Races may be closer than they appear

Eric Grenier's by-election projections certainly make for an interesting read, and it's a plus to see at least some commentary based on voting patterns rather than conjecture. But with two of the three seats projected to be won by 40+ points, I have to wonder whether the nature of a by-election will cause problems for a model developed for general elections.

Without having crunched any numbers myself, it seems that by-elections generally tend to produce closer races - which makes sense as the parties throw everything they have into one riding, eliminating the gap in expenditures and narrowing the difference in manpower necessary for a candidate to run away from the pack. And that would lead me to expect both Dauphin-Swan River-Marquette and Winnipeg North to produce far closer races than projected, even if the projected winners sound about right absent some major change over the course of the campaign.