9.3 Criteria Setting, Splitting vs. Lumping & Rate of Reinforcement

9.3 Criteria Setting, Splitting vs. Lumping & Rate of Reinforcement

Quick Answer: A training criterion is the explicit, operationalized standard of performance required on an individual repetition for the learner to earn reinforcement. Successful canine instruction relies on splitting (breaking complex behaviors into micro-approximations) rather than lumping (demanding multi-parameter leaps). Trainers must manipulate the Three Ds (Distance, Duration, Distraction) by advancing only ONE D at a time while dropping the remaining parameters to baseline. Maintaining an optimal rate of reinforcement (10–15 rewards per minute during new acquisition) sustains focus and behavioral momentum. Criterion shifts are governed by the 80% Rule: advance when performance hits $\ge 80%$ (4/5 or 8/10), maintain at $60% - 70%$, and immediately lower criteria when success drops below $60%$.

---

Defining the Training Criterion

In animal learning science, a criterion (plural: criteria) represents the precise behavioral specification that must be fulfilled on a single trial for the dog to earn a reward. It serves as an unwritten contract between trainer and learner.

Ambiguous versus Operationalized Criteria

When trainers operate with ambiguous criteria, the animal experiences unpredictable reinforcement delivery, creating frustration and confusion:

• Ambiguous Criterion: "The dog stays nicely on the mat."
• Operationalized Criterion: "The dog maintains both elbow joints and both hock joints in contact with the rectangular canvas mat for 5 consecutive seconds while the handler stands 4 feet away with arms still."

If the dog meets or exceeds the criterion, the handler fires the conditioned reinforcer (clicker or verbal marker "Yes!") and delivers primary reinforcement. If the dog fails to meet the criterion, no marker is sounded, no aversive correction is delivered, and the trainer resets the dog for the next trial with adjusted criteria.

---

The Pathology of "Lumping" versus the Science of "Splitting"

The concepts of splitting and lumping are central to the work of pioneering animal trainer Bob Bailey and behavioral psychologist B.F. Skinner.

                                [ TEACHING COMPLEX BEHAVIOR ]
                                              │
         ┌────────────────────────────────────┴────────────────────────────────────┐
         ▼                                                                         ▼
    [ LUMPING ]                                                               [ SPLITTING ]
• Large, multi-parameter leaps                                            • Tiny, single-variable micro-steps
• "I bet he can do it!"                                                   • Systematic shaping ladder
• Error rate > 40%                                                        • Success rate ≥ 80%
• Reinforcement rate drops (< 3 RPM)                                      • High reinforcement rate (10-15 RPM)
• High frustration, displacement sniffing, shutdown                       • High dopamine, engagement, clear feedback

The Anatomy of "Lumping"

Lumping occurs when a trainer bundles multiple learning variables into a single trial, or advances the criteria in large, dramatic leaps rather than gradual increments. Lumping is the most pervasive mechanical error committed by novice trainers and pet owners.

Why Handlers Lump

• Human Impatience: Handlers focus on the terminal goal (e.g., a 3-minute stay during family dinner) and resist spending time on 3-second foundation steps.
• Anthropomorphism & False Attribution: "He did a 10-second stay once in the living room, so he knows what 'stay' means at the dog park!"

Behavioral Fallout of Lumping

When a dog is subjected to lumped criteria, error rates skyrocket, leading to severe behavioral fallout:

1. Extinction Bursts & Frustration Vocalization: When expected reinforcement ceases due to unattainable criteria, the dog emits intense, variable responses—barking at the handler, pawing at the treat pouch, or biting the leash.
2. Displacement Behaviors: Under elevated conflict and confusion, the dog redirects into innate displacement activities: sniffing the ground intently, lip licking, yawning, scratching, or shaking off.
3. Behavioral Shutdown / Learned Helplessness: If repeated trials produce zero reinforcement, the dog disengages completely, freezing, lying down, or walking away to avoid the frustrating interaction.
4. Superstitious Behavior Chaining: In high-error blocks, if a dog accidentally earns a reward after performing an extraneous movement (e.g., head bobbing or stepping backward) while trying to solve the puzzle, that superstitious movement becomes welded to the cue.

