Which groups each system sounds like
How closely each system's default answers match each group's real survey answers. For each question we compare the share of a system's answers on each option with the shares for the group's real respondents: match = 100 × (1 − Jensen-Shannon distance), averaged over the questions the group was asked. 100 means identical shares. No group is named in the prompt.
Which end of each scale systems sound like
Every system, side by side
Every group
Show as a table
Telling a system which group to answer as
Method and caveats
Data. Group answers come from the same surveys as the benchmark: Pew's American Trends Panel (OpinionsQA, SubPOP) and Pew Global Attitudes and the World Values Survey (GlobalOpinionQA). OpinionsQA groups are unweighted respondent counts, like the overall answers the benchmark uses. A group is left out of a question if fewer than 30 of its members answered it.
Intervals. 95% bootstrap intervals from resampling questions ( resamples). A difference whose interval includes 0 could be noise.
Agreement within a group. Groups whose members agree with each other are easier to match by giving the most common answer. The majority-answer reference shows how much of a gap that explains.
Countries. GlobalOpinionQA asks each country a different set of questions, so country differences partly reflect which questions were asked.