Chapter 5 The Case Study and Single Case Design

Screencasted Lecture Link

This lesson provides an overview of the case study approach to research ranging from the anecdotal/uncontrolled case study to experimentation in N of 1 studies. We also spend a moment considering how we might integrate some of these methodologies in practice.

5.2 The Anecdotal and Uncontrolled Case Study

Has a rich and deep place in history and offers a continuum between anecdotal and uncontrolled to true experiments.

5.2.1 Why Bother with the Anecdotal/Uncontrolled Case Study?

  • source of hypotheses
  • source for developing therapy/treatment techniques
  • study of rare phenomena
  • providing a counter-instance for notions that are considered to be universally applicable
  • selected systematically to prove a point (also a weakness)

5.2.2 Limitations of the (uncontrolled, anecdotal) Case Study

  • self-report, anecdotal nature questions objectivity
  • threats to internal validity
    • history, maturation, testing, instrumentation, statistical regression, selection biases, attrition, combo, diffusion or imitation of treatment
  • threats to external validity
    • absence of standard or replicable procedures.

5.3 Single Case Designs

“As true experiments, single-case designs can demonstrate causal relations and can rule out or make implausible threats to validity with the same elegance of group research” (Kazdin, 2017, p. 273).

Three elements to the single case experiment:

  • Continuous Assessment: Observations on multiple occasions over time prior to and during the period in which the intervention is administered.
    • a distinguishing difference between group (between-group) and single-case research. In the between-group research, there is usually a pre- and post- test given to a treatment and control group.
    • provides several observations over time to allow the comparisons of interest.
  • Baseline Assessment: Assessment for a period of time prior to implementation of the intervention.
    • allows descriptive (provides info about the extent of the client’s problem) and predictive (serves as basis for establishing predictions for performance) functions.
  • Continuous Assessment: Observations on multiple occasions over time prior to and during the period in which the intervention is administered.
    • a distinguishing difference between group (between-group) and single-case research. In the between-group research, there is usually a pre- and post- test given to a treatment and control group.
    • provides several observations over time to allow the comparisons of interest.

5.4 Experimental Design Strategies

5.4.1 ABAB Designs

A family of experimental arrangements in which continuous observations of performance are made over time for a given client (or group of clients). Over the course of the investigation, changes are made in the experimental conditions to which the client is exposed.

  • sometimes termed a reversal design because the phases are reversed to demonstrate the effect of the program.
  • causality is demonstrated by showing that the behavior reverts to or approaches the original baseline after the intervention is withdrawn or altered (during the second A phase).
    • no reversal, no causality
  • ethical issues:
    • Do we withdraw a treatment that’s working well?
    • Do we really shout for joy when withdrawal of the treatment results in deterioration of behavior?

5.4.2 Multiple Baseline Designs

Demonstrates the effectiveness of an intervention by showing that the behavior change accompanies introduction of the intervention at different points in time.

  • That is, once the intervention is presented, it need not be withdrawn or altered to reverse behavior to or near baseline levels. Thus, the clinical utility of the design is not limited by the problems of reverting behavior to pretreatment levels.
  • Key feature: evaluation of change across different baselines. The intervention is introduced to the different baselines at different points in time. Ideally, change occurs when the intervention is introduced in sequence to each of the baselines.
  • Each behavior is observed and graphed separately.
  • Many versions: Sometimes baselines may represent different behaviors, individuals, situations.
  • Ideally behaviors still exposed to the baseline condition do not change until the intervention is applied. If they do, it suggests that the intervention may not have been responsible for change. Rather, extraneous factors (history, testing) may have led to the change. Sometimes there are generalized effects; not ideal.

