Single-Subject Design Methodology Research Paper

1. Single-Subject Design: Overview And Rationale

At the core, all scientists ultimately deal with particulars, whether cells, atoms, microbes, or the behavior of organisms. It is from such particulars that the facts of science and generality of findings are derived (Sidman 1960). This is true whether one aggregates a limited set of observations from a large sample of cases into groups that differ on one or more variables, or situations involving a few cases and a large number of observations of each over time (see also Hilliard 1993). The former is characteristic of hypothetico-deductive group design strategies that rely on inferential statistics to support scientific claims, whereas the latter idiographic and inductive approach is characteristic of single-subject methodology. As with other research designs, single-subject methodology is as much an approach to science as it is a set of methodological rules and procedures for answering scientific and practical questions. As an approach, single-subject designs have several distinguishing features.

1.1 Subject Matter Of Single-Subject Designs

Perhaps the most notable feature of single-subject methodology is its subject matter and sample characteristics; namely, the intensive study of the behavior of single (i.e., N = 1) organisms. This feature is also a source of great misunderstanding (Furedy 1999), particularly with regard to the generality and scientific utility of research based on an N of 1. It should be stressed that single-subject methodology gets its name not from sample size per se, but rather because the unit of analysis is behavior of individuals (Perone 1991), or what Gottman (1973) referred to as ‘N-of-one-at-a-time’ designs. Skinner (1966) described it this way: ‘… instead of studying 1000 rats for one hour each, or 100 rats for 10 hours each, the investigator is likely to study one rat for a 1000 hours’ (p. 21); an approach that served as the foundation for operant learning (Skinner 1953, 1957) and one that, with the increasing popularity of behavior therapy, set the stage for a new approach to treatment development, treatment evaluation, and increased accountability (Barlow and Hersen 1984). Though single-subject research often involves more than one subject, the data are evaluated primarily on a case-by-case basis. The idiographic nature of this methodology is well suited for applied work where practitioners often attempt to influence and change problematic behavior(s) of individual clients (Hayes et al. 1999). Though not a requirement, the inclusion of more than one subject helps establish the robustness of observed effects across differentially imposed conditions, individuals, behaviors, settings, and/or time. Hence, direct or systematic replication serves to increase confidence in the observed effects and helps establish the reliability and generality of observed relations (Sidman 1960).

1.2 Analytic Aims Of Single-Subject Designs: Prediction And Influence (Control)

As with other experimental design strategies, the fundamental premise driving the use of single-subject methodology is to demonstrate control (i.e., influence) via the manipulation of independent variables and analysis of the effects of such variables on behavior (dependent variables). This emphasis on systematic manipulation of independent variables distinguishes single-subject designs from other small N research such as case reports, case studies, and the like which are often lucidly descriptive, but less controlled and systematic. With few exceptions (see Perone 1991), the general convention of single-subject design is to change one variable at a time and to evaluate such effects on behavior. This ‘one-variable-at-a-time’ rule positions the basic or applied researcher to make clear causal statements about the effects of independent variables compared with a previous state (e.g., naturally occurring behavior during baseline), to compare different independent variables across or within conditions, and only then to evaluate interaction effects produced by two or more independent variables together.

Indeed, single-subject methodology is unique in that it typically involves an intensive and rigorous experimental analysis of the behavior (i.e., thinking, emotional responses, overt motor acts) across time, and an equally rigorous attempt to isolate and control extraneous sources of variability (e.g., maturation, history) that may confound and/or mask any effects caused by systematic changes in the independent variable(s) across conditions and time (Sidman 1960). This point regarding the handling of extraneous sources of variability cannot be overstated. Iversen (1991) and others (Barlow and Hersen 1984: Sidman 1960) have noted issues concerning how variability is handled in-group design research and single-subject methodology. With regard to such issues, Iverson put it this way:

[in-group design research] the fit between theory and observation often is evaluated by means of statistical tests, and variability in data is considered a nuisance rather than an inspiration …. By averaging over N subjects the variability among the subjects is ignored N times. In other words, a data analysis restricted to only the mean generates N degrees of ignorance … Single-subject designs are used because the data based on the individual subject are more accurate predictors of that individual’s behavior than are data based on an averaging for a group of subjects (p. 194).

