Effects of feedback on performance and response latencies in untimed reasoning tests

Jens F. Beckmann and Nadin Beckmann ran a controlled study to find out what happens when test-takers are told, item by item, whether they got an answer right or wrong. The question matters because many assumptions about how people behave during ability tests rest on untested ideas about confidence, anxiety, and the time a person spends on each problem.

The sample came to 120 students in the eighth and ninth grades. Each was given a set of 12 number series problems in an open-answer format, meaning they had to generate the solution rather than pick from a list. Assignment to one of two conditions was random. In the standard condition students worked through the problems and received nothing back. In the feedback condition they were told after each item whether their response was correct or incorrect.

Before and after the test, the researchers measured two things that students carry with them into any assessment: task-related self-confidence, and worry. The point was to see whether feedback shifted these states, and whether any shift showed up in how students performed or in how long they lingered over each question.

The results did not flatter the idea that correct/incorrect feedback helps. Simple right-or-wrong signals gave examinees little useful information to work with. What the feedback mostly did was raise worry. And higher worry pulled performance down rather than up, so the students who were told how they were doing tended to do worse than those left alone.

Response latencies told a quieter story. Providing feedback produced no systematic change in how test-takers managed their time. Students in the feedback condition did not consistently speed up or slow down relative to the others.

That last finding speaks to the I greater than C phenomenon and the broader claim that time behavior in untimed tests is shaped, at least in part, by things other than reasoning ability. If self-confidence and worry really steered how long people spend on problems, manipulating those states through feedback should have moved the timing data. It did not. The evidence here offers no backing for treating non-intellectual variables as drivers of time behavior in untimed performance tests.