5.4.3 Changing-Criteria Designs

  • Demonstrates the effect of an intervention by showing that behavior changes in increments to match a performance criterion.
  • A causal relation between an intervention and behavior is demonstrated if behavior matches a constantly changing criterion for performance over the course of treatment.
  • Once performance consistently meets the criterion, the criterion is made more stringent. The effect of the intervention is demonstrated if behavior matches a criterion as the criterion is changed.
  • Kazdin writes that the changing-criterion design is less powerful because the effects of extraneous events could account for a general increase or decrease in behavior. If the researcher can show a unilateral change in behavior over time, causality is better supported.

5.5 The Big Debate: Evaluating Data in Single-Case Designs

Visual Inspection Statistical Evaluation
(traditional) preference of n = 1 investigators available
seems sacrilegious to many provides greater assurance of objectivity (at least on the face of it)
more consistent with the clinical picture

Kazdin’s text identified four criteria for evaluating the single-case design (2017).

Criteria for Data Evaluation in Single-Case Designs
Changes in means (averages) The mean rate of the behavior shows a change from phase to phase in the expected direction.
Change in level When one phase changes to another, a level refers to the change in behavior from the last day of one phase (baseline) and the first day of the next phase (intervention). An abrupt shift facilitates data interpretation.
Change in slope (trend) The speed with which change occurs once the conditions are changed (baseline to intervention, intervention back to baseline).
Latency of change The speed with which change occurs once the conditions are changed (baseline to intervention, intervention back to baseline).

5.5.1 Evaluating Single Case Study Designs

The lecture reviews a series of graphs and describes how they connect back to the criteria. Owing to concerns of including copyrighted material in this printed copy, I am omitting those from this portion of the lesson.

These, though, are the notesthat go with each of the examples:

In the weight loss example (a graph showing a steady deline of weight over 5 weeks):

  • Data is objective
  • Not continuous
  • Stability of problem not demonstrated with abaseline period of observation
  • Lacks multiple cases
  • Cannot rule out: history, maturation, testing, instrumentation, statistical regression toward the mean.

In the “flatulent thoughts” example, there is a period of baseline characterized by 12-21 thoughts per day with a good deal of variability. Once treatment starts, there is a variable progression downard. There is a break and then follow up observation where the thoughts remain low (5-7 a day).

  • Data is objective
  • Continuous observation
  • Stability of problem is demonstrated
  • Immediate and marked effects demonstrated
  • Lacks multiple cases
  • Cannot rule out: history, maturation
  • Good controls for testing, instrumentation, statistical regression toward the mean

In the agoraphobia example there is a period of baseline where there was no actiity spent outside the house. There is a period of intervention where there is a substantial increase. Then aperiod of two-month followup with increased activy and agreat deal of variability. Then 18-months follow-up.

  • Data is objective
  • Continuous observation
  • Stability of problem demonstrated
  • Lacks multiple cases
  • Cannot rule out: history, maturation,
  • Controls for testing, instrumentation, statistical regression toward the mean.

In the bedwetting example there is pretraining during the baseline period. There is an substantial and immediate drop in level when trainng begins and continued success that follows.

  • Data is objective
  • Continuous observation
  • Stability of problem demonstrated
  • Contains multiple cases
  • Controls for history, maturation, testing, instrumentation, statistical regression toward the mean

5.6 Case-Based Time-Series Analysis

“…the practitioner-generated case-based time-series design with baseline measurement fully qualifies as a true experiment and that it ought to stand alongside the more common group designs (e.g., RCT) as a viable approach to expanding our knowledge about whether, how, and for whom psychotherapy works.” (Borckardt et al., 2008, p. 77).

5.6.1 The Trouble with Interpreting N = 1

  • Neither visual inspection nor conventional statistics are to be relied on for analyzing single-patient time-series studies” (p. 82).
  • The problem: auto-correlation
    • Observations are auto-correlated if the value of one observation depends (at least in part) on the value of one or more of the immediately preceding observations.
  • Simulation-modeling analysis (SMA) as a technique for analyzing N = 1, time-series data

Borckardt et al. (2008) provided an example of pain intervention. Please refer to the tables and figures in the Borckardt et al. article.

