Saturday, September 15, 2012

History of the ElectoMatic and what I'm doing

Two seemingly unrelated topics but I will explain how they are one. First I want to explain the history of the ElectoMatic.

I've told you before about the game I play, Politics Canada. I founded this game in late 2003, and even that is a bit of a spin on events. Politics Canada had actually been alive before I ever found it, and, I did not even "find" it myself. What I found was a dead forum and a game that was inoperative. How I found it was though a similar game called Politics UK, or, PolUK. PolUK was founded in 2001 and had created various other games like PolUSA, PolAus, PolNZ and so forth. When I stumbled across PolCan it was a dead game that I revived myself.

One of the first problems was how to simulate elections. As admins, we would read all the election advertisements made by players and from that deduce a popular vote score for each province. How to turn this into seats though? After some research and much experimentation I came up with the formula of X^2 I will explain how it works.

Lets use the Quebec election as our example. The Liberals took 31.20% the PLQ 31.96%, the CAQ 27.06% and the QS 6.03%. First, we need to get rid of the decimal and get rid of the % sign. This puts the Liberals at 3120, the PQ at 3196, the CAQ at 2706, and the QS at 603

Now, we square it.

The Liberals end up with 9,734,400 the PQ at 10,214,416 the CAQ with 7,322,436 and the QS at 363,609. Now we total these numbers, and use these numbers to determine the share of seats. In this case, 35.23% for the Liberals, 36.96% for the PQ, 26.50% for the CAQ, and 1.31% for the QS. The simplest way to turn this into seat is to multiply this by 125. 44.0 for the PLQ, 46.2 for the PQ, 33.1 for the CAQ, and 1.6 for the QS. All get rounded down except for QS which gets rounded up. The final result is:


Not quite exactly the real end-results.

We can also throw other provinces into the mix, lets try this with Ontario 2011

or Alberta 2011

You can begin to see how this is just not accurate enough. One thing it does do, however, is give a very rough indication of how many votes is needed for a majority. If you are ever in a crunch, use this "just square it" formula to figure out what will happen.

So we needed something better.

While that was going on, I was made an admin at PolUK, and, the Elections Admin vanished at the same time. I was asked to run the election, and used my formula. As expected, the other admins were not big fans, so we set out to find a better way. This is when we stumbled across a program called UK Elect. It is a wonderful program that I still have, and suggest you get even if you are not British. The program allows editable scenarios, and you can use this to create your own fictional election tests, or even, put in another countries data and run elections over there. This is what we did for Canada. The problem however is that doing this was a very complicated process, and, since Elections only happened every 5 months or so in our game, I would quite literally forget how to do it each time.

Meanwhile, I was running election projections here on this blog, my earlier posts are at about this time, 2008. I've been a fan of the UBC-ESM Forecasters ever since I first saw them, and use them almost daily, even now. I realized after playing with the numbers that party-to-party swings were not working the way I thought, so I invented a new method. I would project on a curve. The party that did the best, compared to last election, would be set at 1.000 and all other parties would have their numbers reduced by the proper amount, so that the projected popular vote would be the same as the share of their real vote. In effect, I was applying a ratio method.

I had dual needs. I needed to find an alternative to UK Elect, and, I had to find a much easier less math-heavy way to do forecasts that did not require the UBC website to be online. This is when I started work on the ElectoMatic. The intent of the ElectoMatic was to mimic the results I would get in the UBC forecaster by using by "ratio" method. Eventually I figured out the math of how to do this. If I need a 5% increase, I don't apply a raw 5% to each riding, rather, I apply 1.05. The ratio method works wonders in places like Canada where we have "2.5" major parties in each province. The simple raw swing works great where you have two parties, but breaks down when you add more.

My first ElectoMatic was done based on the 1993 election, and was to be used in PolCan for our round based on the 1993 election. The program did what it had to for PolCan purposes and that was that.

However it occurred to me... My program is designed to take old election results, new poll results, and project the result. It seems to do that just fine. Why not use it for real elections, elections that have yet to occur. Thus midway though the 2008 election I introduced the ElectoMatic, and it worked. I later created an ElectoMatic based on the 1979 election and throw in 1980 popular vote results. The errors were rather minor.

So there you have it, the history of the ElectoMatic, a file created originally to give alternate history results of past elections. This is when we get into what I've been doing. You see alternate history, especially election alternate history, has always been something I'm interested in. People say that this or that party almost did this or that well in this or that election. With the ElectoMatic you can find out just how close they came.

That brings me to what I've been working on. User calivancouver has been helping me finish up the maps from the 1950s and 60s. The maps gave me an idea to test, how close did Social Credit come to being the official opposition in 1957. St. Laurent implied it was rather close. The problem is I don't quite yet have an ElectoMatic for this era. Part of the reason is the lack of Social Credit candidates in Ontario, giving spotty results.

At the same time the old questions I've had of how to compare the 1980 and 1984 PC vote in Quebec came to my mind. I wanted to know how much of the Social Credit vote went PC, so, I ran a basic test and the answer is not much. I've also been trying to see if I can patch the holes in Ontario for Social Credit in the 50's and 60's and the answer is I can. While they only ran 68 candidates in 1962, in the ridings they did not run, they ran candidates at other points during the era. Only 3 ridings did not have SC candidates during this time, and they can be estimated based on neighbouring ridings and other vote trends.

Thus brings me to what I've been doing. Along with the above, I want to create a new ElectoMatic for past elections. Most of the files I created were for the purposes of the PolCan game. Now I'm more interested in general alternate history. My experiments in Quebec in the 80s show that a single ElectoMatic can be used for the 1979, 1980 and 1984 elections. I will tinker around with the differences between 1988 and 1993 to find if a single file can do this as well, or, if the party-to-party swing is too great. 1997 and 2000 may also require individual files. I'll also take a look at the 4 most recent elections with the same things in mind, and of course, look at past elections such as 1968, 1972, and 1974, as well as 1953 though to 1965.

Another concern of mine is that there are some gaps.

See, Nationalism in Quebec prior to Levesque was a right-wing force. Having SC represent that trend is logical right up to the mid 70s. Prior to the constitution, left-wingers in Quebec were more than willing to trust Trudeau even if they did not care for his ideas on federalism, and there is a strong enough NDP presence in the province during this time for those who did not. 1984 saw a PNQ party, the first attempts at a Bloc Quebecois, and it is these numbers that can be used. In effect, this will allow me to simulate, in alternate history, a Quebec based party for all elections from future elections, going back clear to 1953.

A larger problem is that of Social Credit and Reform. While Reform did run in 1988 and can be forecast without problem to the year 2000, and, while Social Credit did run strong elections in the west to 1965, we still have gaps in the west. One thing the maps have made clear is that where SC did well, Reform did well, and vice versa. SC had a presence in 1968, and this could be used for elections on those ridings, but for 1979 and 1980, as well as elections since 2004, trying to figure out where a western based party would stand is difficult. Even more difficult, given that throughout the entire history of Social Credit, Reform, and the Alliance, only 3 MPs were ever elected from Ontario, is seeing where that party would stand in that province. Early information shows that in many of the areas Reform did well in Ontario in the 90s, Social Credit did well in the 60s, which is encouraging.

I will continue to work on these problems and hopefully have a shiny new set of ElectosMatic for all of you in no time.