People often reject algorithms after seeing them err. I tested whether having a say at the start changes how willing they are to rely on an outside judge.
800 people judged whether each of 10 high-school students scored above the 75th percentile on a math test. They earned bonus pay for being right. Then they could keep their own call or use an outside judge’s call.
Overall, 67% chose the outside judge. The outside judges were described as 80% accurate in every condition.
67 of 100 chose the outside judge
All four conditions combined, N = 800.
The four conditions differed in the outside judge: one algorithm, three algorithms, one group of lay people or three groups. About 200 people were in each condition.
The autonomy effect reported by Dietvorst and colleagues (2018) did not replicate here. I read this as a null result, not as evidence that choice does nothing.
Those who chose the outside judge reported higher confidence than those who kept their own call.
Many people chose to use an outside judge. Two in three did, whether it was an algorithm or a group of people.
Limits. This is a conference result and has not been peer reviewed. The result is null or inconclusive, so I do not claim that choice raises or lowers use of algorithms.