Observation procedure, observer gender, and behavior valence as determinants of sampling error in a behavior assessment analogue.

The researchers set up a lab task to test how well people record behavior.

Some used a clear step-by-step plan. Others just watched and wrote notes.

They also looked at whether the observer’s gender and the type of behavior changed the scores.

The group with the plan made fewer errors.

Unsystematic watching led to shaky data.

Observer gender and whether the behavior was good or bad also shifted the numbers.

Spanoudis et al. (2011) give you the next step. They show how to calibrate observers with scripted videos until every score is within 0.1 responses per minute. Their method turns the 1980 “use a plan” idea into a day-one training tool.

Kazdin (1977) warned that drift and expectancies spoil data. Locurto et al. (1980) proved it by showing gender and valence bias in real numbers.

Abuin et al. (2026) extend the same worry to treatment fidelity. They found that 50 % fidelity can outscore 100 % fidelity unless you switch conditions fast and without cues. Both papers scream the same message: tiny procedural details move the graph.

Before you let a new tech collect data, make them watch a gold-standard video and score within 0.1. Pick reference clips that mix positive and negative behaviors so gender or likeability bias shows up early. Recalibrate each month. These two steps take twenty minutes and save you from bad calls on treatment change.