14.1 Visual Analysis of Single-Case Data (Level, Trend, Variability)

Visual Analysis of Single-Case Data (Level, Trend, Variability)

Exam Tip: On the QASP-S exam, visual analysis questions test your ability to interpret equal-interval line graphs, cumulative records, and scatterplots without relying on inferential statistics. You must know how to calculate and interpret Level (mean vs. median), Trend (direction, degree, and the split-middle line of progress / quarter-intersect method), and Variability (stability envelope). Pay close attention to rules regarding data paths (never cross phase lines!), the mathematical formula for Percentage of Non-Overlapping Data (PND), and how cumulative records depict response rates through slope.

In Applied Behavior Analysis (ABA), visual analysis of time-series data represents the primary method for evaluating intervention efficacy and establishing experimental control. Unlike traditional psychological research that relies on inferential statistics (e.g., p-values, ANOVA, t-tests) derived from large group designs, behavior analysis focuses on the behavior of the individual over time. Single-case research methodologies allow clinicians and supervisors to continuously monitor behavioral changes, make responsive data-driven clinical adjustments, and observe the immediate and sustained effects of environmental manipulations.

Visual analysis is intentionally conservative. Because behavior analysts require robust, socially significant behavior change, an intervention is considered demonstrably effective only if its impact is readily apparent upon visual inspection of a graph. Subtle statistical blips that require mathematical transformations to detect rarely translate into meaningful real-world improvements for autistic individuals and their families.

---

Anatomy of the Equal-Interval Line Graph

The equal-interval line graph (also known as a frequency polygon or Cartesian coordinate plane) is the most ubiquitous visual display in clinical ABA. In an equal-interval line graph, equal distances on each axis represent equal amounts of the dimension being measured. To prevent optical illusions and biased interpretations, behavior analysts adhere to strict construction standards established by Cooper, Heron, and Heward (2020).

Core Structural Components:

1. Horizontal Axis (Abscissa / X-axis): Depicts the passage of time, represented in successive observation sessions, clock hours, calendar days, weeks, or experimental trials. The x-axis scale must use equal intervals throughout.
2. Vertical Axis (Ordinate / Y-axis): Represents the target behavior's quantifiable dimensional measure, such as frequency, rate (responses per minute), duration (seconds/minutes), latency, inter-response time (IRT), or percentage of opportunities/intervals.
3. The 2:3 or 3:4 Aspect Ratio Rule: To prevent optical distortion—such as flattening a trend by stretching the x-axis or artificially exaggerating a trend by stretching the y-axis—the vertical axis height should be approximately two-thirds to three-fourths (66% to 75%) the length of the horizontal axis.
4. Origin and Scale Breaks: The intersection of the abscissa and ordinate is the origin $(0, 0)$. If there is a prolonged gap in chronological time or an intentional discontinuity in the measurement scale, a scale break (a pair of parallel slashes or zigzag lines) must be drawn on the relevant axis to alert the clinician.
5. Condition Change Lines (Phase Lines): Vertical lines drawn upward from the horizontal axis indicate changes in environmental conditions or independent variables:
   • Solid Vertical Lines: Signify major experimental condition changes (e.g., moving from Baseline to Treatment, or from Treatment to Reversal).
   • Dashed Vertical Lines: Signify minor environmental modifications or parameter adjustments (e.g., changing a reinforcement schedule from FR1 to FR3, or altering a prompt hierarchy).
6. Condition Labels: Descriptive titles positioned horizontally across the top of each experimental phase (e.g., "Baseline," "DRA + Extinction," "Maintenance").
7. Data Points: Geometric figures (circles, squares, triangles) plotted at the exact coordinate intersection of the measurement session (x) and behavioral value (y).
8. Data Paths: Straight solid lines connecting successive data points within a condition.

Critical Rules for Drawing Data Paths:

• NEVER connect data points across a condition change line (solid or dashed).
• NEVER connect data points across a scale break or significant lapse in continuous time (e.g., a 3-week hospitalization or school break).
• NEVER connect data points across consecutive observation sessions in which data were not collected.
• NEVER connect a data point to an axis line or origin unless a data value of zero was actually recorded at that exact session coordinate.

---

The Core Dimensions of Visual Analysis

Visual analysis is conducted systematically in two stages: (1) analyzing data within a single phase to characterize behavioral stability, and (2) analyzing data across adjacent phases to determine the presence, magnitude, and reliability of an intervention effect. Behavior analysts inspect three fundamental properties: Level, Trend, and Variability.

