On Tue, Oct 12, 2021 at 12:02 PM Kaarel Vaidla kvaidla@wikimedia.org wrote:
Additionally, we are piloting a so-called “Election Compass https://mcdc-election-compass.toolforge.org/” for this election. Click yourself through the tool and respond to the 19 statements, and you will see which candidate is closest to you!
Hi, thank you for facilitating this process and for sharing the interesting "election compass" experiment. After trying the tool, I urge you to take it offline. Its algorithm is opaque, and in my opinion very unlikely to give a helpful result. It's explicitly meant to influence how we vote, but without us having done any validation of what it's actually calculating. If you want to test this tool, you could position it as an "exit poll", to compare the tool's results with how each person actually voted, or you could turn off the "alignment" scoring.
My suspicions started with the fact that I answered "strongly support" or "support" to almost every question, which suggests that the axes were not chosen in a way that differentiates between the candidates. Instead, it seems like it's going to amplify tiny differences like "strongly" vs "support"—is this true?
Was the tool analyzed with this sort of concern in mind? Are there reasons to believe that the "alignment" scores are meaningful in our scenario?
Kind regards, Adam Wight [[mw:User:Adamw]] Writing in my volunteer capacity.