How did the Riding Projection models fare?

Ontario Election 2007

Hill & Knowlton

Election Prediction

LISPOP WLU

Nodice

Democratic Space

GP Murray Research

8-model Avg

Oct 11th  2007 Reflections:  Pollsters luv to use the caveat phrase "Our work reflects the sentiment of today only" and they dread predictions.  Yesterday's results underscore the oft dilemma.  A long term trend was in place.  When asked to forecast an Election Nite outcome, do u extrapolate the trend in play or do u revert to the mean ... or go with what ya got?

Those three choices befell every modeler that attempted to take the projections and modify them for a potential Election Nite result.  Methinx we all trust that the projections were accurate for the preceding few days.  But 17 years of this shows us that forecasting ahead is a crapshoot.

Evalina and i agonize before each Election on which way to go.  In a rolling poll it is augmented by the fact that we see changes in trends before they show up on conventional charts or releases.  And do u go with the small sample changed trend or the variance of the larger sample?  It's tuff and becomes an art.

For this Election, and against the advice of Barry Kay, we were still feeling the frustrations of the last two Cndn Fed episodes where there were last moment recoils to the mean.  Those that went with the outliers or beyond to extend the trend were humiliated.  But what works Federally may not Provincially i was warned ... and we found out!  In short, there are no hard and fast rules of what is happening with last minute strategic voting, going with second choices and/or the dispersal of the Undecided component.

And so we await the Throne Speech.  See it on CPAC if u are satellite/cable challenged.  And if u are a modeler, Harris-Decima has just posted two new Federal polls.  Please send to us your conversions for our Canada 2009 Chart.  It may become "2007"...

Note that at dissolution, there were 67 Liberals, 25 PCs, 10 Dippers & 1 vacancy (total of 103).  There will be 4 new seats due to redistribution this year.  The Projection chart has totals of 107 after 2005.

Election Day Blog:

Oct 10th  2007  5pm:  The polls close at 9pm EDT.

Oct 10th  2007  4pm:  All the models' final numbers have been logged on our Scoreboard.  We'll check once again before the counting begins...

Oct 10th  2007  3am:  Congrat's to David MacDonald of NODICE on his awesome Newfoundland Election Seat Projection.  He used the Antweiler application but the only poll data was from August!

Oct 10th  2007  2am:  A late "revision" to the Morrow "Final Revision" changes some of the chart and Scoreboard numbers.  And more bleeding...

Oct 10th  2007  1am:  I go back a few decades, but i still can't remember a day when four major polls were released (Harris-Decima, Strategic Counsel, Environics & SES.  In fact it's six when we include the Thanxgiving Weekend releases by Angus Reid & Ipsos.  Wow!

Nik Nanos of SES Research has owned the bragging rights in the polling circles since his coup in January 2006.  This week's barrage is all about "unseating" him.  But whoever's poll u read today, the bad news for John Tory is that the hemorrhaging continues.

Altho the AVG of all the Riding Projections says 68 Liberal MPPs will be elected today, our TrendLines Research prediction is 63 and is based on our intuition that the Undecided and Strategic Vote factors will dampen the voting pattern as represented by the dotted target trendlines in the upper chart.  These "target" figures are in the table to the left.

The finest Seat Predictions in the past two Cndn Federal Elections were by Milton Chan, Greg Morrow, Ekos & the UBC Stock Forecaster.  Come back tonite to see how the 8 2007 Riding Projections each did in the end!  We've also enhanced our Riding Projection Scoreboards for the past two Canadian Federal Elections ... a dozen modelers in all are measured:

2007-Ontario-Models-Scoreboard  2006-Canada-Models-Scoreboard  2004-Canada-Models-Scoreboard

As mentioned yesterday, the 2007 Campaign exits the Thanxgiving weekend with the Liberal Party at its highest Seat Projection since March 2005 ... and vice versa for the PCs.  The NDP have been in a 9 - 14 range over those 31 months as seen in the long term chart above.