  • Use Pearson r to calculate the Lag 1 autocorrelation. Result for baseline & treatment: r = .85.
    • A Lag 1 correlation is the degree to which an observation at Time K predicts the observation that comes immediately after it (at Time K + 1). The Lag 1 correlation is simply the correlation between each data point and the point immediately following it.
  • The research question: Is the decrease in pain from baseline (M = 7.78) through tx (M = 3.29) statistically significant?
  • SMA says yes: r (44) = -.69, p = .049, even after controlling for autocorrelation.

SMA answers the question, “mere random variation of pain reports is an unlikely explanation of the phase difference from baseline to treatment, EVEN AFTER controlling for autocorrelation (that is, even after autocorrelation, there is systematic variation in the data and we will attribute it to our intervention).

Another example: What would behavioral activation theory suggest about the relationship between mood and social engagement? Cross-lagged correlations can be used to determine the direction of change. In this case, does change in mood precede behavioral activation, or vice versa? The article includes three examples, an instructional appendix, and a link to free SMA software.

5.7 Incorporating Systematic Evaluation into Clinical Practice

  • Assess and establish treatment goals(Morgan & Morgan, 2001)
    • Clear, specific, explicit.
  • Specify and assess procedures and processes
    • Name the means and processes that are expected to lead to therapeutic change.
  • Select credible measures
    • Identify or develop instruments, scales, measures.
  • Regular, continuous, ongoing assessment
    • Any concerns from internal/external threats to validity? Testing effects? Not an issue if they are part-n-parcel to treatment.
  • Evaluate data/outcomes
    • Maybe SHARE it with client.
    • Linda McDaniels’ example of the Parent Trust client who was so excited to rate herself at the next level.

5.7.1 Kazdin’s (2017) Gloria Case

Kazdin retells the case of clinical that asked clients to self-assess at each visit. Kazdin’s text had three graphs – two of established measures and a third of the “G-scale,” which represented Gloria’s self-rating of progress.

A copy of the scale is shown in the lecture and available in Kazdin’s text. Consider these questions:

  • How did the Clinician attempt to systematic ally evaluate the Clinical Practice?
  • How are evaluation/methods issues similar/different to clinical/case conceptualization issues.
  • How credible is the G scale?
  • Can you imagine doing this in YOUR practice?
  • With what types of clients/circumstances might this work or not work?
  • Looking at Figure 11.5, what can you say about means, level, slope, and latency of change?

5.8 From the Perspective of Philosophy of Science

Morgan and Morgan (2001) addressed the push and pull between statistical/group designs and the single case design.

  • Role of variability is the key.
  • Variability is an irreducible property of natural phenomena (Gould 1996).
    • Empirical statistical studies seek to neutralize variance. We owe this to Plato who saw variation as accidental; ideals lived on a higher plane.
      • N = 1 studies exploit variability, to provide pivotal information about the impact of the IV over time. We owe this to Darwin’s grand unification theory (genetic variation serves as the raw materials on which selective pressures can operate over time).

"It is difficult to envision a clinical application, at least within psychology, medicine, or other service –oriented disciplines, for which the purported goal is the alteration of group means. (2001), p. 123].

References

Borckardt, J. J., Nash, M. R., Murphy, M. D., Moore, M., Shaw, D., & O’Neil, P. (2008). Clinical practice as natural laboratory for psychotherapy research: A guide to case-based time-series analysis. American Psychologist, 63(2), 77–95. https://doi.org/10.1037/0003-066X.63.2.77

Kazdin, A. E. (2017). Research Design in Clinical Psychology, 5th Edition. /content/one-dot-com/one-dot-com/us/en/higher-education/program.html

Morgan, D. L., & Morgan, R. K. (2001). Single-participant research design: Bringing science to managed care. American Psychologist, 56(2), 119–127. https://doi.org/10.1037/0003-066X.56.2.119