University of British Columbia
Sauder School of Business

Alberta 2005
Provincial Election Forecaster


developed by Prof. Werner Antweiler


UBC-ESM Home Page | UBC-ESM Election Forecasting

Voter Migration Matrix

You must guess the probability with which a 2001 voter will vote for a party in 2005. The sum of each row must equal one. Once you have completed your guess of the voter migration matrix, press the FORECAST button. The forecaster program will apply your voter migration matrix to the 2001election results to find out which party will win in each riding. Leaving the matrix unchanged will display the 2001 election results.

Probability to migrate
from party in 2001 ...
... to party in 2005
PCP LIB NDP ALA OTH
Alberta Progressive Conservative PartyPCP
Alberta Liberal PartyLIB
Alberta New Democratic PartyNDP
Alberta AllianceALA
All Other PartiesOTH
Show results for individual constituencies:
Press to apply your voter migration matrix.
Press to set the voter migration matrix to the identity matrix.


Methodology:

Elections in Canada employ a "first-past-the-post" voting system. Thus the prediction of the election result in terms of the seats distribution in provincial and federal parliaments must be based on a prediction of the election result in each constituency. The use of a voter migration matrix reflects the notion that voters change their opinion about candidates and parties in a similar way across the entire province. That is, a party that loses in one constituency tends to lose in another constituency and vice versa. In the simplest form, this "swing" can be applied across all constituencies.

However, applying this simple voter migration matrix to the election results from a previous election is a crude way of forecasting the outcome of a forthcoming election. It is crude in three ways. First, it does not reflect the change in population. Some children have reached voting age, some (mostly older) people have died, and other people have moved away or into a jurisdiction. Second, it does not reflect the possibility that voters from the last election opt to abstain in the forthcoming election, or abstainers in the last election opt to vote in the forthcoming election. Third, the transition matrix is applied identically across all constituencies, thus ignoring important local factors. A more sophisticated approach would take care of these three problems.


This web page and the election forecasting software are © 1995-2004 by Werner Antweiler.