Project 3 Measurement

Sarah Cowie introduces herself and the guests, two real-life behavior analysts.
Topic of discussion: Measurement and issues in measurement within applied behavior analysis (ABA).

Automatic recording of behavior in a controlled setting.
- Example: When a pigeon interacts with a key, it activates a microswitch that logs the event onto a computer.
Ease of recording discrete behaviors is facilitated by technology in the lab setting.
Behaviors in real-world settings are more complex and difficult to measure.

Katrina introduces the topic of measurement issues in applied behavior analysis.
Discussion on the potential for automation of recording behaviors in applied settings:
- Examples:
- Counters for people entering/leaving through doors.
- Traffic flow measurements on roads.
Historically, socially significant behaviors have relied on direct observation methods.
- Direct observation involves human observers noting specific behaviors.
- Traditional recording methods: pen and paper, tallying, sampling methods.

Impact of technology on measurement methods in ABA:
- Shift from pen and paper to electronic reporting.
- Electronic methods expedite data analysis but retain labor-intensive observation needs.
Challenges faced in measuring behaviors:
- Requires constant presence of observer during behavior occurrence.
- Time-intensive training process for observers to ensure accurate behavior recording.
- Human observers may deviate from set definitions and criteria in measurement.
- Excitement for technology as a means to reduce variability in human observation.

Javier explains computer vision and its role in behavior measurement.
Challenges with applying computer vision in real-world scenarios:
- Recording topographically diverse behaviors in expansive settings is complex.
- Example of complexity: Assessing whether a task was performed correctly during an academic behavior observation.

Mention of project initiated due to COVID-19 focusing on face touching as a health behavior.
- Face touching as a behavior: topographically diverse and identifiable by computer vision.
- Initial project involved analyzing 600 hours of video with human observers, was time-consuming and challenging.
- Decision to develop computer code for automatic detection of face touching behaviors.
Process of creating the computer vision code:
- Feeding the algorithm with examples of behavior and non-behavior events (e.g., face touching).
Performance of the computer vision system:
- Accuracy rate above 95% when tailored to specific faces for the study.
- Limitations: Need for specific footage to develop accurate detection algorithms.
Use of freeware and libraries to enhance computer vision algorithms.

Riveting class project leveraging the developed code for face touching detection.
Aim of the class project: Provide a diverse array of face touching events for the system to analyze.
- Activities to include scripted examples of face touching events (e.g., touching lips, nose, eyes, forehead).
Duration and structure of video data collection:
- Participants will record themselves for five minutes, engaging in specified behaviors every ten seconds, totaling 30 events.
Inclusion of near-target events and nonevents for accuracy testing (e.g., touching hair, near touching face).
Collection of diverse inputs essential for enhancing the code's sensitivity and specificity.
- Results will illustrate if the system can accurately detect behaviors as intended.

Reflections on the nature of data collection in research versus applied settings.
Importance of human observation as a standard for accuracy, though it may remain fallible due to human error (e.g., observer drift).
- Machine learning systems are designed to avoid human-like observer drift but can become overinclusive with excessive data input.
- Balance needed to ensure the technology remains functional without introducing excessive false positives.
Some socially meaningful behaviors may soon become measurable through advanced technology (e.g., nail biting).
Acknowledgment of ongoing need for human observers in behavioral analysis, despite advancements in technology.
Emphasis on the potential for technology to streamline and enhance measurement processes in future ABA research and application.

Acknowledgment of significant strides in measurement technology over the years.
Anticipation of a future where behavioral measurement may resemble an automated operant conditioning chamber.
Closing remarks: Excitement about sharing the project’s prospects and results with the class.