Treatment Fidelity Checklist for ABA, With Scoring

A learner's data can only be interpreted if you know the procedure that produced it. If a program is flat, there are two very different explanations: the procedure does not work for this learner, or the procedure was not run the way it was written. Treatment fidelity is how you tell those apart. This post defines the terms, gives a working treatment fidelity checklist for ABA that you can copy, and shows exactly how to score it.

Treatment fidelity vs integrity in ABA: the terms

Searchers use treatment fidelity, treatment integrity, procedural fidelity and procedural integrity as near-synonyms, and in much of the applied literature they are used interchangeably. All of them refer to the extent to which an intervention is implemented as it was designed. Some authors draw finer lines. A common one reserves procedural fidelity for the specific steps of a procedure, and treatment integrity for the broader question of whether the whole intervention package was delivered as planned. Because usage varies by author and by payer, define the term you use in your own documents, and do not assume a reader shares your meaning.

A second distinction is more useful in practice. Fidelity can be looked at along several dimensions, and the literature on implementation commonly discusses at least these:

• Adherence: were the steps done, and done in order?
• Quality: were they done well, such as prompts delivered promptly and with the right level?
• Exposure or dosage: did the learner get the planned number of opportunities?

The checklist below mostly measures adherence, with a few items for quality and exposure. The ideal measure for your setting is for the BCBA to decide.

Why measure it at all

Three reasons, all practical.

1. It protects clinical decisions. Changing a program because it "isn't working" when it was run at half strength wastes time and can lead to the wrong change.
2. It targets training. Item-level scores show which step staff skip, which is far more useful than a general "needs more training."
3. It documents the service. A record that procedures were implemented as written supports the claim that the service delivered matched the plan.

Fidelity data also overlaps with the supervision and competency work of the RBT's month. See RBT competency assessment and supervision for how observation and feedback are usually structured.

How to build a checklist for any procedure

1. Task-analyze the written procedure. Break the protocol into observable steps in the order they occur. If the protocol does not state a step, add it to the protocol first.
2. Write each item as one observable action. "Delivers the instruction once, in a neutral tone" is checkable. "Uses good prompting" is not.
3. Include errors of commission as well as omission. Failing to deliver reinforcement is an omission. Delivering reinforcement after an incorrect response is a commission, and it is just as damaging.
4. Decide what "not applicable" means. Some steps do not occur in a given observation, for example error correction when there were no errors. Mark those N/A and leave them out of the score.
5. Decide the observation unit. Score per trial, per opportunity, or once for the session. Say which on the sheet.

A working checklist: discrete trial teaching

This checklist is for a single target in a discrete trial program with a prompt-delay procedure. Score each item for each trial observed, then summarize. Adapt it to your written protocol; items here are illustrations and the exact steps for any learner come from that learner's program.

#	Item (observable)	Dimension
1	Materials ready and the learner attending before the trial begins	Adherence
2	Instruction delivered once, with the exact wording in the protocol	Adherence
3	Prompt delivered at the correct level and after the correct delay (for example, immediately in early teaching, after the planned delay in later teaching)	Quality
4	Correct response followed by reinforcement within 3 seconds, as defined in the protocol	Adherence
5	Reinforcement is the specified item or praise, not a different one	Adherence
6	Incorrect response followed by the planned error correction (for example, model, re-present, prompt, reinforce the prompted response)	Adherence
7	No reinforcement delivered after an incorrect response	Adherence (commission)
8	Data recorded for the trial before the next trial begins	Adherence
9	Intertrial interval within the planned range	Quality
10	Number of trials in the block matches the plan	Exposure

Item 10 is scored once per block. Items 1 to 9 are scored on each trial where they apply.

A second checklist: differential reinforcement for escape

Fidelity also applies to behavior plans. Here is a short checklist for a plan in which the learner requests a break instead of escaping a demand by aggression.

#	Item
1	Demand presented using the planned wording and visual
2	Break card visible and within reach at the start of the work period
3	When the learner requests a break, a break is given within 5 seconds
4	Break lasts the planned time and the demand then resumes
5	When problem behavior occurs, the demand is not removed contingent on it
6	Staff follows the safety steps in the plan if aggression occurs

The point of item 5 is that the most common fidelity failure in escape plans is the well-meant removal of a demand after the learner escalates. It silently reinforces the behavior the plan is trying to reduce.

