design methodology is the quality of the measurement. How to choose and how often to use the measurement tools is addressed in the measurement section of this overview.

The possible influence of extraneous variables (maturation, changes in environment, etc.) obviously limits the A-B design as a source for making inferences about "cause". However, there is an exception. When there is no evident change between baseline and treatment, it is usually reasonable to infer that your treatment is not having causal influence through the period being considered. Even that inference cannot be absolute (inferences seldom are) because it is theoretically possible that the treatment effect was masked by a negative effect of other variables.

Alternating-treatment and multiple-baseline designs have more internal validity controls and will be considered next. But, before moving to those designs, there is an enhancement of the A-B design which warrants some attention in the practice of school psychology and counseling.

Baseline data have been gathered. Treatment has been provided, and the analysis suggests a significant positive change. Treatment (for example, counseling) has been terminated. When circumstances allow it, it will be very helpful, after a month or so, to again gather data in a follow-up phase using the same measurement tool. This is, in effect, an A-B-A design with a different perspective. In this case, you are hoping not for a reversal but for a continuation of the positive outcome.

When control of extraneous variables is an especially important concern, the best alternative in most school psychology and counseling applications will be an alternating-treatments design. Baseline data is optional (but recommended if circumstances allow). The essence of this design is a random presentation of at least two forms of treatment and continuous monitoring of the effects.

Usually the different treatments will be provided from session to session. The random order for the different treatments is important to increase the likelihood that the observed effects are the result of the treatment rather than some extraneous influence.

To illustrate, let's return to the student with symptoms of elevated anxiety in the school setting. Two forms of treatment will be compared: 1) weekly individual counseling sessions focused on building self-concept and 2) participation in an after school group guidance activity with open membership.

It will be helpful to have some baseline data on the frequency and/or severity of the anxiety symptoms before beginning either of the treatments. This can be as simple as a self-report, perhaps on a scale of ten, of the level of anxiety being experienced by the student. You may also find it helpful to obtain a comparable report from the teacher who referred the student.

Before treatment begins, a counterbalanced random schedule for treatment participation should be determined. For illustration, let's assume a ten-week treatment period during which, in random predetermined order, the student will participate in five individual weekly counseling sessions and five weekly group guidance activities.

You will notice some assumptions in this example. It is assumed that the either of the treatment alternatives could have an impact on the level of anxiety and that the impact would not be long delayed. That assumption seems reasonable in this example.

It is also assumed that the two forms of treatment are, in fact, different treatment modalities and that the carryover effect will not be significant. That assumption may be difficult to support in this (and many other) illustrations, but the concern is not fatal.

When and how to gather the repeated measures assessment data is considered in some detail in this overview section on measurement. For this illustration, let's assume that the data are gathered once per week, on the day immediately following the treatment.

It is highly recommended that you begin graphing the measures at the onset of the treatment. One of the marked advantages for the practitioner of the single-case method as opposed to other research designs is that you do not have to wait until the end of a study to have information which could indicate a need to make immediate changes in your original plans.

Assuming that the original plan continued through the ten weeks, you are then ready to complete the statistical analysis comparing the effect of the two treatments. Details are in the statistical analysis section of this program. In essence, you will first check for autocorrelation. Then, if supported by the outcome of the autocorrelation analysis, data for the two conditions can be contrasted with a traditional t-test or a nonparametric Mann-Whitney U.

The alternating-treatments design controls for extraneous variables (internal validity) through assumption that such variables would be expected to influence both of the treatments. External validity, the generalizability question, requires replication.

This design is very flexible and often can be implemented within the framework of existing school resources. For example, this design could be used to contrast the impact of counseling sessions which do and do not include parent participation. When a student is experiencing problems in a specific academic content area, this design can be used to compare the impact of treatment focused specifically on this content and treatment with a broader focus. Of particular value is the capability to concurrently implement both areas of focus.

Another design choice when you want to establish (or refute) a causal or functional relationship between the independent (treatment) variable and dependent variable is the multiple-baseline. Multiple-baseline studies can investigate the effect of the treatment with:

1. additional student(s)
2. additional problems for the same student, or
3. the same problem and same student in different settings.

