Classroom Observation — Systematic Classroom Observation

Structured observation uses a predefined coding instrument — a fixed set of categories into which the observer places each observed event (e.g., teacher asks a question = category 4 in FIAC). It supports reliable quantification and cross-site comparison but only captures what the instrument was designed to see. Unstructured (ethnographic) observation uses open-ended field notes to capture the full complexity of the social scene; it is more flexible but harder to compare systematically across classrooms. Semi-structured approaches combine a guiding framework with space for open notes.

Research on the reliability of observation-based teaching measures suggests that a minimum of two to four observation sessions per teacher, each lasting 30–60 minutes, is typically needed to obtain a stable estimate of instructional practice. Fewer sessions produce estimates with wide confidence intervals that may not reliably differentiate teachers or track change over time.

Train all observers on the coding protocol using anchor videos that have been pre-coded by the instrument developers or a gold-standard team. Conduct reliability checks by having two observers independently code the same session and computing agreement statistics — Cohen's kappa for categorical coding or intraclass correlation coefficient (ICC) for rating scales. An ICC or kappa above 0.70 is the commonly accepted threshold before field data collection begins.

High-stakes teacher evaluation using observation scores requires very high reliability and multiple raters or multiple observations. Research — including the Gates Foundation MET project — has shown that single observations have limited reliability and predictive validity for student outcomes. When used for personnel decisions, observation must be combined with other evidence sources (student achievement data, portfolios) and collected under carefully standardized conditions.

Both have advantages. Video allows repeated viewing, retrospective coding by multiple raters, and archiving for later analysis or professional development use. However, cameras can themselves cause behavioral change, raise privacy concerns, and may not capture the full sensory environment of the classroom. In-person observation allows real-time coding and contextual awareness the camera may miss. Many rigorous studies use both.