Though the issue of when and how to aggregate data (if at all) is controversial in single-subject research (Hilliard 1993), there is a consensus that variability is to be pinpointed and controlled to the extent possible. This is consistent with the view that variability is imposed and determined from without, and not an intrinsic property of the subject matter under study (Sidman 1960). Efforts are deliberately made to control such noise so that prediction and control can be achieved with precision and scope. Consequently, those employing experimental single-subject design methods are more likely to commit Type II errors (i.e., to deny a difference and claim that a variable is not exerting control over behavior, when it is) compared with more classic Type I errors (i.e., to claim that a variable is exerting control over behavior, when it is not; (see Baer 1977).

1.3 Single-Subject Design: Measurement Issues

Single-subject design methodology also can be distinguished by the frequency with which behavior is sampled within conditions, and across differentially imposed conditions and time. It is common, for instance, for basic and applied researchers to include hundreds of samples of behavior across time (i.e., hours, days, weeks, and sessions) with individual subjects. Yet, a minimum of at least three data points is required to establish level, trend, and variability within a given phase or design element (Barlow and Hersen 1984, Hayes et al. 1999). Changes in level, trend, and at times variability from one condition to the next, and particularly changes that are robust, clear, immediate, and stable from behavior observed in a prior condition, help support causal inferences.

Here, stability is a relative term and should not be confused with something that is static or flat. Rather, stability provides a background by which to evaluate the reliability of changes produced by an independent variable across conditions and time. By convention, each condition is often continued until some semblance of stability is observed, at which point another condition is imposed, and behavior is evaluated relative to the previous condition or steady state (cf. Perone 1991). Such changes, in turn, are often evaluated visually and graphically relative to a prior less controlled state (i.e., baseline), or adjacent conditions involving another independent variable. In singlesubject methodology, such stability can be evaluated within a condition (i.e., from one sample or observation to the next) and across conditions (i.e., changes in level, trend, and variability). The pragmatic appeal of such an approach, particularly in applied contexts, rests in allowing the practitioner to evaluate the effectiveness of imposed interventions in real time, and consequently to modify their intervention tactics as the data dictate. Indeed, unlike more formal group designs that are followed strictly once set in place, single-subject methodology is more flexible. Indeed, it is common for design elements to be added, dropped, and/or modified as the data dictate so as to meet scientific and applied goals. This feature has obvious parallels with how practitioners work with their clients in designing treatment interventions (see also Hayes et al. 1999).

In sum, all single-subject designs focus on individual organisms and aim to predict and influence behavior via: (a) repeated sampling and observation; (b) manipulation of one or more independent variables while isolating and controlling sources of extraneous variability to the extent possible; and (c) demonstration of stability within and across levels of imposed independent variables (see Perone 1991). Discussion will now turn to an enumeration of such features in the context of more popular single-subject designs.

2. Varieties Of Single-Subject Methodology

Most single-subject designs can be generally classified as representing two main types: within-series and between-series designs. Though each will be discussed briefly in turn, it is important to recognize that neither class of design precludes combining elements from the other (e.g., combined-series elements). In other words, a within-series design may, for some purposes, lead the investigator to add between-series elements and viceversa, including other more complex elements (e.g., interaction effects, changing criterion elements).

2.1 Within-Series Designs

The basic logic and structure of within-series designs are simple; namely, to evaluate changes within a series of data points across time on a single measure or set of related measures (see Hayes et al. 1999). Stability is judged for each data point relative to other data points that immediately precede and follow it. By convention, such designs include comparisons between a naturally occurring state of behavior (denoted by the letter A) and the effects of an imposed manipulation of an independent variable or intervention (denoted by different letters such as B, C, and so on). A simple AB design involves sampling naturally occurring behavior (A phase), followed by repeated assessment of responding when the independent variable is introduced (B phase). For example, suppose a researcher wanted to determine the effects of rational-emotive therapy on initiating social interactions for a particular client. If an A-B design were chosen, the rate of initiations prior to treatment would be compared with that following treatment in a manner analogous to popular pre-to-post comparisons in-group outcome research. It should be noted, however, that A-B designs are inherently weak in controlling for threats to internal validity (e.g., testing, maturation, history, regression to the mean). Withdrawal reversal designs control for such threats, and hence provide a more convincing case of experimental control.