Having converted sentiment gauged over only the last 19 days, the 8 Riding Projection models are in agreement that the Liberal Party is currently in the 64 - 73 MPP range;  PC support today would indicate 24 - 35 winnable Ridings;  and the NDP with 8 - 13 seats.

Today's chart status of 68/28/11 (and all figures since the 2003 Election) are the compiled avg of available recognized Riding Projections mentioned in the footnotes & in our new 2007 Scoreboard.  Each projection model has its own methodology of converting polling/canvassing data to estimates of winnable Ridings by the Parties.   The "today" or "October" conversions on both charts are based all 8 models and their analysis of recognized polling/canvassing completed Sept 21 - Oct 9.

Since the 2003 Election, TrendLines Research has been plotting the avg of available projections and then extrapolating each Party's long term trend to project their probable status forward to Election Day.  We call this evolving number our Election Day Target.

Albeit our Riding Projection chart illustrates their status at 68 potential MPPs as of today, present core momentum (represented by the same chart's long term trendline) would indicate that the Liberal Party may lose 5 additional Seats on the way to a reduced Majority of 63 on Election Nite.  The PC's will likely rise to 33 MPPs & the NDP are mired at 11.   I've developed this trendline methodology to better predict Election Day surprises which accounted for errors of about 28 Ridings in the last two Canadian Federal  Elections as measured by the TrendLines AVG Projection.  We have our finger crossed, eh!

The Campaign period or Writ Chart reflects that since June 4th, TrendLines Research has been predicting a Liberal Party Majority with Election Day targets ranging from 59-64 MPP's.  The TrendLines model blends the data from available pre-Campaign Riding Projections and generates high/low Election Day Targets for the Party most likely to prevail.  The 8 Riding Projections' data during the course of the Writ slowly refines this high/low range.  For the Liberals, the figures had narrowed to 69/54 by Sept 18th.  The low of 54 MPPs had obvious significance and barring breaking scandal or an unconventional event ... on Sept 18th TrendLines Research declared that a Liberal Majority was inevitable from that point in the Campaign.

Oct 7:  Going into the final days of the Campaign, we are pleased to wrap up this project with the addition of an 8th Riding Projection Model ... by Graham Murray.  TrendLiners may remember that Alan Hall of GP Murray was a model contributor to our 2004 Canadian Federal Election tracking.

Oct 4:  With his awesome accuracy, Nikita Nanos of SES Research blew away the other pollsters in the wrap-up series of finishing polls at the end of the 2006 Federal Election campaign.  It is said that the late stage nuances are crucial to good Riding Projection conversions in the final count ... to catch moves to second choice and discovery of where to the Undecided camp are breaking off.

At TrendLines, our intuition is that there is a "return to the mean" phenom in play in those final days (or even when standing in the ballot box) that assists in explaining some major variances we have seen as the Vote is tallied.

Meanwhile, at LISPOP, Barry Kay's studies confirm the accuracy of Riding Projections during the Campaigns.  One of his contributions was to have us compare Advance Poll results (easily available from Elections Offices) with the Projections as they were on those dates.  Kay goes on to confirm that when final votes are tallied and converted, they confirm the methodology.

The art of Projections seems to be in the ability to predict any change that may occur in those last 72 hours in the absence of poll data.  The question arises:  Does the last trend continue or do folks pull back to a certain comfort level?  Because it may be the latter, some Projection models do not put alotta weight on what may be developing and thus revealed in those late days...

It is for this reason that the TrendLines Election Day targets are not tracking "up" in harmony with the recent Riding Projections that are inferring a Liberal Party win in the mid to high 60's ... just as we stayed in the "62" range when there was a flurry of Minority Gov't predictions early in the Campaign.

Oct 2:  With more models data, it is confirmed today that the PC Party is at its lowest support level (from a Riding Projection perspective) since March 2005.

Sept 27:  With all 7 Riding Projection Models digesting post-Debate sentiment via canvassing and/or the new Angus Reid & SES surveys, their consensus thus far is that there was a net loss of three (Seats) from the PC's camp to the Liberals.  The 3-5 poll rolling avg's by most models shields the actual outcome.

