Stick To Your Role! Leaderboard
Structure
This image shows the circular value structure projected on a 2D plane. This was done by computing the intercorrelations between different values this space was then reduces with a SVD based approach and varimax rotation (`FactorAnalysis` object from `scikit-learn`). The theoretical order is shown in the top left figure. The distance is computed as the average distance of each value to it's rank in the theoretical order. The minimal distance with the theoretical order in the clockwise and counter-clockwise direction was taken as the final distance.
 }})
Pairwise Rank-Order stability
This image shows the Rank-Order stability between each pair of context chunks. Rank-Order stability is computed by ordering the personas based on their expression of some value, and then computing the correlation between their orders in two different context chunks. The stability estimates for the ten values are then averaged to get the final Rank-Order stability measure. Refer to our paper for details.
 }})
Confirmatory Factor Analysis metrics
This tables show the metrics resulting from the Magnifying class CFA procedure: for each context chunk four CFA models are fit (one for each high level value). The average of the metrics for those four CFA models are shown for each context chunk.
Visualizing the order of simulated personas
This image shows the order of personas in each context chunk for each value. For each value (row), the personas are ordered on the x-axis by their expression of this value in the `no_conv` setting (gray). Therefore, the Rank-Order stability between the `no_conv` chunk and some chunk corresponds to the extent to which the curve is increasing in that chunk.
 }})