The logic of the design is essentially the same whether the study is designed across students, across problems, or across settings.

To illustrate, assume that you have two referred students, each of whom has (perhaps among other concerns) a problem with elevated anxiety symptoms. You begin gathering baseline data at the same time for each student. The treatment will be participation in a group counseling activity.

To use the multiple-baseline design in this situation, you would have only one of the students begin the group counseling, continuing to gather the anxiety data for both students. The second student would not begin the group counseling treatment until there was indication that anxiety symptoms were being reduced in the student receiving the treatment. When improvement was evident in the first student, you would then begin the treatment with the second student (remember that you are continuing to use the repeated measures with both students throughout the process). In effect, the second student has a longer "baseline" period.

If the treatment is in fact directly related to the anxiety reduction, the effect is expected to be evident in the data from the two students. Symptoms in the second student would continue at the baseline level until the onset of treatment.

In school psychology and counseling applications, there is, unfortunately, the potential for a serious ethical problem in this design. If you actually feel that the treatment will be effective in reducing the symptoms, withholding treatment from the second student in order to comply with a research design is a questionable practice at best. However, in actual practice, there may be times when the second student, for other reasons, cannot begin treatment at the same time as the first one. If the baseline data gathering can begin at the same time for both students, circumstances would allow you to use this design.

You would not necessarily want to discard the treatment even if success is not evident with the second student. The group counseling could in fact have been the causal feature for the success with the first student while other factors could have prevented the effect in the second student. It is also possible that factors other than the treatment could have resulted in apparent success with both students. For this reason it is usually recommended that you have at least three baseline procedures (three problems, three persons, or three settings).

When a student is referred to you, it is probably unusual for there to be only one concern. If you plan to use the same form of treatment for both problems and if the problems are not directly correlated and if you plan to focus on one problem at a time, a multiple-baseline design across problems may work especially well. (The grammar in the preceding sentence may be shaky, but you get the point.)

For example, a student is referred to you with concern about maintaining on-task behavior in the classroom and concern about anxiety apparently associated with being recently moved to a new school. Records indicate that the on-task problem was evident in the prior school as well. You believe that a treatment based on brief counseling to build self-confidence will be effective with both problems and decide to focus first on the anxiety symptoms.

To implement the multiple-baseline approach, you gather baseline data on both on-task behavior and anxiety symptoms, beginning the data gathering for both problems at the same time. After a stable baseline has been obtained (more on this in statistical analysis section), you begin treatment of the anxiety symptoms, continuing to gather baseline data on the on-task behavior problem. When there is some evidence of treatment success with the anxiety symptoms, you change your focus in treatment to the on-task behavior.

If success with the anxiety symptoms is the result of something other than your treatment, that "other" factor may well have comparable effect on the on-task behavior. For example, perhaps it is just the increased attention being given to the student. If so, the baseline data for the on-task behavior could show marked improvement at about the same time as the improvement is evident in the anxiety.

The third option within the multiple-baseline approach is the contrast of the same problem with the same student but across different settings. For example, the presenting problem might be an oppositional behavior (not following directions) which was evident both in the classroom and at home.

The treatment you propose is based on manipulating contingencies of reinforcement. First, you gather baseline data on the problem in both settings. Treatment begins with a focus on rewards for the desired behavior but only in the classroom setting. Baseline data gathering continues in the home setting. When there is some evident success in the classroom setting, the treatment focus changes to providing rewards for following directions at home. As in the other forms of multiple-baseline, the impact is assumed to be evident in inspection of the data.

In actual applications in school psychology and counseling, the simplified A-B time series will probably be most useful, and the alternating-treatments design will typically be the best choice for increased control of extraneous variables. Remember, however, that this material is intended only as a brief overview of the research designs. The references included at the end of this overview provide more detailed information about these and other alternatives.

_________________________________________________________________________

continue to measurement                 research design review questions

access statistical analysis tools         back to program guide