Technology and Decisions

Overconfidence effect

Having too much confidence in the accuracy of your judgement, even when facts suggest otherwise

Dana et al

A: role that interviews play in decision making

P: University students/given information on other students’ course selection and past GPA/told that past GPA is the best predictor of future GPA/asked to predict students’ future GPA/participants met and interviewed students for GPA predictions or just make predictions based on course selection and past GPA alone/in half the interviews, students answered honestly/in the other half, they were only allowed to ask yes or no questions and students responded dishonestly

F: students made more accurate predictions for the students they didn’t interview/interviews were counterproductive/participants realised when the students were dishonest

C: findings support the overconfidence effect/pariticipants were confident that they could get an accurate impression from an interview and be able to predict future GPA/participants made more accurate predictions if they did not interview the student

Evaluate Dana et al

• Strong experimental design: established a casual relationship between whether or not the interviews took place and the accuracy of the future GPA predictions

• All university students who did not have experience in interviewing

• Suggests the limits of human judgement: predictions based on data alone are more accurate

Hoffman et al

A: compare hiring decisions of human managers with computer algorithms

P: research carried out across 15 businesses who employ low-skilled service workers/high worker turnover/computer algorithm used to predict job performance of 300,000 applicants, based on questions about skills and personality/three categories/1. green (high potential for success 2. yellow (medium) 3. red (low)/ hiring managers were allowed to overrule algorithm and hire from the yellow red categories if they disagreed

F: algorithms predictions were statistically significant/ green stayed an average of 12 days longer than yellow who stayed an average of 17 days longer than red/median length of job stay was 99 days/when hiring manager overruled algorithm, yellow candidate still ended up staying for 8% less time than green

C: computer algorithms can make accurate predictions about employees staying based on data/human intuition is counterproductive/judgement was worse than if algorithms decisions were followed

Evaluate Hoffman et al

• High ecological validity: thousands of applicants from real businesses

• Only low skilled service workers: low generalisability to other types of jobs

• Only measured how long, not job performance: more difficult to quantify and accurately predict