Single-System Design
- Introduced by
- Martin Bloom, Joel Fischer & John G. Orme (codification in social work)
- Year
- 2009
- Type
- Time-series design for evaluating intervention with a single client system
- Family
- Single-case evaluation
What does it do?
A single-system design is a time-series approach to evaluating practice in which a single client system — an individual, family, group, or organization — is measured repeatedly on a clearly defined target before and during (and sometimes after) an intervention.
By tracking the same system over time rather than comparing a treatment group to a control group, it lets a practitioner judge whether their own intervention is associated with change in the people they actually serve. It is the methodological backbone of the 'accountable professional' tradition codified by Bloom, Fischer, and Orme.
The intuition behind the method
Instead of asking 'does this treatment work on average across many people?', a single-system design asks 'is this client getting better while I work with them?'. You first watch the target behavior or problem without intervening — the baseline (A) — to establish the natural pattern. Then you introduce the intervention (B) and keep measuring. If the trend, level, or variability of the target shifts in the desired direction when, and only when, the intervention is in place, that change becomes the evidence. The baseline is the client serving as their own comparison, which is why a stable, well-documented baseline is everything.
History
Single-system logic descends from the experimental analysis of behavior in mid-20th-century psychology, where researchers like B. F. Skinner studied behavior change within individual organisms over time. Social work adopted and reframed the approach during the 1970s empirical-practice movement, which urged practitioners to be accountable by measuring their own outcomes. Martin Bloom, Joel Fischer, and John Orme's textbook Evaluating Practice, first published in the 1980s and revised across multiple editions, became the canonical social-work codification, and Alan Kazdin's work systematized the broader single-case methodology for clinical and applied settings.
When to use it
Use a single-system design when you need to evaluate whether a specific intervention is helping a specific client system and a group comparison is impossible, unethical, or irrelevant — the typical situation in direct clinical and community social work practice. It fits targets that can be measured repeatedly and frequently with the same instrument. It is poorly suited to outcomes that can only be measured once, to problems with no stable baseline (acute crises requiring immediate action), or to questions about average effects across a population, where group designs are appropriate. Carryover, history, and maturation threats must be considered when interpreting any single case.
Limitations and considerations
- Findings apply to the single system studied; external validity (generalization to other clients) rests on replication across cases, not on a single design.
- A basic AB design cannot rule out history or maturation as alternative explanations; stronger logic requires withdrawal (ABAB) or multiple-baseline structures.
- An unstable or trending baseline can make any later change uninterpretable, yet ethically a baseline cannot always be extended.
- Autocorrelation in time-series data violates the assumptions of many ordinary statistical tests, so naive significance testing can mislead.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Integrates evaluation directly into ongoing practice, so the practitioner gets continuous feedback on whether the current client is improving.
- Requires only one client system, making rigorous outcome evaluation feasible in everyday casework where randomized groups are impossible.
- Each client serves as their own comparison, removing between-person confounds such as differing histories or severity.
- Repeated measurement reveals the timing and trajectory of change, not just a before-after snapshot, supporting timely adjustment of the intervention.
Applications
- A clinical social worker tracking a client's weekly depression or anxiety scale scores across baseline and treatment to judge therapeutic progress.
- Evaluating a behavior-management plan for a child by counting a target behavior at home and at school before and after the plan is introduced.
- Monitoring a family-preservation intervention by repeatedly rating family functioning during service delivery.
- Assessing an organizational or community intervention — such as a new intake procedure — by tracking a process indicator over successive weeks.
- Teaching evidence-based practice by having students evaluate their own field placement cases with a structured AB or ABAB design.
How it works
How it works
The practitioner defines a single, observable, repeatedly measurable target — a behavior count, a symptom score, a standardized scale, or a self-anchored rating — together with how, when, and by whom it is recorded. A good target is specific, sensitive to change, and feasible to measure on every occasion.
Before intervening, the target is measured on several occasions to form the baseline phase A. The baseline ideally shows a stable level or a trend running opposite to the hoped-for change, so that later improvement cannot be dismissed as a continuation of what was already happening. Three or more data points are a common minimum.
Formula
A = \{x_{1}, x_{2}, \dots, x_{n_A}\}The intervention is introduced and the same target continues to be measured on the same schedule to form phase B. Holding measurement constant across phases is what makes the two phases comparable; only the presence of the intervention is meant to change.
Formula
B = \{y_{1}, y_{2}, \dots, y_{n_B}\}The phases are plotted on a single chart and read by eye for changes in level (the jump between phase averages), trend (the slope within each phase), variability, and the immediacy and overlap of data points. Visual analysis is the primary decision tool in single-case work; a clear, non-overlapping shift coinciding with the intervention is the strongest within-design evidence of effect.
When visual analysis is ambiguous, simple statistical aids supplement it. A common rule plots a band of two standard deviations around the baseline mean and flags intervention points that fall outside it; related tools include the celeration line, the proportion/frequency approaches, and effect sizes such as percentage of non-overlapping data. These quantify, rather than replace, the judgment that change is real.
Formula
\bar{x}_A \pm 2\,\text{SD}_A \;\Rightarrow\; \text{flag } y_t \text{ outside band}Common pitfalls
- Starting the intervention before the baseline is stable, so improvement cannot be distinguished from a pre-existing trend.
- Changing the target definition or measurement procedure between phases, which makes the phases non-comparable.
- Reading a small overlap between phases as 'no effect' when level and trend clearly shifted, or vice versa — over-relying on a single feature of the graph.
- Applying ordinary t-tests or regression to short, autocorrelated phase data without accounting for serial dependence, inflating apparent significance.
Frequently asked questions
How many baseline data points do I need?
There is no universal number, but three is a common practical minimum and more is better when the data are variable. What matters is not the count but stability: the baseline should show a flat level or a trend opposite to the desired change, so that subsequent improvement cannot be explained as a continuation of the baseline pattern. In crisis situations where withholding intervention is unethical, a retrospective or concurrent baseline from records may substitute.
What is the difference between an AB design and an ABAB design?
An AB design has one baseline phase (A) followed by one intervention phase (B); it documents change but cannot strongly rule out coincidental causes. An ABAB (withdrawal/reversal) design removes and reintroduces the intervention, so if the target improves under B, worsens when B is withdrawn, and improves again when B returns, the repeated correspondence is much stronger evidence that the intervention — not history or maturation — drove the change.
Can I use ordinary statistics on single-system data?
With caution. Time-series data from one system are typically autocorrelated, meaning consecutive observations are not independent, which violates the assumptions of standard t-tests and regression and tends to inflate significance. Practitioners more often rely on visual analysis supplemented by tools designed for the setting — the two-standard-deviation band, celeration lines, and non-overlap effect sizes — or on methods that explicitly model serial dependence.
Related methods
Sources
Bloom, M., Fischer, J., & Orme, J. G. (2009). Evaluating Practice: Guidelines for the Accountable Professional (6th ed.). Pearson/Allyn & Bacon.
ISBN: 9780205458066Copied.Kazdin, A. E. (2011). Single-Case Research Designs: Methods for Clinical and Applied Settings (2nd ed.). Oxford University Press.
ISBN: 9780195341881Copied.