┌─────────────────────────────────────────────────────────────────────────────┐
│                     THE THREE DIMENSIONS OF VISUAL ANALYSIS                 │
├─────────────────┬───────────────────────────────────────────────────────────┤
│ 1. Level        │ The vertical position of the data along the y-axis.       │
│                 │ Measured as mean (average) or median (midpoint) level.    │
├─────────────────┼───────────────────────────────────────────────────────────┤
│ 2. Trend        │ The overall direction and slope of the data path over     │
│                 │ time (increasing, decreasing, zero/stationary).           │
├─────────────────┼───────────────────────────────────────────────────────────┤
│ 3. Variability  │ The degree of scatter or bounce around the trend line or  │
│                 │ mean level; evaluated via the stability envelope.         │
└─────────────────┴───────────────────────────────────────────────────────────┘

1. Level

Level refers to the value on the vertical ordinate scale around which a set of behavioral measures converges. It indicates the overall magnitude or frequency of the behavior during an experimental phase.

• Mean Level Line: Calculated as the arithmetic average of all data point values in that condition. A horizontal line is drawn across the phase at that coordinate.
• Median Level Line: Calculated as the middle value when all data points are arranged in ascending numerical order. The median level is clinically preferred over the mean whenever the data set contains extreme outliers or is significantly skewed.
• Change in Level Across Phases: Evaluated by examining the difference between the data value of the last session in the preceding phase and the data value of the first session in the succeeding phase. An immediate shift in level upon introducing the independent variable provides powerful evidence of experimental control.

2. Trend

Trend describes the general direction and degree of slope taken by the data path across time.

• Direction:
  • Increasing (Accelerating): The data path demonstrates a persistent upward trajectory over time.
  • Decreasing (Decelerating): The data path demonstrates a persistent downward trajectory over time.
  • Zero Trend (Stationary): The data path remains horizontal without a sustained upward or downward drift.
• Degree / Slope: Describes the steepness of the trend line (e.g., steep acceleration, gradual deceleration).
• The Split-Middle Line of Progress (Quarter-Intersect Method): While trend lines can be approximated visually ("freehand"), scientific rigor requires an objective, mathematical method known as the Split-Middle Line of Progress (White & Haring, 1980; Cooper et al., 2020):
  1. Step 1: Divide Data Chronologically into Halves. If there are 10 data points, divide them into two equal halves of 5 points each. (If an odd number of points exists, the middle point is included in both halves).
  2. Step 2: Divide Each Half into Quarters. Find the mid-rate (median y-value) and mid-date (median x-session) for each half. For a half with 5 sessions (Sessions 1, 2, 3, 4, 5), the mid-date is Session 3. Find the median behavioral value of those 5 sessions.
  3. Step 3: Plot the Two Quarter Intersects. Mark the intersection coordinate of mid-date and mid-rate for the first half, and mark the intersection coordinate for the second half.
  4. Step 4: Connect the Quarter Intersects. Draw a straight line passing through both quarter-intersect points across the entire phase.
  5. Step 5: Adjust the Line (Split-Middle Adjustment). Count the total number of data points falling above the line and below the line. Shift the line up or down parallel to itself until an equal number of data points fall above and below the line (e.g., for 10 data points, 5 above and 5 below).

3. Variability

Variability refers to the degree of bounce, dispersion, or fluctuation in the data around the trend or mean level. High variability indicates that unmeasured or uncontrolled environmental variables are influencing the client's behavior.

• Stability Envelope: A visual guideline used to determine whether a data path is stable enough to permit valid experimental conclusions. Commonly, an envelope is established by drawing parallel lines at $\pm 15%$ to $\pm 20%$ of the median level (or around the trend line).
• The 80/20 Stability Standard: A baseline is considered stable if at least 80% of the data points fall within a $\pm 20%$ range around the median level line. High variability during baseline obscures intervention effects and mandates further investigation into environmental setting events before introducing treatment.

---

Immediacy of Effect & Overlap Metrics

When evaluating the functional relation between an independent variable and target behavior across phases, two advanced parameters are essential: Immediacy of Effect and Data Overlap.

Immediacy of the Effect

Immediacy refers to the temporal latency between the introduction (or removal) of the independent variable and the observed change in behavior.