Withdrawal designs represent a replication of the basic A-B sequence a second time, with the term withdrawal representing the imposed removal of the active treatment or independent variable (i.e., a return to second baseline or A phase). For example, a simple A-B-A sequence allows one to evaluate treatment effects relative to baseline responding. If an effect due to treatment is present, then it should diminish once treatment is withdrawn and the subject or client is returned to an A phase. Other variations on withdrawal designs include B-A-B designs and A-B-A-B (reversal) designs (see Barlow and Hersen 1984, Hayes et al. 1999). A-B designs and withdrawal/reversal designs are typically used to compare the effects of a finite set of treatment variables with baseline response levels. Data regarding stable response trends are collected across several discrete periods (e.g., time, sessions), wherein the independent variable is either absent (baseline) or present (treatment phase). Phase shifts are data-driven, and response stability determines the next element added to the design, elements that may include other manipulations or treatments either alone (e.g., A-B-A-C-A-B-A) or in combination (e.g., A-B-A-B + C-A-B-A). As the manipulated behavior change is repeatedly demonstrated and replicated across increasing numbers of phase shifts and time, confidence in the role of the independent variable as a cause of such changes increases.

This basic logic of within-series single-subject methodology has been expanded in sophisticated and at times complex ways to meet basic and applied purposes. For instance, such designs can be used to test the differential effects of more than one treatment. Other reversal designs (i.e., B-C-B-C) involve the comparison of differential, yet consecutive, treatment interventions across multiple phase shifts. These designs are similar to the above reversals in how control is evaluated, but differ primarily in that baseline phases are not required. Designs are also available that combine features of withdrawal and reversal. For example, A-B-A-B-C-B designs allow one to compare A and B phases with each other (reversal; A-B-A-B), and a second treatment to B (reversal; e.g., B-C-B), or even to evaluate the extent to which behavior tracks a specified behavioral criterion (i.e., changing criterion designs; see Hayes et al. 1999). Changing criterion designs provide an alternative method for experimentally analyzing behavioral effects without subsequent treatment withdrawal. Criterion is set such that optimal levels given exposure can be met and then systematically increased (or decreased) to instill greater demand on acquiring new repertoires. For example, a child learning to add may be required to calculate 4 of 10 problems correctly to earn a prize. Once the child successfully and consistently demonstrates this level of responding, the demand increases to 6 of 10 correct problems. Changing criterion designs serve as a medium to demonstrate learning through achieving successive approximations of the end-state.

2.2 Between-Series Designs

Between-series designs differ primarily from within-series designs in that data are first grouped and compared by condition, and then by time, whereas the reverse is true for within-series designs. Between-series designs need not contain phases, as evaluation of level, trend, and stability are organized by condition first, but not by time alone (Hayes et al. 1999, p.176). Designs within this category include ‘alternatingtreatment design’ and ‘simultaneous-treatment design.’ The basic logic of both is the same, in that a minimum of two treatments are evaluated concurrently in an individual case. With alternating-treatments design, the concurrent evaluation is made between rapid and largely random alternations of two or more conditions (Barlow and Hayes 1979). Unlike within-series designs, alternating-treatment designs contain no phases (Hayes et al. 1999). The same is true for simultaneous-treatment designs; a design that is appropriate for situations where one wishes to evaluate the concurrent or simultaneous application of two or more treatments in a single case. Rapid or random alteration of treatment is not required with simultaneous-treatment design. What is necessary is that two or more conditions are simultaneously available, with the subject choosing between them. This particular design is not conducive for evaluating treatment outcome, but is appropriate for evaluating preference or choice (see Hayes et al. 1999). Alternating-treatment designs, by contrast, fit well with applied work, as therapists must often routinely target multiple problems concurrently, and thus need to switch rapidly between intervention tactics in the context of therapy.

