Yes, single subject designs (also called single case designs) require baseline data in nearly every form. The baseline phase is so central to this methodology that it serves as the primary tool for establishing experimental control. Without it, there is no way to determine whether an intervention actually caused a change in behavior or whether the change would have happened on its own.
Why Baseline Data Is Essential
The baseline phase records the existing levels and patterns of a behavior before any intervention begins. This creates a benchmark, a prediction of what the behavior would look like if nothing changed. When an intervention is introduced and the behavior shifts in a way the baseline data would not have predicted, that contrast is what allows researchers to claim the intervention worked.
This comparison between baseline and intervention phases is the foundation of experimental control in single subject research. Every major design variant, from simple A-B designs to more complex arrangements, relies on some form of this comparison. The “A” in the familiar A-B notation stands for the baseline phase, and it comes first for a reason: you need to know where someone started before you can measure where an intervention took them.
What Makes a Baseline “Good Enough”
Not all baseline data serves its purpose equally well. Researchers typically wait for the baseline to show minimal or no trend before introducing the intervention. If the data are already moving in the direction you’d expect the treatment to produce (say, a person’s target skill is already improving before you intervene), any further improvement during treatment could simply be a continuation of that existing trend rather than an effect of the intervention.
Stability matters in two ways. First, the data should not show a strong upward or downward slope. Second, the data points should not bounce around too wildly. A graph can be perfectly flat on average but still unstable if individual data points swing far above and below the mean. Researchers evaluate this spread, sometimes using cutoff values like a standard deviation of 1.0 or 1.5 calculated from the last three baseline data points, to decide whether the phase is stable enough to move forward.
This means baseline phases don’t have a fixed length. You collect data until the pattern is clear and stable. If the first few sessions show erratic results, you keep collecting. The What Works Clearinghouse guidelines for single case research specifically recommend waiting to see whether a baseline series stabilizes as more data are gathered before proceeding.
How Different Designs Use Baseline Data
The role of baseline data shifts slightly depending on which single subject design you’re using, but it never disappears entirely.
Reversal (A-B-A-B) Designs
These designs use baseline data twice. You collect baseline, introduce the intervention, remove the intervention (returning to baseline conditions), then reintroduce the intervention. If the behavior improves during intervention, returns toward baseline levels when the intervention is removed, and improves again when the intervention is reinstated, that pattern provides strong evidence of a functional relationship. The repeated return to baseline conditions is what gives this design its power.
Multiple Baseline Designs
In a multiple baseline design, baseline data collection is staggered across two or more participants, settings, or behaviors. One participant might have three baseline sessions before intervention begins, another might have six, and a third might have nine. If the behavior changes only when the intervention is introduced for each participant (and not before), the staggered baselines rule out coincidence.
More recent variations stagger baselines across two dimensions simultaneously, for example across both participants and settings. This creates more opportunities to demonstrate experimental control. If baseline trends happen to be problematic in one dimension, the researcher can still detect treatment effects across the other. Each combination of participant and setting gets its own assigned baseline length, often arranged using a structured matrix to ensure at least three different baseline lengths within each dimension.
Alternating Treatments Designs
This is the one design where a traditional extended baseline phase is not always strictly required. Alternating treatments designs compare two or more interventions by rapidly switching between them across sessions. Because the comparison is primarily between treatments rather than between baseline and treatment, some researchers begin alternating conditions without a separate initial baseline phase. That said, many implementations still include a brief baseline to establish a starting point.
How Researchers Analyze Baseline Data
Single subject designs rely heavily on visual analysis rather than statistical tests. Researchers graph the data and look for visible changes across phases. Three features guide this analysis.
- Level: the overall amount of the behavior relative to the vertical axis. A jump in level from baseline to intervention (for example, going from averaging 2 correct responses to 8) suggests an effect.
- Trend: the direction the data are moving over time. Flat, increasing, or decreasing patterns within a phase tell you whether the behavior was already changing before the intervention started.
- Variability: how much the data points spread or fluctuate around the trend line. High variability in the baseline makes it harder to detect a real treatment effect because the “noise” obscures the signal.
Researchers examine these three features within each phase and then compare them across phases. A convincing result shows a clear, immediate change in level, trend, or variability at the exact point where the intervention was introduced, one that would not be predicted from the baseline pattern alone.
When Baseline Collection Gets Complicated
There are practical situations where collecting an extended, stable baseline is difficult. In clinical or educational settings, a behavior might be dangerous (severe self-injury, for instance), making it ethically uncomfortable to simply observe without intervening. In these cases, researchers sometimes use a B-A-B design, starting with treatment first, then withdrawing it to see if the behavior worsens, and reintroducing treatment. This sacrifices the initial baseline phase but still uses a no-treatment phase (the “A” in the middle) to demonstrate that the intervention is responsible for the change.
Even in these modified designs, baseline data is not eliminated. It is repositioned. The withdrawal phase functions as a delayed baseline, providing the same type of comparison point. The logic of single subject design always returns to the same core question: what does the behavior look like without the intervention? Baseline data, whether collected at the start or embedded later in the design, is how that question gets answered.