Cool your jets. I’m not deriding 338 or the usefulness of polling in general. Modelling is a very difficult science and 338 are very, very good at it.
By their own numbers their likelihood of correctly projecting a seat goes down dramatically when one parties odds of a win aren’t higher than 70%. Translated backwards that means that the projected riding level poll numbers you see become significantly less meaningful.
What this means is that while models like 338 are very good at predicting the overall trend of the election, in close or complicated races like the one being discussed they are not as indicative as people would like to think when it comes to, say, making tactical voting decisions. They are, ultimately, just making somewhat informed guesses about what these numbers are. In the aggregate those guesses tend to be right, but that doesn’t mean you’re getting accurate and direct data about how your particular riding is leaning, and people need to understand that.
What this means is that while models like 338 are very good at predicting the overall trend of the election, in close or complicated races like the one being discussed they are not as indicative as people would like to think when it comes to, say, making tactical voting decisions. They are, ultimately, just making somewhat informed guesses about what these numbers are. In the aggregate those guesses tend to be right, but that doesn’t mean you’re getting accurate and direct data about how your particular riding is leaning, and people need to understand that.