MITI Coding Manual 4.2.1

• MITI 4.2.1 is a standardized framework that assesses clinician adherence to motivational interviewing principles through both global dimensions and behavioral codes.
• It operationalizes metrics such as technical and relational global scores, reflection-to-question ratios, and MI-adherent ratios for rigorous research and training feedback.
• The manual guides human coder calibration and supports automated coding systems, enhancing reliability and facilitating advancements in MI practice evaluation.

The Motivational Interviewing Treatment Integrity (MITI) Coding Manual 4.2.1 is a structured framework for the assessment of clinician fidelity to Motivational Interviewing (MI) principles in counseling practice and research. MITI 4.2.1 provides operationalized global ratings and discrete behavioral codes, defining both what constitutes MI-consistent technique and how proficiency should be quantified for both research validation and competency feedback.

1. Structure and Purpose of MITI 4.2.1

MITI 4.2.1 standardizes the evaluation of MI practice, enabling reliable assessment of both technical and relational therapist skills. Its primary goals are: (1) to facilitate systematic research on MI efficacy and mechanisms, (2) to guide training and supervision by providing actionable feedback, and (3) to serve as a quantitative ground truth for automated skill assessment platforms. The manual operationalizes core MI constructs by specifying global dimension ratings and granular behavior codes to be applied to transcript segments, typically full counseling sessions or multi-turn dialogues (Hu et al., 17 Dec 2025, Kiuchi et al., 28 Jun 2025, Flemotomos et al., 2021).

2. Global Dimensions and Rating Anchors

The core of MITI 4.2.1 is its set of four global dimensions, each rated on a 1–5 Likert-type scale (half-point increments allowed in some studies):

• Cultivating Change Talk: Degree to which the clinician actively encourages client statements in favor of change.
• Softening Sustain Talk: Degree to which the clinician reduces, redirects, or avoids reinforcing client discourse in favor of maintaining the status quo.
• Partnership: Extent of collaboration and power-sharing between clinician and client.
• Empathy: Depth, accuracy, and consistency of the clinician's understanding of the client's explicit and implicit perspectives.

Anchor statements for each score precisely define behavioral expectations at each scale point. For example, a score of 5 on Cultivating Change Talk requires "marked and consistent effort to increase the depth, strength, or momentum of the client’s language in favor of change." Partnership and Empathy dimensions operationalize MI’s underlying relational principles (Kiuchi et al., 28 Jun 2025).

Several studies also supplement these globals with composite metrics:

• Technical Global = $$(\text{Cultivating Change Talk} + \text{Softening Sustain Talk}) / 2$$
• Relational Global = $(\text{Partnership} + \text{Empathy}) / 2$ (Hu et al., 17 Dec 2025)

Some research introduces custom overall gestalt scores (e.g., "Overall Evaluation" paralleling MI spirit and acceptance), but these are not standard to MITI (Kiuchi et al., 28 Jun 2025).

3. Behavioral Coding and Quantitative Metrics

MITI 4.2.1 prescribes the coding of discrete counselor behaviors at the utterance level, each corresponding to MI-consistent (adherent), MI-inconsistent (non-adherent), or neutral acts. Standard codes include:

• Giving Information
• Persuading with Permission
• Asking Questions
• Simple Reflections
• Complex Reflections
• Affirming
• Seeking Collaboration
• Emphasizing Autonomy
• Persuading (Non-Adherent)
• Confronting (Non-Adherent)

Coders tally the frequency of each target behavior within sessions. MITI defines several derived ratios central to both research evaluation and therapist feedback:

Metric	Formula (LaTeX)	Derived Components
Complex Reflections Ratio	$$\dfrac{\#\,\text{Complex Reflections}}{\#\,\text{Simple Reflections} + \#\,\text{Complex Reflections}}$$	Reflection subtype counts
Reflection-to-Question Ratio	$$\dfrac{\#\,\text{Simple} + \#\,\text{Complex Reflections}}{\#\,\text{Total Questions}}$$	Reflection and question totals
Total MI-Adherent Ratio	$$\dfrac{\text{Seeking Collaboration} + \text{Affirming} + \text{Emphasizing Autonomy}}{\text{Seeking Collaboration} + \text{Affirming} + \text{Emphasizing Autonomy} + \text{Persuading} + \text{Confronting}}$$	MI-adherent and non-adherent behaviors

These composite metrics enable summative judgments of technical skill (e.g., favoring reflection over questioning, privileging complex reflections) and adherence to MI spirit (e.g., collaboration, autonomy support) (Hu et al., 17 Dec 2025, Flemotomos et al., 2021). Not all studies implement the full set of behavior codes or derived ratios (Kiuchi et al., 28 Jun 2025).

