what is the right benchmark for caring conformist
Problem
What is the right benchmark to test whether the caring conformist type really exists and is not just a random fluke?
Initial consideration
- Caring type behavior should not be observed at all. Correct benchmark is zero.
- However, people make errors in experiment. What is the percentage of errors that we expect
First Suggestion: 50% benchmark
- assumption: behavior among conformists is completely random
- then, we expect 50%, but we get 63% (87 observations of which 55 are caring)
- although this test is conservative, it is significant at 1% level
Rado is unhappy with this solution because
- it is "too" conservative.
- it creates illusion that 50% of the found 63% could be attributed to error
Second Suggestion: inconsistency on first tow choices
- take switches between first individual decision and vote as benchmark
- technically no one should switch here, still around 12% do
- problem with this approach: what is the right test
further ideas
What is the benchmark in the ECNTR paper?
Edited by Wendelin Schnedler