2.3 Combining Within- And Between-Series Elements: Multiple-Baseline Design

Multiple-baseline designs build upon and integrate the basic logic and structure of within and between-series elements. The fundamental premise of multiple-baseline designs is to replicate phase change effects systematically in more than one series, with each subsequent uninterrupted series serving as a control condition for the preceding interrupted series. Series can be compared and arranged across behaviors, across settings, across individuals, or some combination of these (see Hayes et al. 1999). Such designs require that the series under consideration be independent (e.g., two functionally distinct behaviors), and that intervention be administered sequentially beginning with the first series, while others are left uninterrupted as controls. For example, a client might present with three distinct behavior problems all requiring exposure therapy. All behaviors would be monitored during baseline (A phase), and after some semblance of stability is reached the treatment (B phase) would be applied to the first behavior series, while the remaining two behaviors are continuously monitored in an extended baseline. Once changes in the first behavior resulting from treatment reach stability, the second series would be interrupted and treatment applied, while the first behavior continues in the B phase and the third behavior is monitored in an extended baseline. The procedure followed for the first two series elements is then repeated for the third behavior. This logic can be similarly applied to multiple-baseline designs across individuals or settings (see Barlow and Hersen 1984, Hall et al.). More complex within-series elements (e.g., A-B-A-B, or B-C-B, or counterbalanced phases such as B-A-B and A-B-A) can be evaluated across behaviors, settings, and individuals.

Note that with multiple-baseline designs the treatment is never withdrawn, rather, it is introduced systematically across a set of dependent variables. Independent variable effects are denoted from how reliably behavior change correlates with the onset of treatment for a particular dependent variable. For such reasons, multiple-baseline designs are quite popular, owing much to their ease of use, strength in ruling out threats to internal validity, built-in replication, and fit with the demands of applied practitioners working in therapy, schools, and institutions. Indeed, such designs can be useful when therapists are working with several clients presenting with similar problems, when clients begin therapy at different times, or in cases where targets for change in therapy occur sequentially.

3. Recent Trends In The Application Of Singlesubject Methods

Single-subject methodology has a long historic affiliation with basic and applied behavioral research, and the behavior therapy movement more generally (Barlow and Hersen 1984). Though the popularity of single-subject methods in published research appearing in mainstream behavior therapy journals appears to be on the decline relative to use of group design methodology (Forsyth et al. 1999), there does appear to be a resurgence of interest in single-subject methodology in applied work. There are several possible reasons for this renewed interest, but only two are mentioned here. First, trends in treatment-outcome research have increasingly relied on the now popular ‘randomized clinical trial’ group design methodology to establish empirical support for psychosocial therapies for specific behavioral disorders. Practitioners, who work predominantly with individual clients, are quick to point out that group-outcome data favoring a given intervention, however convincing statistically speaking, includes individuals in the group that did not respond to therapy. Moreover, the ‘average’ group response to treatment may not generalize to the specific response of an individual client to that treatment. Thus, practitioners have been skeptical of how such research informs their work with individual clients.

The second, and perhaps more important, reason for the renewed interest in single-subject methodology is driven by pragmatic concerns and the changing nature of the behavioral health care marketplace. Increasingly, third-party payers are requiring practitioners to demonstrate accountability. That is, to show that what they are doing with their clients is achieving the desired effect (i.e., good outcomes) in a cost-effective and lasting manner. Use of single-subject methodology has a place in assisting practitioners in making clinical decisions based on empirical data, in demonstrating accountability to third party payers and consumers. Further, such methodology, though requiring time and sufficient training to implement properly, is atheoretical and fits nicely with how practitioners from a variety of conceptual schools work routinely with their individual clients (Hayes et al. 1999). Thus, there is great potential in single-subject methodology for bridging the strained scientist–practitioner gap, and ultimately advancing the science of behavior change and an empiricallydriven approach to practice and treatment innovation.

