ABA Graph Analysis – Beyond Just Fidelity Data in ABA

By Jazzmyn Mijic, M.S., B.C.B.A. | Posted: July 2026 | Category: ABA Data Analysis & Clinical Supervision

When BCBAs analyze a learner’s progress, we are not simply looking at whether the data are increasing or decreasing. We are also looking for patterns that may help us understand why the learner is responding differently across sessions, providers, environments, or teaching conditions.

One tool I frequently use during graph analysis is the per-provider view in CentralReach. This allows me to compare the learner’s rate of responding with each person implementing the program.

A graph like the one shown above may immediately raise questions. One provider may have consistently high percentages, while another provider’s data may be variable or considerably lower.

The first thing many BCBAs consider is treatment fidelity—and fidelity is absolutely important. However, fidelity is not the only variable that should be examined.

The graph tells us that there may be a difference. It does not automatically tell us why that difference exists.

Treatment Fidelity Is an Important Starting Point

Treatment fidelity refers to whether the program is being implemented as written.

A BCBA may review whether each provider is:

• Delivering the correct discriminative stimulus or instruction
• Following the written teaching procedures
• Using the designated prompt hierarchy
• Providing reinforcement according to the program
• Implementing error-correction procedures correctly
• Recording data according to the operational definitions

If one provider is consistently implementing the program differently, this may contribute to lower or more variable responding.

However, we should be careful not to assume that lower data automatically mean poor implementation. Before reaching that conclusion, the BCBA should examine the entire clinical context.

Consider the Number of Learning Opportunities

Percent-correct data can look very different depending on the number of trials conducted.

For example, a provider who conducted two trials may have data that appear extremely high or extremely low based on only a small number of responses. Another provider may have conducted 15 or 20 trials across different activities, materials, and environments.

A percentage alone does not show the BCBA how many opportunities were presented.

When reviewing per-provider data, consider:

• How many trials did each provider conduct?
• Were enough trials conducted to represent the learner’s current performance accurately?
• Were the trials distributed across the session?
• Were natural opportunities included?
• Did one provider conduct substantially more acquisition trials than another?

Trial counts provide important context for interpreting percentages and trends.

Evaluate Stimulus Control Across Providers

A learner may respond correctly with one provider because the skill is under the control of that person’s voice, body position, materials, gestures, or other unintended cues.

This can create the appearance that a skill has been learned when it has not yet generalized across people.

The BCBA should evaluate whether the learner responds to the relevant instruction rather than responding only when a particular provider presents it.

Questions to consider include:

• Does the learner respond when different providers present the same instruction?
• Are providers using the same wording?
• Is one provider unintentionally giving additional cues?
• Does the learner wait for a gesture, facial expression, or movement before responding?
• Has the skill been systematically generalized across adults?

Differences across providers may indicate that additional generalization programming is needed rather than simply more acquisition teaching.

Review Prompt Levels and Prompt-Fading Procedures

Two providers may technically be running the same program while using very different levels of assistance.

One provider may use full verbal prompts, partial verbal prompts, phonemic prompts, gestures, or modeling. Another provider may wait for an independent response.

If prompt levels are not clearly defined and consistently documented, the resulting data may not represent the same response requirement.

For example, one provider may count a prompted response as correct, while another provider records only independent responses as correct. Both providers may believe they are following the program, but their data will not be comparable.

The BCBA should review:

• Which prompt levels were used?
• Were prompts recorded accurately?
• Were prompts delivered according to the written hierarchy?
• Was the learner given enough time to respond independently?
• Did providers fade prompts consistently?
• Are all providers using the same definition of a correct response?

Clear prompt definitions and examples within the program instructions can significantly improve consistency.

Examine Data-Collection Procedures

Variability may also be related to differences in how providers collect data.

One provider may record the first response. Another may record the final response after several prompts. One provider may score a partially correct response as correct, while another scores it as incorrect.

Before making major program changes, the BCBA should confirm that everyone is measuring the same behavior in the same way.

This may require:

• Reviewing the operational definition
• Conducting direct observations
• Comparing provider scoring during the same trial
• Completing interobserver agreement checks
• Clarifying when a response should be scored as correct, incorrect, prompted, or no response
• Retraining providers on the data-collection system

A graph is only as accurate as the data entered into it.

Compare Materials and Teaching Conditions

Materials can also influence responding.

One provider may use familiar pictures, while another introduces unfamiliar examples. One provider may present materials in a clean visual field, while another presents them among several distracting items. The size, quality, position, and familiarity of the stimuli may all affect performance.

