To study the relationship between decision similarity of group members and collective accuracy, we randomly sampled, from each dataset, 100 unique groups of n individuals (n = 3 and 9). For each group, we calculated the (i) average individual decision similarity (i.e., the average percentage agreement of all possible pairwise combinations of group members), (ii) average individual accuracy of the group members, and (iii) accuracy of the majority rule (Fig. 5, B to E, and fig. S7).

To test whether decision similarity can be used to predict high-performing groups, we performed a cross-validation procedure using a training and test set procedure. Within each dataset, we randomly drew m cases to create a training set (varying m from 0 to 60 in steps of 5) and used part of the remaining cases to form a test set using the procedure described above. Within a training set, we calculated the average decision similarity of each rater and ranked individuals according to their decision similarity. We then created groups of three raters, calculated their average decision similarity, and tested the performance of these groups in the test set using different similarity thresholds (Fig. 6 and fig. S8). For example, in Fig. 6, the top 25% corresponds to groups with the 25% highest decision similarity values in the training set. We repeated each number of training images 1000 times in each dataset and report the average values.

Note: The content above has been extracted from a research article, so it may not display correctly.

Please log in to submit your questions online.
Your question will be posted on the Bio-101 website. We will send your questions to the authors of this protocol and Bio-protocol community members who are experienced with this method. you will be informed using the email address associated with your Bio-protocol account.

We use cookies on this site to enhance your user experience. By using our website, you are agreeing to allow the storage of cookies on your computer.