4. Summary And Conclusions

Methodology is a way of knowing. Single-subject methodology represents a unique way of knowing that has as its subject matter an experimental analysis of the behavior of individual organisms. Described here were the assumptions driving this approach and the main varieties of design options available to address basic experimental and applied questions. Perhaps the greatest asset of single-subject methodology rests with the flexibility with which such designs can be constructed to meet the joint analytic goals of prediction and control over the behavior of individual organisms. This asset also represents one of the greatest liabilities of such designs in that flexibility requires knowledge of when and how to modify and/or add design elements to achieve analytic goals, and particularly skill in recognizing significant effects, sources or variability, and when one has demonstrated sufficient levels of prediction and control over the behavior of interest.

Bibliography:

1. Baer D M 1977 Perhaps it would be better not to know everything. Journal of Applied Behavior Analysis 10: 167–72
2. Barlow D H, Hayes S C 1979 Alternating treatments design: One strategy for comparing the effects of two treatments in a single subject. Journal of Applied Behavior Analysis 12: 199–210
3. Barlow D H, Hersen M 1984 Single-Case Experimental Designs: Strategies for Studying Behavior Change, 2nd edn. Allyn & Bacon, Boston, MA
4. Busk P L, Marascuilo L A 1992 Statistical analysis in single-case research: Issues, procedures, and recommendations, with applications to multiple behaviors. In: Kratochwill T R, Levin J R (eds.) Single-Case Research Design and Analysis. Plenum, New York, pp. 159–85
5. Forsyth J P, Kollins S, Palav A, Duff K, Maher S 1999 Has behavior therapy drifted from its experimental roots? A survey of publication trends in mainstream behavioral journals. Journal of Behavior Therapy and Experimental Psychiatry 30: 205–20
6. Furedy J J 1999 Commentary: On the limited role of the singlesubject design in psychology: Hypothesis generating but not testing. Journal of Behavior Therapy and Experimental Psychiatry 30: 21–2
7. Gottman J M 1973 N-of-one and N-of-two research in psychotherapy. Psychological Bulletin 80: 93–105
8. Hall R V, Cristler C, Cranston S S, Tucker B 1970 Teachers and parents as researchers using multiple baseline designs. Journal of Applied Behavior Analysis 3: 247–55
9. Hayes S C, Barlow D H, Nelson-Gray R O 1999 The Scientist- Practitioner: Research and Accountability in the Age of Managed Care. Allyn & Bacon, Boston, MA
10. Hilliard R B 1993 Single-case methodology in psychotherapy process and outcome research. Journal of Consulting and Clinical Psychology 61: 373–80
11. Huitema B E 1986 Statistical analysis and single-subject designs: Some misunderstandings. In: Poling A, Fuqua R W (eds.) Research Methods in Applied Behavior Analysis. Plenum, New York, pp. 209–32
12. Iversen I H 1991 Methods of analyzing behavior patterns. In: Iversen I H, Lattal K A (eds.) Experimental Analysis of Behavior, pt 2. Elsevier, New York, pp. 193–241
13. Kazdin A E 1982 Single-Case Research Designs: Methods for Clinical and Applied Settings. Oxford University Press, New York
14. Parsonson B S, Baer D M 1992 The visual analysis of data, and current research into the stimuli controlling it. In: Kratochwill T R, Levin J R (eds.) Single-Case Research Design and Analysis. Plenum, New York, pp. 15–41
15. Perone M 1991 Experimental design in the analysis of free-operant behavior. In: Iversen I H, Lattal K A (eds.) Experimental Analysis of Behavior, pt 1. Elsevier, New York, pp. 135–71
16. Sidman M 1960 Tactics of Scientific Research: Evaluating Experimental Data in Psychology. Basic Books, New York
17. Skinner B F 1953 Science and Human Behavior. Macmillan, New York
18. Skinner B F 1957 The experimental analysis of behavior. American Scientist 45: 343–71
19. Skinner B F 1966 Operant behavior. In: Honig W K (ed.) Operant Behavior: Areas of Research and Application. Appleton-Century-Crofts, New York, pp. 12–32