The Science of "Splitting": Shaping by Successive Approximations

Splitting is the systematic decomposition of a terminal behavior into minute, progressive micro-steps (approximations) along a shaping gradient. Each slice is so small that the animal has a greater than 80% probability of immediate success on the first repetition.

[!TIP] Bob Bailey's Universal Maxim: "Think, Plan, Do." Think about what you want. Plan your micro-steps and reinforcement delivery mechanics on paper before touching the animal. Do the training with robotic precision and timing.

Practical Splitting Protocol: Basket Muzzle Acclimation

A classic example of splitting versus lumping is training a dog to wear a basket muzzle happily:

• The Lumping Error: Forcing the muzzle onto the dog's face, buckling the strap behind the ears, and handing the dog a treat while the dog thrashes, paws at its face, and panics.
• The Splitting Ladder:
  • Step 1: Muzzle held at 12 inches; mark/treat any head turn or eye orientation toward muzzle.
  • Step 2: Muzzle held at 6 inches; mark/treat forward weight shift toward muzzle.
  • Step 3: Mark/treat dog's nose approaching within 1 inch of muzzle opening.
  • Step 4: Mark/treat dog's nose touching the outer rim of the muzzle.
  • Step 5: Dog places nose 0.5 inches inside muzzle opening for 0.5 seconds to lick peanut butter.
  • Step 6: Dog pushes snout halfway into muzzle for 1 second.
  • Step 7: Dog pushes snout fully into muzzle basket for 2 seconds.
  • Step 8: Snout fully seated; handler lifts straps toward ears for 1 second without buckling.
  • Step 9: Snout seated; handler buckles strap, marks, feeds 5 continuous treats through front grid, unbuckles immediately.
  • Step 10: Muzzle buckled for 10 seconds while dog walks across the room catching treats.

---

The Three Ds: Distance, Duration, Distraction (& Diversity)

Every canine behavior is subject to environmental and operational parameters known collectively as the Three Ds (expanded in modern pedagogy to include Diversity as the Fourth D).

                                   [ THE THREE Ds ]
                                          │
         ┌────────────────────────────────┼────────────────────────────────┐
         ▼                                ▼                                ▼
   [ DISTANCE ]                     [ DURATION ]                    [ DISTRACTION ]
• Space from handler             • Time position held            • Environmental sights
• Space from target/mat          • Time before terminal bridge   • Auditory movements
• Space from trigger/stimulus    • Continuous motor output       • Scent competition / food
         │                                │                                │
         └────────────────────────────────┼────────────────────────────────┘
                                          ▼
                               [ THE GOLDEN RULE ]
                Train ONE "D" at a time — Drop all others to baseline!

Definitions of the Parameters

1. Distance: Physical separation between dog and handler, or physical proximity between dog and an environmental stimulus/trigger.
2. Duration: The elapsed time that a dog maintains a stationary posture (stay, chin rest, station) or continuous motor output (loose-leash walking, heeling) before the terminal marker sounds.
3. Distraction: External environmental variables competing for the dog's attention, including auditory noises, visual motion (bouncing balls, running children), odors, and conspecifics.
4. Diversity (The Fourth D): Generalizing the behavior across diverse floor surfaces (carpet, gravel, linoleum), weather conditions (rain, snow, heat), and handler clothing/postures.

The Iron Rule of the Three Ds: Train ONE Parameter at a Time

When training a behavior, never increase more than one "D" simultaneously. Whenever you elevate one dimension, you must deliberately relax or drop the remaining dimensions back to an easy foundation baseline.

Clinical Example: Progressing a Sit-Stay

• Current Baseline: Dog reliably holds a sit-stay for 20 seconds with the handler standing 2 feet away in a quiet room with zero distractions.
• The Lumping Mistake: Handler steps 15 feet away and drops a tennis ball, expecting a 30-second stay. The dog immediately breaks and chases the ball.
• The Systematic Splitting Progression:
  • Goal: Increase Distance. Increase distance from 2 feet to 6 feet, but drop Duration from 20 seconds to 2 seconds and maintain zero distractions.
  • Once 6-foot distance is fluent at 2 seconds, gradually rebuild Duration (4s, 8s, 15s, 20s).
  • Goal: Introduce Distraction. Drop Distance back to 1 foot (stand directly beside dog) and Duration back to 2 seconds. Introduce a low-level distraction (handler gently raises right hand). Mark and reinforce stability.
  • Only when distance, duration, and distractions have been conditioned independently at high fluency should the trainer begin recombining them.

