Claims of harassment, public disavowals by influencers, and fractured campaign management threaten to further derail the NDP’s performance with just weeks to go — all while insiders tell the story of a disorganized campaign failing to communicate with candidates.
SmartVoting isn’t simply data, however. They have a “proprietary algorithm” that they refuse to share.
Not only that, but if the true goal of the site is strategic voting to keep Conservatives out of office, then you’d think that in a tied race between an NDP and Liberal candidate, with the Cons a distant third, there wouldn’t be a strategic vote recommendation at all.
And yet, in a riding in that exact situation, with a NDP incumbent tied with a Liberal candidate yet to be named, it was recommending people vote for the Liberal up until a day or two ago (Hamilton Centre, Matthew Green’s riding).
Then you start digging a little deeper and realize that the SmartVoting site is mostly (but not entirely - they have their secret algorithm!) based on the projections from 338. Which are reasonably accurate overall. However, in the recent Ontario election they under-projected the NDP’s seat count significantly, projecting 16 seats, whereas they actually ended up winning 27.
I’m not a statistician, but that seems like a remarkable jump. If I had to guess, it’s that 338’s model is not capturing something about the NDP or their voter base, whether that’s the party’s ground game, incumbency, or some other hard-to-quantify aspect.
So I mean, if it were just data then sure, don’t shoot the messenger! But the problem is more that the modeling and recommendations seem like they have real flaws, probably unintentional, but flaws nonetheless. And projections and recommendations can and will influence voter behaviour.
SmartVoting isn’t simply data, however. They have a “proprietary algorithm” that they refuse to share.
Not only that, but if the true goal of the site is strategic voting to keep Conservatives out of office, then you’d think that in a tied race between an NDP and Liberal candidate, with the Cons a distant third, there wouldn’t be a strategic vote recommendation at all.
And yet, in a riding in that exact situation, with a NDP incumbent tied with a Liberal candidate yet to be named, it was recommending people vote for the Liberal up until a day or two ago (Hamilton Centre, Matthew Green’s riding).
Then you start digging a little deeper and realize that the SmartVoting site is mostly (but not entirely - they have their secret algorithm!) based on the projections from 338. Which are reasonably accurate overall. However, in the recent Ontario election they under-projected the NDP’s seat count significantly, projecting 16 seats, whereas they actually ended up winning 27.
I’m not a statistician, but that seems like a remarkable jump. If I had to guess, it’s that 338’s model is not capturing something about the NDP or their voter base, whether that’s the party’s ground game, incumbency, or some other hard-to-quantify aspect.
So I mean, if it were just data then sure, don’t shoot the messenger! But the problem is more that the modeling and recommendations seem like they have real flaws, probably unintentional, but flaws nonetheless. And projections and recommendations can and will influence voter behaviour.