A small, synthetic data audit

synthetic-pilot.csv was constructed for teaching. It is not human participant data and is not intended to demonstrate a real Stroop effect.

Filter to task = stroop and block = test. Empty CSV cells represent missing values here; parse numeric and Boolean columns deliberately rather than treating every cell as a string. The corresponding experiment JSON uses null, true, and false.

Before viewing the key below, calculate the number of test trials, omissions, accuracy, and mean RT for correct responses overall and by condition. Explain what happens if you include the instruction row or incorrect trials.

Answer key

Eight test trials; two omissions. Four correct responses divided by eight trials gives 50% accuracy. Mean correct RT is (500 + 700 + 900 + 1100) / 4 = 800 ms. Each condition has 50% accuracy; congruent correct RT is 600 ms; incongruent correct RT is 1000 ms. The descriptive difference is 400 ms in these invented numbers, with no inferential significance.

If all four trials in a condition timed out, accuracy would be 0% and mean correct RT would be N/A, not 0 ms. If a condition had no trials at all, both accuracy and RT would be N/A. These are different situations.

The course has no automatic trimming threshold. A real study needs justified, documented exclusion rules and a suitable design/sample; do not select cutoffs after seeing which result they produce.