Criteria Progression Matrix: The Three Ds in Action

---

Rate of Reinforcement: Reinforcers Per Minute (RPM)

Rate of reinforcement refers to the frequency of reinforcers delivered within a standardized unit of time, typically calculated as Reinforcers Per Minute (RPM).

RPM = Total Reinforcers Delivered / Total Minutes of Active Training

The Target RPM for New Acquisition

During the acquisition phase of any new behavior, the target rate of reinforcement is 10 to 15 reinforcers per minute. This equates to delivering one reward every 4 to 6 seconds.

Why High RPM Accelerates Learning

1. Dopaminergic Activation: Frequent delivery of primary rewards produces steady pulsatile releases of dopamine in the ventral tegmental area and nucleus accumbens, creating intense motivational focus and optimistic engagement.
2. High Behavioral Momentum: A rapid delivery cadence prevents "dead time" between repetitions. When handlers pause 30 to 45 seconds between trials, dogs look around, sniff the carpet, or check out mentally.
3. Clear Informational Feedback: Learning is an information game. A rate of 12 RPM gives the dog 12 discrete data points per minute regarding which physical body positions yield reinforcement.

[!IMPORTANT] Low RPM Symptoms: If a dog begins yawning, sniffing the floor, walking away, or offering random frantic behaviors, the cause is almost universally an excessively low rate of reinforcement combined with criteria lumping. Fix: Split criteria into smaller increments to immediately restore 10–15 RPM.

---

The 80% Rule (Criteria Progression Mechanics)

To prevent subjective guesswork, trainers use the 80% Rule (often operationalized as the 4-out-of-5 Rule or 8-out-of-10 Rule) to make data-driven decisions during training sets.

                [ RUN A SET OF 5 OR 10 TRIALS ]
                               │
         ┌─────────────────────┼─────────────────────┐
         ▼                     ▼                     ▼
   [ ≥ 80% SUCCESS ]     [ 60% - 70% SUCCESS ]   [ < 60% SUCCESS ]
   (4/5 or 8/10 reps)    (3/5 or 6-7/10 reps)    (≤2/5 or ≤5/10 reps)
         │                     │                     │
         ▼                     ▼                     ▼
  RAISE CRITERION       MAINTAIN CRITERION      LOWER CRITERION
  Advance to next       Repeat set; solidify    Drop to previous step;
  micro-step on ladder  fluency and latency     split into smaller step

The Three Action Rules

1. Success Rate $\ge 80%$ (4/5 or 8/10 correct): RAISE THE CRITERION. The dog has demonstrated cognitive comprehension. Advance immediately to the next micro-step on your shaping ladder. Lingering at 100% success for extended sets is counterproductive; it builds habituation to low effort and creates resistance when criteria are finally raised.
2. Success Rate $60% - 70%$ (3/5 or 6-7/10 correct): MAINTAIN THE CRITERION. The behavior is emerging but unstable. Run another set of 5 or 10 trials at the exact same criterion to build fluency, muscle memory, and latency.
3. Success Rate $< 60%$ (2 or fewer out of 5, or $\le 5$ out of 10 correct): LOWER THE CRITERION. The task is too difficult; the trainer lumped. Do not repeat the failed criterion. Immediately drop the criterion back to the previous successful approximation, or split the gap by inserting an intermediate micro-step.

Practical Mechanics: Clean Splitting in Precision Shaping

When shaping an animal through successive approximations, trainers document each concrete milestone in advance to avoid criteria creep. By recording pass/fail percentages across each 10-trial block, the practitioner ensures criteria are elevated strictly on objective performance data rather than optimistic intuition. If the dog shows early conflict signs such as delayed response latency or tentative weight shifts, the trainer immediately treats this as an operational cue to split the current step into two smaller, highly achievable micro-slices.