Table 1 Summary of Experimental Findings.

From: Humans rely more on algorithms than social influence as a task becomes more difficult

Claim Experimental support
1 2 3
Subjects will rely more on algorithmic advice than equally good human advice Yes Partial* Yes
Subjects receiving algorithmic advice will rely more on advice in difficult questions Yes Yes Yes
Highly accurate subjects will rely more on algorithmic advice than inaccurate subjects** No No No
Bad advice from an algorithm more strongly reduces weight on advice than bad advice from a crowd N/A N/A Yes
  1. *supported in baseline model without interactions and controls.
  2. **This hypothesis was preregistered only for Experiment 3. When using an alternative measure of accuracy that allowed for significant outliers, we observed a positive and significant effect of the interaction between accuracy and algorithmic advice in both Experiment 1 and 2, so we preregistered a hypothesis about this effect for Experiment 3. We then observed that the observed post-hoc effect in Experiment 1 and 2 were due to outliers. We thus changed the accuracy measure to be the percentile rank of accuracy on a question, to prevent outliers from strongly influencing results.