Ok… In a perfect timing moment, I do loads of work on social media and crowd sourcing for the general election and just as the election is called I bugger off to the West Indies. I see England win the Twenty20 world cup and generally dick about in the sun and take a well earned break, so I don’t care enough about the poor timing :-)

Now we have a government (well kind off) and I’m back from the Caribbean; I thought it was about time to write something about the tmg election tool.

Here are the basic details of the tool:

  • The system used crowd sourcing of information instead of the class polling method to collect the feel of the next election
  • The methods used resulted in over 3 Million pieces of information being collected over the month of ‘polling’
  • The system would check the expected outcome of each seat (though various sources) and ranked each of the standing parties into a win percentage chance matrix
  • Each time a new data collection occurred, the new win percentage was ranked against the last win percentage
  • The change to the previous to the current win percentage was calculated and then projected forwards to the election date -this is like a horse closing in on the rails in a horse race; so you a calculating if the current leader is going to be overtaken before the finishing line
  • A hybrid seat predictor was also created, this applied the classic polls over the seat prediction to add a drag factor to the crowd sourced information to see what the difference were
  • The system also took the seat prediction and turned that into a national vote share (which ended up following the classic polls rather well)

Observations:

  • The system initially showed a Tory victory however, over the month the number of seats they were expected to win dropped to below the 326 mark
  • The second option for the Tories to join up with the DUP (with their 7 or 8 seats expected) also started to fall away
  • This resulted in the fact that the Tories were likely to either form a minority government or do a pact with another party -Namely the Lib Dems (called on the 28th April on this blog)
  • The system showed that a Lib/Lab pact would not work as their combined total would also be short of 326 seats (called on the 28th April on this blog)
  • The system, at the time of doing the ‘final call’ and ‘seat by seat call’ was done too early as there was a shift away from the Tories in the final evening before the polls opened. This put the Tories on 315 instead of 321 (with the Hybrid poll going from 317 seats to 310 seats)
  • The ‘web’ (and the classic polls) clearly overplayed the Lib Dem success of the leader debates. The seat predictor had the Lib Dems getting 55 seats before the debate, where is shot up to 83 seats by the ‘final call’. Here we can see people getting over excited about something different. It shows that the leader debates didn’t actually have that much of an effect on the result and it was only the Twitter and media types that were wowed by it all. This lead to the Lib Dem call being too high and the resulting Labour vote being too low (either this or the comments on the illegal immigration amnesty scared off a lot of Lib Dem votes right at the end of the campaign -as some commentators have noted)

Conclusion:

  • All in all the seat predictor should be considered a success. It predicted results like the Greens getting Brighton Pavilion when people thought this was clearly wrong. It also predicted some unexpected Tory gains where the old media was saying it wasn’t possible.
  • The predictor can pick up local issues in a seat, which the national polls can’t -this is an advantage of the old media polls
  • The predictor was right to not include votes in Northern Ireland as the political landscape there is too different to the rest of the UK (for example with the DUP losing Belfast East with a 23% swing away from them)
  • Both the old media polls and the seat predictor (and social media along with the web in general) reacted too strongly to Nick Clegg’s leaders debate performance. Therefore you can say that the seat predictor got that wrong; however so did everyone else (with news papers polling the Lib Dem’s and Labour getting about the same percentage share of votes)
  • As a result the number of Labour seats was too low and the number of Lib Dem seats too high
  • At midnight before the election the Tory seat number was 315, and the hybrid was 310 -this was closer than ANY UNS projection from the old media polls
  • The data collected shows the feel of the election over the month, so this information could be linked to the news to see what each party did right and wrong on a day to day basis
  • The seat predictor can work in a multi party system; the polls and the UNS only really works in a 2 party system -therefore it is more progressive than the old media methods. If more elections end up having a strong third party, the UNS will only show it’s weakness once again

Therefore the tmg seat predictor was closer to any UNS swing from polls published in the papers and in other media outlets with only the exit poll (which is post election so doesn’t count) being closer to the actual result. The tmg seat predictor didn’t cost the volumes of money spent by parties and papers to collect information (which wasn’t as correct anyway). So this method is both cheaper and more accurate than the polls.

(click for full size)

It wasn’t 100% correct (annoyingly), however at midnight (the final call and seat by seat call was at 4pm on the eve of the election, due to trying to rush out the seat by seat results with still being able to go to bed) the results calculated was the most accurate call of the election result; beating all the polls, all the papers, all the TV news. It also lead to the conclusion that the Tories would either try and form a minority government or join with the Lib Dems, as well as highlighting that a Lib/Lab pact would not collectively reach the magical number of 326.

(click for full size)

Therefore I would call it a success!!

Roll on the next election… maybe in October? :-)