Thursday, May 29, 2014

Swing-o-meter

I've been playing around with the idea of a swingometer for the upcoming provincial election. The efficacy of this is complicated by the high poll numbers for the NDP.

The concept has been used in the UK for half a century.

A simple explanation is this: they line up the ridings in order of strongest for major party A to weakest for major party A (and therefore strongest for major party B). They then compute the swing required for each seat to flip to the other party (half of the margin between the parties +1). They can therefore determine how many seats each party should win based on a particular swing in the popular vote.

Based on my seat model, the last election under the new boundaries would have resulted in 38 PCs and 11 Liberals. The PCs won the popular vote by 14.4 percentage points, and therefore a swing between the parties of 7.2 points in favour of the Liberals would result in a tie of the popular vote.

According to my calculations however, it would actually take a swing of 7.7 percentage points for the Liberals to win a majority government by taking Moncton South as the 25th and decisive seat.

A swing to the PCs of over 11.6 percentage points would be required for them to sweep every seat in the legislature. But a swing to the PCs of just over 1.7 percentage points would reduce the Liberals to 5 seats - showing just how close the last election was to a near wipe-out for the Liberals.

Conversely, the Liberals would need a swing of over 23.3 percentage points to take all of the seats.

The above is blind to the NDP. By design, the swingometer is a two-dimensional creature and can compare only two parties. However, the strength of the NDP changes the ordering slightly even if they're ignored for swingometer purposes.

This is what the swingometer looks like if we use the last CRA poll as a base. That poll had the Liberals at 43% and the PCs at 31% (the NDP was at 21%). That would indicate a result moving from PCs +14 at the 2010 election to PCs -12, a change of 26 percentage points and a swing of 13 percentage points. Using these numbers as a base, we would expect an election to yield 39 Liberals, 8 PCs and 2 NDPers.



As you can see, both of the NDP seats come at the expense of what would otherwise have been Liberal seats with this big of a swing. However, if we look at them in terms of road to majority government, the NDP is doing equal damage to both parties: taking an otherwise safe Liberal seat (Saint John Harbour) and an otherwise leans PC seat (Fredericton West-Hanwell).

In order to recover to a majority government from this poll number, the PCs need a swing of over 5.7 percentage points to win Southwest Miramichi-Bay du Vin and take 25 seats.

Wednesday, May 14, 2014

What are New Brunswick's tipping point ridings?

Nate Silver did some interesting work in the last U.S. presidential election on tipping point states. He was looking to find the state that would be most likely to deliver the deciding electoral vote to win the presidency, regardless of the total margin. It is not unlike the concept of a winning goal in a hockey game; in a 4-0 game, the first goal is decisive; in a 5-4 game it is the winning team's fifth goal that matters most.

To win a majority government in the 2014 New Brunswick election, a party will need to take 25 seats. The question: which seat is most likely to be that 25th seat?

I am not as sophisticated as Nate Silver. My analysis was relatively simple. I took all of the ridings and sorted them by the expected PC margin of victory/defeat and then did the same for the Liberals and for the NDP. This ranks the ridings from strongest to weakest for each party. In theory, the 25th strongest seat for the PCs would be their tipping point. I did this with two sets of data, one a simple average of results across the past several elections along the new boundaries, and a second with this data adjusted for incumbency and leadership factors as described here.

Because this data is imprecise and local factors may skew close ridings in one direction or the other, I initially focused on a range of 10 tipping point ridings. Some ridings showed up on all four lists, while others were wishy-washy as to whether or not they could be the tipping point.

A total of 7 ridings fall on all four of the lists between the two major parties, these being the PC and Liberal adjusted and non-adjusted lists of seats in the range of 20-30th best. According to the data, it is in these 7 ridings that the election is most likely to be decided:
  • Saint John Portland
  • Kent South
  • Moncton South
  • Quispamsis
  • Southwest Miramichi-Bay du Vin
  • Memramcook-Tantramar
  • Miramichi Bay-Neguac
The party which wins all or most of these seats is very likely to win the election.  If there is a 4-3 split in these seats, it will be a long election night.

Other contenders for PC-Liberal tipping point seats are:
  • Rothesay (appeared on 3 lists)
  • Fredericton West-Hanwell (appeared on 3 lists)
  • Moncton East (appeared on 2 lists)
  • Charlotte-Campobello (appeared on 2 lists
  • Saint John Lancaster (appeared on 1 list)
  • Carleton-Victoria (appeared on 1 list)
If the Tories or Liberals win in a landslide, all or most of these ridings will go for them. In a close election, they will be decisive. Back to the hockey game analogy, we're looking for the winning goal. In a 4-0 game, we might not pay that much attention to the first goal, but it was the most important. In a 5-4 game, that winning goal will be all we talk about.

Based on the reality of the past several years some of these seats seem misplaced, but that is less so the case when you consider the following fact. Something occurred to me while I was working on this list that I guess I knew but which I hadn't quite realized. The Liberals have not won the popular vote in an election in New Brunswick since 1995.

Ridings like Saint John Portland and Rothesay which are held by senior cabinet ministers and have been PC since 1999 would not seem on their face to be tipping point seats. But they were both won very narrowly in the close elections of 2003 and 2006. It is no surprise to see Quispamsis on this list, it (and its predecssor Kings West) has a perfect record of voting with the winning party going back to the establishment of single member ridings in 1974.

What about the NDP?

It is unlikely that the NDP will win the election. Really, they are looking to ensure they win seats in the legislature for the first time since the 2003 election. So, I've looked at their first 10 seats.

Nine seats show up on both the adjusted and non-adjusted list for the NDP:
  • Fredericton West-Hanwell
  • Saint John Harbour
  • Hampton
  • Fredericton South
  • Tracadie-Sheila
  • Saint John East
  • Oromocto-Lincoln
  • Saint John Portland
  • Kings Centre
The contenders for the NDP's tenth best seat are Miramichi Bay-Neguac and Moncton East.