• Immediate Shift: If self-injurious behavior drops from 20 episodes/hour in the final baseline session to 1 episode/hour in the very first treatment session, the immediacy of effect is high, strongly confirming stimulus control.
• Delayed / Latent Shift: If a behavior changes only after 6 or 7 treatment sessions, the clinician cannot easily rule out history, maturation, or external confounding events as alternative explanations for the change.

Percentage of Non-Overlapping Data (PND)

Data overlap refers to the proportion of data points in the intervention phase that fall within the range of values observed during the baseline phase. High overlap suggests weak or questionable treatment efficacy.

The most widely recognized quantitative metric in single-case visual analysis is the Percentage of Non-Overlapping Data (PND) (Scruggs, Mastropieri, & Casto, 1987):

$\text{PND} = \left( \frac{\text{Number of Intervention Data Points Exceeding the Extreme Baseline Value}}{\text{Total Number of Intervention Data Points}} \right) \times 100$

• For Behavior Reduction (Deceleration): Count the number of intervention data points that are strictly lower than the lowest baseline data point.
• For Skill Acquisition (Acceleration): Count the number of intervention data points that are strictly higher than the highest baseline data point.

Established Clinical Interpretation Guidelines (Scruggs & Mastropieri):

• PND > 90%: Highly effective intervention; strong evidence of clinical significance.
• PND 70% to 90%: Moderately effective intervention; acceptable clinical outcome.
• PND 50% to 70%: Questionable or weak effectiveness; requires modification or secondary support.
• PND < 50%: Ineffective treatment; no demonstrated functional relation.

Caveat: PND is exceptionally sensitive to extreme baseline outliers. If a single uncharacteristic baseline session reached 0 episodes, PND for a deceleration target will automatically be 0%, even if all treatment data are near zero. In such instances, clinicians utilize Percentage of Data Exceeding the Median (PEM).

---

Cumulative Records vs. Equal-Interval Graphs

Developed by B.F. Skinner in the experimental analysis of behavior, the cumulative record is a specialized graphical display that shows the total accumulated number of responses emitted over time.

                               CUMULATIVE RECORD DYNAMICS
                 ▲ 
                 │                                          / (Steep Slope = High Rate)
  Total          │                                         /
  Accumulated    │                                        /
  Responses      │                             ──────────/ (Flat Line = Zero Responding)
                 │                            /  
                 │                           / (Moderate Slope = Moderate Rate)
                 │                          /
                 │                         /   *Note: Line NEVER decreases*
                 └────────────────────────┴────────────────────────────────────────►
                                        Time / Sessions

Defining Characteristics of Cumulative Records:

1. Monotonically Increasing: The cumulative data path can NEVER decrease. Each response emitted is added to the previous running total ($Y_{n} = Y_{n-1} + R$).
2. Slope Depicts Response Rate: The steepness (slope) of the line directly indicates the rate of responding:
   • Steep Slope: Rapid, high-rate responding.
   • Moderate Slope: Steady, moderate-rate responding.
   • Flat Horizontal Line: Complete absence of responding (rate = 0). It does not mean the behavior went down; it means zero responses occurred during that time.
3. Pen Reset: In traditional electromechanical recorders, when the pen reaches the top of the chart paper, it resets instantly back to the baseline zero line and resumes accumulation.
4. Clinical Utility in ASD: Ideal for tracking cumulative mastery of skills (e.g., cumulative vocal tacts acquired across months) or monitoring responding under precise operant reinforcement schedules.

---

Scatterplot Data Displays

Pioneered by Touchette, MacDonald, and Langer (1985), a scatterplot is an assessment data display that plots behavioral occurrences across standardized intervals of time (e.g., 15-minute or 30-minute blocks) across successive days.

Clinical Mechanics and Patterns:

• Time of day is arranged on the vertical y-axis (e.g., 8:00 AM to 4:00 PM in 30-minute rows), and calendar days are arranged on the horizontal x-axis.
• Cells are shaded according to behavioral density: empty (zero occurrences), slash (low frequency; 1-2 occurrences), or filled black (high frequency; 3+ occurrences).
• Clinical Value: Scatterplots do not demonstrate functional relations, but they reveal temporal, environmental, and activity-correlated patterns. For instance, if severe aggression consistently clusters between 11:30 AM and 12:00 PM Monday through Friday, the QASP-S investigates specific antecedent events occurring during that daily window (e.g., chaotic cafeteria transitions, hunger MOs, or noisy unstructured gym environments).

---

Summary Matrix of Visual Analysis Parameters