Sept 25:  TrendLines applauds Nikita Nanos of SES Research for again publishing their Ontario poll today with the inclusion of a Margin of Error (MoE) based on the Decided voters.  He went the extra mile even though the Undecided factor was a respectable 14%.  By doing so, SES joins with us in shaming the pollsters hiding poor quality control and methodology via MoE's that disguise a high Undecided component and base the MoE on "total" respondents ... not committed voters.

Sept 20:  Note that the seven Seat Projection Models that we track have all purged pre-September polling data.

Sept 19:  The seven Riding Projection Models are currently processing only 18 days of polling data!  All but a couple of days of August have been purged.  This allows us to introduce for the first time today a writ-length trendline.  It is significant  in that it illustrates trend courses that are presently opposite of the post 2003 era (which are the basis of the TrendLines Election Day targets).

Sept 18:  Results of the fifth of the seven Riding Projection Models to digest last week's tighter polls brings down our target for the Liberal Party a notch.  Our model foresees a high/low range for each Party's finish on Election Day based on their performance in past and present blended Riding Projections (as many as seven recognized Models).  As of today, this low number for the Liberal Party is 54 Seats.  Thus, TrendLines Research declares a Liberal Majority is inevitable at this stage of the Campaign!

Sept 17:  Thursday's Leaders Debate takes on more significance with recent polling.

Sept 16:  We know that one third of the populace makes their vote decision based on Leadership qualities, another third on Party platform/ideology and a third are influenced by the preferred local candidate.  One revelation in the Angus poll was that 54% favour Tory as Premier.  But it also confirmed that most folks are agin faith based education funding ... the major issue.  We seem to see a situation developing where the Leadership factor is rising to the surface and affecting sentiment numbers.

With more McGuinty/Tory confrontations in the media, the chance at comparison seems to be going in Tory's favour.  This will tick off most of the pundits that seem by their writings to have an open disgust for Tory.  That being said, on top of the 2% Angus Reid poll & 4% Environics poll, yesterday Ipsos (Sept 4-13) confirmed the trend with their revelation of only a 3% lead for the Liberals.  Folks, we may have the signals of a contest! 

Sept 15:  Well, the Angus Reid poll (Sept 7-8) certainly spiced things up to end this week.  With its result that the PC's are within 2% of the Liberals, at least one Model projects a Tory Minority based on that one survey alone.  Is it a rogue poll or is the Tory Blue Machine finally making an appearance?

Later surveys will confirm the trend that both it and Environics imply.  The Environics (Sept 6-9) survey shows the Liberals with only a 4% lead; but it was a only a 501-voter poll with a 4.4% Margin of Error amid a 21% Undecided factor.  In other words, only 396 callers were Decided.

This brings to issue the larger question of integrity of data, ethics and methodology.  Angus Reid & Environics seem to belong to a camp of pollsters that are resistant to base their Margin of Error (MoE) on "Decided Voters" rather than the number of folks called.  With respect, imho the Angus Reid's 28% Undecided Rate should be interpreted as a 4% MOE ... not the 3.6% that it boasts. Only 605 respondents were Decided participants yet the Angus Reid MOE is based on 725 surveys.  Pollsters that hide poor methodology and/or quality of service via an archaic MoE calculation are being less than truthful about their accuracy.

To the extreme, there is Angus Reid's protégé Donna Larsen of DataPath Systems publishing political polls with Undecided components as high as 31%, 37%, 48%, 49% & 72% ... all with respectable looking MoE's however 'cuz that calculation is based on the gross number of respondents.  Come on boys and girls, let's all use the "net figure" in 2008.  Be brave.  Be ahead of the curve.  Use a footnote that your MoE is based on the Decided component only.  It's time for the polling industry to review this dated metric and not hide potential inaccuracy via the wiggle room allowed by flawed MoE latitudes.  OK, rant over!