How to score it

The standard calculation is simple:

Fidelity (%) = steps implemented correctly / total applicable steps, times 100.

Worked example. An observer watches one 10-trial block of the DTT checklist and scores the items on the first three trials. For brevity, take the totals across those three trials: 27 item opportunities were applicable (nine items across three trials). Of those, 6 items were N/A (items such as error correction, which did not occur when the learner responded correctly), leaving 21 applicable. Staff implemented 18 correctly and missed 3: one late prompt, one inconsistent reinforcer, and one missed data entry. So fidelity is 18 / 21 = 85.7 percent.

A fuller sheet separates the numbers so the percentage cannot hide a pattern:

Trial	Applicable items	Correct	Score	Missed items
1	9	8	88.9%	#3 prompt late
2	7 (2 N/A)	6	85.7%	#8 data not recorded
3	5 (4 N/A)	4	80.0%	#5 wrong reinforcer

Totals: 21 applicable, 18 correct, 85.7 percent. Note that the trial rows come out differently from the total because each trial has different applicable items; always report the total from the sums, not by averaging trial percentages, or say that you averaged.

Reading the score

• Look at the missed items first. A score of 85.7 percent made of three different misses is a different training problem than 85.7 percent made of the same miss three times.
• Set the criterion with the BCBA, not from the internet. You will see numbers like 80 or 90 percent quoted online. They are conventions, not rules, and the right level depends on the procedure, the learner and the risk. Write the criterion you chose in the plan.
• Treat a low score as information about the system. Poor fidelity can reflect an unclear protocol, too little training, or a program that is too hard to run as written.
• Score enough sessions to be believable. One observation is a snapshot. Many practices sample across staff, programs and times of day; how often is a clinical decision.

Fidelity and agreement are different checks

Fidelity asks whether the procedure was run correctly. Interobserver agreement (IOA) asks whether two observers saw the same thing. You need both: fidelity data scored by one observer is only as trustworthy as that observer. See how to calculate IOA. If you also want the fidelity observation itself checked, a second observer can score it, and the same agreement formulas apply, and ABA IOA software covers the agreement side.

Who scores, and when

The person who scores fidelity is usually the supervising BCBA or a senior clinician, observing live, in person or by a real-time video connection where your payer and state allow it. Peer scoring works for training purposes. Self-scoring is the weakest option because the person delivering the procedure is least able to see the misses, though a self-check against the same list is a useful habit between observations.

Observe at different times. A fidelity score taken on a day the supervisor is visibly watching tends to be higher than an ordinary day, so vary when and which program you sample. Try to observe each staff member on each high-stakes procedure at least occasionally, with extra attention after a protocol change, after a new hire's first weeks, and after any period of flat data. How often is a clinical and organizational decision; state your schedule in a written policy rather than copying a number from a blog.

Caregivers can be scored too. When a parent runs a bedtime plan, the same approach applies: list the steps, watch a rehearsal or a live attempt, score the steps, give feedback, and re-score. The caregiver-training section of a treatment plan is the natural home for those criteria.

A blank template to build your own

Header field	Entry
Procedure and learner code	Name the written protocol and its version or date
Staff observed and observer	Names and roles
Date, setting, duration	Actual times
Unit scored	Per trial, per opportunity, or per session
Items	One observable action each; mark Y, N or N/A
Total	Correct, applicable, percent
Missed items and feedback given	Specific, with the date of re-score
Criterion	The level the BCBA set, and the action if below it

What to do with the result

1. High fidelity and flat data: the procedure itself needs changing. This is the clinical decision fidelity makes safe to make.
2. Low fidelity and flat data: fix implementation first. Retrain with behavioral skills training (instruction, modeling, rehearsal, feedback) on the missed items, then re-score.
3. Low fidelity and good data: the learner is progressing despite the deviation. Investigate before assuming the written protocol is the cause; the staff member may be making a useful adjustment that should be written into the protocol, or may be getting lucky.

Document the feedback and the re-score. A fidelity check with no follow-up is just a number. For the teaching procedures themselves, see errorless learning and prompt hierarchies and prompt fading.

Software and fidelity

Paper checklists are enough to start. If you want the scoring done and trended per item, Wilma lets you build your own checklist template and score it item by item against a session and see the result trended by item; see ABA treatment fidelity software. The checklist content, and the criterion, remain the clinician's to decide.