visual-analysis

# Visual Analysis Visual analysis is the primary method for evaluating behavior change in single-subject experimental designs. It relies on systematic inspection of graphed data to determine whether a functional relationship exists between the independent variable and the dependent variable. ## Foundations Visual analysis is preferred in behavior analysis because: - It is conservative—only detects robust, clinically meaningful effects. - It maintains the individual as the unit of analysis (no group averaging). - It is accessible to practitioners without advanced statistical training. - It allows ongoing, real-time decision making throughout treatment. Visual analysis is not impressionistic or casual—it follows a structured, systematic process. ## Within-Condition Analysis Examine the data within each phase (baseline, intervention, etc.) on three characteristics. ### Level The central tendency of the data within a phase. - **Mean**: Average of all data points. Sensitive to outliers. - **Median**: Middle value when data are ordered. Preferred when data contain outliers or are skewed. - **Range**: Distance from lowest to highest data point. Draw a level line (mean or median) through the data in each phase to facilitate between-condition comparisons. ### Trend The systematic direction of the data path over time. - **Direction**: Increasing (accelerating), decreasing (decelerating), or zero (flat). - **Magnitude**: Steepness of the trend line. A steep trend represents rapid change. - **Stability**: How consistently data points fall near the trend line. #### Split-Middle Technique for Trend Estimation 1. Divide the phase data in half (by the number of data points). 2. Find the median data value and median session number for each half. 3. Draw a line connecting the two median points. 4. Adjust the line up or down so that half the data points fall above and half below. ### Variability The extent to which data points deviate from the level and trend. - **Stability envelope**: Draw a band around the median (or mean) equal to ±20% of the median value. If ≥80% of data points fall within this band, the data are considered stable. - **High variability** within a phase makes it harder to detect treatment effects and suggests the need for longer phases or investigation of uncontrolled variables. ## Between-Condition Analysis Compare adjacent phases to evaluate the effect of the independent variable. Six features guide the comparison. ### 1. Change in Level - **Absolute level change**: Difference between the last data point of one phase and the first data point of the next. - **Relative level change**: Difference between the median of the last half of one phase and the median of the first half of the next. - Large, immediate level changes in the expected direction support a functional relationship. ### 2. Change in Trend Compare the trend direction and magnitude across phases. - If baseline shows a flat or counter-therapeutic trend and intervention shows a therapeutic trend, this supports the treatment effect. - If baseline already shows a therapeutic trend, introducing an intervention makes it difficult to attribute further change to the IV. ### 3. Change in Variability Compare the stability of data across phases. - Ideally, variability decreases in the intervention phase, indicating consistent treatment effects. - Increased variability in the intervention phase may suggest inconsistent implementation or uncontrolled variables. ### 4. Immediacy of Effect How quickly does behavior change after the phase change? - A rapid, immediate change following introduction or withdrawal of the IV strongly supports a functional relationship. - Delayed effects are more ambiguous—they may reflect the IV or other coincidental variables. - Some interventions have expected delays (e.g., extinction bursts before decrease). ### 5. Overlap Between Phases The proportion of data points in adjacent phases that share the same range. - **Percentage of Non-overlapping Data (PND)**: Percentage of intervention data points that exceed the most extreme baseline data point. Higher PND = stronger effect. - **Percentage Exceeding the Median (PEM)**: Percentage of intervention data points that exceed the baseline median. More robust to outliers than PND. - **Tau-U**: Non-overlap index that also controls for baseline trend. Provides a p-value and confidence interval. Low overlap between adjacent phases supports a functional relationship. ### 6. Consistency of Data Patterns Across Similar Phases In designs with replication (A-B-A-B, multiple baseline), examine whether similar data patterns replicate. - Behavior should change in the same direction each time the IV is introduced. - Behavior should revert (or remain changed, in multiple baselines) consistently. - Three demonstrations of the effect at three different points in time establish a functional relationship. ## Practical Guidelines ### Minimum Data Points - **Per phase**: 3 data points minimum, 5 preferred for trend estimation. - **Baseline stability**: Do not introduce intervention until baseline is stable or shows a counter-therapeutic trend. A therapeutic trend in baseline precludes demonstration of experimental control. ### Order of Analysis 1. Examine within-condition features for each phase separately. 2. Compare between-condition features for adjacent phases. 3. Evaluate consistency across all replications in the design. 4. Integrate all features into an overall judgment about functional relationship. ### Common Errors - Introducing intervention when baseline is unstable or trending in the therapeutic direction. - Changing phases with too few data points. - Ignoring variability when interpreting level changes. - Over-relying on a single between-condition feature (e.g., only looking at level change and ignoring overlap). - Failing to demonstrate at least three replications of effect. ## Summary Table | Feature | Within-Condition | Between-Condition | |---------|-----------------|-------------------| | Level | Mean, median, range | Absolute and relative level change | | Trend | Direction, magnitude, stability | Change in direction, magnitude | | Variability | Range, stability envelope | Change in variability | | Immediacy | — | Rapidity of change at phase line | | Overlap | — | PND, PEM, Tau-U | | Consistency | — | Replication across phases/tiers | ## Key References - Cooper, J. O., Heron, T. E., & Heward, W. L. (2020). *Applied Behavior Analysis* (3rd ed.). Pearson. - Fisher, W. W., Kelley, M. E., & Lomas, J. E. (2003). Visual aids and structured criteria for improving visual inspection and interpretation of single-case designs. *JABA*, 36, 387–406. - Lane, J. D., & Gast, D. L. (2014). Visual analysis in single case experimental design studies. *Neuropsychological Rehabilitation*, 24, 445–463. - Parker, R. I., Vannest, K. J., Davis, J. L., & Sauber, S. B. (2011). Combining nonoverlap and trend for single-case research: Tau-U. *Behavior Therapy*, 42, 284–299.