Sept 14:  Tonie Chaltas's Hill & Knowlton Predictor was added as our 7th model today.  TrendLines already incorporates her studies in our Canadian & Quebec Riding Projections.  Merging her data over the past six months had only minor influence on our chart (gain of 1 for NDP from Liberal August tally).  Pre mid-August data has been purged from all the models.

Sept 12:  We're introducing a higher resolution writ period chart today that tracks the AVG of 6 analysts' conversions of latest opinion polls to likely Riding wins for each Political Party.  It also tracks the TrendLines Target of our prediction (extrapolation) of how many MPPs that the leading Party (currently Liberals) should attain looking forward to Election Day on Oct 10th.  It reflects that since June 4th, TrendLines Research has been predicting a Liberal Party Majority with 59-64 MPP's.

Sept 11:  A 5-model analysis of new September figures confirms an advancing divergence.

Sept 10:  With the first September poll figures available, the purging of June & July data by all the models is facilitated.  A possible closing of the gap could be in progress.

Sept 6:  The Leaders Debate has been announced for Thursday Sept 20th;  6:30pm on CPAC et al.  20 days prior to balloting, it won't appear to change the order of the standings but could affect the final numbers.

Sept 5:  With a major revision of Barry Kay's LISPOP Grid Model, our chart has been amended to more accurately reflect post-2003 sentiment as revealed in over 60 recognized polls.  The net effect has been higher MPP counts for the Liberal Party & NDP at the expense of the PC Party.

Sept 4:  The addition of two more Seat Projection Models over the Labour Day Weekend clarifies the most recent divergence that surfaced in late July.  And a flurry of updated polls to analyze allows all models to purge pre-June data.

Random commentary:

August - Late in 2005, TrendLines Research picked up a change in momentum that allowed us to commence published predictions of a Federal Conservative Minority starting Nov 30th ... seven weeks before Election Day; and in spite of the fact that the Liberal Party was ahead in almost all polls from May to Christmas!  This same technique sensed what appeared as fatal softening of Ontario PC support in early June 2007; and a Tory victory projection was halted.  Since June 4th, TrendLines has been predicting a Liberal Party Majority with 59-64 MPP's.

July - There have been a dozen lead changes since Autumn 2003.  But with a mere 60 days 'til ballot day, time is running out for a PC challenge.  While this ongoing volatility in sentiment should give the PC's a glimmer of encouragement, the NDP are indeed mired.

June - Will strategic voting come into play?  As the electorate begin to grasp that this is only a two-horse race, the prospects of a couple of scenarios that could influence:  (a) Core support leaves the NDP in droves to their second choice, or  (b) Dippers become encouraged that they can drive the Agenda as the Balance of Power in an upcoming Minority Gov't ...

Our 2007 projections are based on a blending of 7 Seat Projection Models (by Barry Kay, Greg Morrow, Milton Chan, David MacDonald, two studies by anonymous Paul Socialist of Paulitics, Tonie Chaltas & Graham Murray ... all who do stellar work for little recognition).  Their studies involve the conversion of sentiment revealed in Canvassing and/or published national/regional Public Opinion Poll data (with avg 3% margin of error) into Projections of how many MPPs each Party would win at the time the data was taken (not Election Day).  The extrapolation of Riding Projections advanced to Election Day is an analysis done by Freddy Hutter of TrendLines Research.

Over 60 post-2003 Election polls have been crunched by some of the models prior to this Campaign for the pre-Campaign tallies; Some of the models' conversions apply the UBC Election Forecaster designed by Werner Antweiler.  Please note that fine tuning of the models sometimes affects older figures on the chart and several chart values have been deleted for clarity.  The Election Day projection will incorporate other recognized models as Provincial polling becomes more frequent.  Only the Yukon Projection includes direct polling by TrendLines.  Scroll around or use the menu for our "Race for the Kremlin", Newfoundland/Labrador, Canadian Federal, Quebec, Yukon, UK & "Race for the Whitehouse" projections.