The slippery slope

CalgaryGrit looks at the latest batch of federal by-elections, and observes:

Turnout was 32.4% in Vaughan, 30.8% in Winnipeg North, and 26.9% in Dauphin, for an overall average of 30%. This is comparable to the 31.1% we saw in the last round of by elections and slughtly below the average of 34.2% we've seen since 1998 - 34.2% is also the average by election turnout in the Harper era so, despite lackluster excitement in the last two rounds of by elections, there's not strong evidence of a long term decline in by election turnout.

Which is true enough... unless you change your definition of "long-term" so that it has a longer, um, term.

in fact, since the end of the Second World War — see, long-term — by-election turnout has been on a steady downward slope.

Now, there is a bit of a problem in examining election turnout generally, since the method of compiling voters lists, and thus of calculating the denominator, has varied over time as election law and practice has changed. Some of the supposed decline in turnout in the last two decades may, in fact, be attributed to phantom electors whose presence on the "permanent" voters list skews the calculation of "turnout" in the normal sense of the word, votes cast as a percentage of voters on the list.

To try and elminate this variability, consider by-election turnout as a percentage of the previous general election turnout in the same riding. If, for example, 10,000 people voted in Merasheen–Mécatina in the 1993 general election, and 9000 did in the 1995 by-election in that riding, the turnout, on this measure, is 90%.

This scattergram shows the 209 contested^(A) federal by-elections since 1945 plotted chronologically left to right, with this definition of turnout plotted vertically. The result is striking. The turnout range, and both its upper and lower limits, have sloped downwards over the years, and especially since the late 1970s.

Since 1945, there have been just 22 by-elections where the number of voters in a by-election exceeded the number of voters in the previous general election^(B). Only one of those has been since 1980. (HappyFun: Spot the outlier!)

On the bottom end of the range, there have also been 22 by-elections where this kind of turnout has been less than half that of the preceding general election. Of those, 13 have been since January 1, 2000.

Aggregated into half-decades starting in "0" and "5" years, it is apparent that the short "long-term" trend is that turnout has stopped falling, at least for now, in the first decade of the century. But that comes after three decades of relentless decline of roughly 1% per year. This graph shows the average turnout of the by-elections held in each of those five-year periods.

The figure for the half-decade starting in 2005 includes the by-elections held in 2010. The effect of the "outlier" by-election on the 2005 half-decade figure is shown in grey; without it, the most recent half-decade result would have been only 0.1% higher than the first half of the 2000s.



(A) There were also a small number of by-elections by acclamation, which are not considered here.

(B) Until 1966, there were a small number of dual-member ridings. Where by-elections were held to fill one seat in a dual-member riding, by-election turnout was calculated as a share of half the turnout of the previous general election.