The BCBA should determine whether providers are using:

• The same program materials
• Similar numbers of stimuli
• Comparable examples of the target
• The same presentation format
• Materials with similar levels of difficulty
• Appropriate distractors

Differences in materials are not always inappropriate. Generalization requires variation. However, the BCBA should know when materials have changed so the resulting data can be interpreted accurately.

Consider Setting Events

Variables outside the immediate teaching interaction may also affect the learner’s performance.

Setting events can temporarily change how likely a learner is to attend, respond, participate, or benefit from reinforcement. These variables may not be visible on the graph.

Examples may include:

• Poor sleep
• Illness
• Hunger
• Changes in routine
• Family stressors
• Changes in school schedules
• Medication changes
• Inconsistent administration of prescribed medication
• Changes in medications used to support sleep
• Missed meals
• Increased sensory discomfort

For example, if a caregiver inconsistently administers a child’s prescribed ADHD medication, the learner’s attending and responding may differ significantly across days. Similarly, inconsistent use of melatonin or another caregiver-managed sleep support may affect sleep quality and performance during the following session.

Providers should not make medication recommendations outside their scope of competence. However, BCBAs can document relevant caregiver-reported changes, consider how those changes may relate to behavioral data, and encourage families to discuss medication concerns with the prescribing medical professional.

Avoid Blaming the Provider Based on the Graph Alone

Per-provider graphs are valuable because they help us identify patterns that require further investigation. They should not be used to label one provider as “good” and another as “bad.”

Lower data with one provider may reflect:

• Fewer teaching opportunities
• More difficult targets
• Less familiar materials
• Reduced stimulus control
• More independent-response requirements
• Differences in prompting
• Inconsistent data collection
• A challenging session
• Setting events
• A genuine need for additional provider training

The purpose of graph analysis is not to assign blame. It is to determine what variables may be affecting learning and what clinical actions are needed next.

Planning the Next Clinical Steps

After identifying variability across providers, the BCBA may:

1. Review the number of trials conducted. Confirm that percentages are based on enough opportunities to support meaningful interpretation.
2. Observe the program directly. Watch each provider implement the target rather than relying on the graph alone.
3. Compare prompting procedures. Confirm that providers use and record the same prompt hierarchy.
4. Review data definitions. Ensure that correct, incorrect, prompted, and no-response trials are scored consistently.
5. Standardize materials when appropriate. Begin with comparable teaching conditions before systematically programming generalization.
6. Assess stimulus control. Determine whether the learner responds to the instruction or to unintended provider-specific cues.
7. Review possible setting events. Consider sleep, illness, schedule changes, caregiver-reported medication variables, and other relevant events.
8. Provide supportive performance feedback. Retrain or clarify procedures when needed without assuming intentional provider error.
9. Program across providers. Intentionally teach and reinforce responding with multiple people.
10. Continue monitoring the graph. Evaluate whether the clinical changes reduce variability and produce a more stable, increasing trend.

Whenever possible, change one major variable at a time. This makes it easier to determine whether the modification improved the learner’s performance.

The Goal Is Consistent Learning Across People

A learner should not demonstrate a skill only with the person who originally taught it.

Our goal is to establish skills that remain strong across providers, caregivers, environments, materials, and naturally occurring situations. Per-provider data help BCBAs identify when that consistency has not yet been achieved.

Fidelity data are an essential part of that analysis—but they are one part of a much larger clinical picture.

Strong graph analysis requires the BCBA to examine the numbers, observe implementation, ask questions, evaluate environmental variables, and make thoughtful program modifications. When we do this well, we are better able to reduce unnecessary variability, strengthen treatment integrity, and support an accelerating trend in meaningful learning.

---

About the Author

Jazzmyn Mijic, M.S., B.C.B.A. is a Board Certified Behavior Analyst with experience in clinical leadership, staff training, caregiver collaboration, program development, supervision, and the treatment of children with autism and related developmental needs. Through Mrs. Jazzmyn BCBA, she creates practical ABA resources, clinical program libraries, and continuing education content designed to support high-quality, ethical, and individualized services.

Disclaimer

This article is intended for educational and informational purposes only. It does not replace individualized clinical assessment, professional judgment, medical advice, or consultation with the learner’s treatment team. BCBAs should make decisions based on direct observation, reliable data, client-specific variables, professional ethics, applicable regulations, and their scope of competence. Medication-related concerns should be directed to the prescribing medical professional. Any clinical image or example used for education should be appropriately de-identified and shared in accordance with privacy requirements and organizational policies.