The trouble with babies and the value of bathwater: Complexities in the use of verbal reports as data.

Irvin et al. (1998) wrote a think-piece, not an experiment. They asked one question: Can we trust what people say about their own thoughts?

The authors looked at protocol analysis. That is the fancy name for asking someone to talk out loud while they solve a task. They warned readers not to treat these words as a magic window into the mind.

The paper found no new numbers. Instead it gave a rule: Treat spoken words like any other behavior. Check if the talk matches what the person actually does. Make sure your theory fits the data you collect.

Israel (1978) said the same thing twenty years earlier. That paper warned that what people say and what they do often fail to line up. Irvin et al. (1998) echo the warning but aim it at a new method.

Chou et al. (2010) moved the warning into the lab. They showed that small hints and rewards can twist what an observer writes down. Their data back up the old advice: verify your data source before you trust it.

Périkel et al. (1974) argued for cleaner stats in single-case work. Irvin et al. (1998) argue for cleaner measurement in verbal work. Both papers push the field to treat every datum, number or word, with the same hard eye.

You already take data on hits, bites, and latency. Do the same when you ask a client, "How did that feel?" Record the words. Then check if those words match what you can see and count. If they clash, trust the visible behavior and keep probing. This habit keeps treatment decisions tied to events, not to stories that may drift.