4. Implementation in Human and Automated Coding

Human application of MITI 4.2.1 typically involves coders (frequently graduate-level or experienced clinicians) working independently and blind to the origin of transcripts (e.g., real vs. AI-simulated dialogues). Coding units are entire multi-turn dialogues, with all counselor statements coded for behavior counts and global ratings administered post hoc for the entire session (Hu et al., 17 Dec 2025).

Automated systems, as described in "Automated Evaluation Of Psychotherapy Skills Using Speech And Language Technologies," adapt MITI/MISC coding to a speech processing pipeline:

1. Voice Activity Detection
2. Speaker Diarization
3. Automatic Speech Recognition (ASR)
4. Speaker Role Recognition
5. Utterance Segmentation
6. Behavior Coding via neural classifiers (e.g., BiLSTM with attention) Session-level metrics (e.g., Reflection-to-Question Ratio, MI-Adherent %) are then computed over labeled utterances (Flemotomos et al., 2021).

A plausible implication is that full automation of MITI coding, while feasible for major codes and composite metrics, remains less reliable for infrequent behaviors and in the presence of upstream ASR/diarization error. Integration of additional multimodal or dialog-context signals is an active development area.

5. Coding Protocol, Rater Training, and Reliability

Human coding protocols under MITI 4.2.1 mandate coders work independently and, ideally, undergo formal calibration sessions and reliability assessment. Coders may be supervised by an experienced MI clinician. In some applications, no detailed calibration/training or inter-rater adjudication is implemented or described beyond summary supervision (Hu et al., 17 Dec 2025, Kiuchi et al., 28 Jun 2025).

Reliability is typically benchmarked using statistics such as interclass correlations (ICC), with thresholds of <0.50 (poor), 0.50–0.75 (moderate), 0.75–0.90 (good), >0.90 (excellent). Reported ICCs for MITI global dimensions vary by rater population and context—for example, Partnership ICCs in one human-coded study reached "excellent" (0.99 for $k$-raters), while Empathy and Cultivating Change Talk exhibited moderate to good reliability (Kiuchi et al., 28 Jun 2025).

It is important to note that some published studies omit reliability reporting altogether, limiting interpretability of their MITI coding results (Hu et al., 17 Dec 2025).

6. Research and Applications Using MITI 4.2.1

MITI 4.2.1 is foundational in contemporary research evaluating both human-delivered and AI-simulated MI. It is used to:

• Benchmark and fine-tune LLMs for MI-consistent behavior, both in Chinese-language settings (Hu et al., 17 Dec 2025) and multi-lingual clinical simulations (Flemotomos et al., 2021).
• Set targets for technical and relational skill acquisition in therapist training and supervision.
• Enable large-scale, expert-validated comparison of automated counselor systems, establishing performance baselines across global dimensions (Kiuchi et al., 28 Jun 2025).

Recent extensions include the evaluation of non-human counselors (e.g., AI agent dialog) and adaptation for automated speech and text analysis systems. The pipeline architecture in automated evaluation research integrates MITI coding at multiple processing stages, from utterance labeling to session-level metrics and advanced summary scores encompassing empathy, spirit, and MI-adherence proportions (Flemotomos et al., 2021).

7. Limitations and Ongoing Developments

Key limitations in current MITI 4.2.1 application include:

• Lack of detailed anchor documentation and rater-training material in some studies, hindering reproducibility (Hu et al., 17 Dec 2025).
• Variable implementation: not all research applies the full set of behavior codes/ratios, and some rely solely on global ratings with no summary metric computation (Kiuchi et al., 28 Jun 2025).
• Incomplete reporting of example-coded transcripts and detailed reliability calculations.
• Automated coding pipelines remain less accurate for rare behaviors and are sensitive to propagation of upstream segmentation or recognition errors (Flemotomos et al., 2021).

A plausible implication is that future directions will emphasize more robust, context-aware, and multimodal behavioral classification systems aligned more closely with the full richness of the MITI manual, as well as standardized reporting and open sharing of anchor sets and coder training protocols.

---

References:

• (Hu et al., 17 Dec 2025) Toward expert-level motivational interviewing for health behavior improvement with LLMs
• (Kiuchi et al., 28 Jun 2025) Evaluating AI Counseling in Japanese: Counselor, Client, and Evaluator Roles Assessed by Motivational Interviewing Criteria
• (Flemotomos et al., 2021) Automated Evaluation Of Psychotherapy Skills Using Speech And Language Technologies