[Ottawa – April 27, 2025] We submit our final poll for the 2025 federal election. We are selecting our 3-day rolling IVR poll as our choice to predict the final vote outcome. Our chosen final poll …
I don’t trust any data from a company that uses basis spline or hobby spline smoothing to represent data without showing the original points and a confidence interval.
It’s an incredibly amateur move to display data this way. All the popular chart libraries include it because people like it, but objectively those people are wrong.
I can rant for hours about this, but the smoothing on their charts actually shows values outside the real range from data. That’s okay if you use a smoothing model like loess with confidence intervals, but it is not okay as a “make it pretty” visualization technique.
These spline smoothed data viz do not show any real trends between the points. The points can force the curve to make sharp jumps that aren’t real during sudden changes or changes in spacing of points. It’s more evident when the lines go below 0, but we don’t have that here.
These charts do not replicate, as new data is added it actually changes the curve, which means you’re not plotting polling results over time anymore. The points are simply anchors on the curve, if you use illustrator for example draw a bunch of points, then hit the smooth button, that’s what’s happening here.
These charts can have curves that go steeper than 90 degrees and actually loop backwards! We don’t live in the Jeremy Bearimy timeline, that can’t happen here. And it did with EKOS charts from a few months ago when they changed the time step of polling.
I don’t trust any data from a company that uses basis spline or hobby spline smoothing to represent data without showing the original points and a confidence interval.
It’s an incredibly amateur move to display data this way. All the popular chart libraries include it because people like it, but objectively those people are wrong.
I can rant for hours about this, but the smoothing on their charts actually shows values outside the real range from data. That’s okay if you use a smoothing model like loess with confidence intervals, but it is not okay as a “make it pretty” visualization technique.
These spline smoothed data viz do not show any real trends between the points. The points can force the curve to make sharp jumps that aren’t real during sudden changes or changes in spacing of points. It’s more evident when the lines go below 0, but we don’t have that here.
These charts do not replicate, as new data is added it actually changes the curve, which means you’re not plotting polling results over time anymore. The points are simply anchors on the curve, if you use illustrator for example draw a bunch of points, then hit the smooth button, that’s what’s happening here.
These charts can have curves that go steeper than 90 degrees and actually loop backwards! We don’t live in the Jeremy Bearimy timeline, that can’t happen here. And it did with EKOS charts from a few months ago when they changed the time step of polling.