Calibration of observational measurement of rate of responding.

The team made short videos with a known number of hand raises per minute. They asked observers to score the clips with a computer key that counted each response. Then they checked how close each observer came to the true rate.

The goal was to see if you can train people to read behavior within half a response per minute.

Most observers landed within 0.4 responses per minute of the real count. Half of them were even tighter, within 0.1. That level of error is small enough for most rate-based decisions you make in clinic or classroom.

Computerized continuous recording beat older stop-watch sampling every time.

Stolz (1977) looked at every JABA paper from 1968-1975 and saw that fewer than 25 percent checked reliability in each condition. Spanoudis et al. (2011) move past that warning by giving you a scripted yard-stick instead of just two people agreeing.

Kazdin (1977) showed that observer drift and expectancies can inflate simple agreement. Calibration against a fixed video removes those biases because the right answer is set in stone.

Locurto et al. (1980) already proved that systematic beats casual sampling. The new study adds a digital tool that reaches the same goal with tighter numbers.

If you run rate-based interventions like differential reinforcement of high rates, you need clean counts. Calibrate new staff with a two-minute reference clip before they collect real data. You will catch drift early and keep your graphs trustworthy.