Predictions

The Predictions Page

This page has not yet been updated to take into account the November 2003 prediction contest, let alone the current June 2004 contest.

1998 contest

This site was first set up to cover the 1998 Assembly elections. I invited visitors to the site, and also readers of soc.culture.irish and uk.politics.electoral, to send me their predictions of the eventual results, and eight people replied. I added in my own best guesses and another two predictions from newspaper sources, to make a grand total of eleven entries, nine of which had specified the result for each party in each of the eighteen constituencies. The best set of predictions in the latter category was made by Keith Mills who got the party representation of eight constituencies completely correct and whose overall total varied by sixteen seats from the real result. However, Brendan O'Leary and Chris McCrudden in the Sunday Business Post had come much closer overall though without predicting the results in individual seats: their variance from the real result was only 14 seats. The two best predictions (for the overall result) were therefore as follows:

Party	DUP	UUP	PUP	UKUP	Oth U	Alliance	NIWC	SDLP	Sinn Fein
Result	20	28	2	5	3	6	2	24	18
O'Leary and McCrudden	22	29	2	3	0	8	0	25	19
Keith Mills	22	30	2	3	1	8	0	26	16
Average of all entries	21	32	3	2	0	9	0	26	16

1999 contest

The next election was the European election of 1999, a much less exciting event. Once again I invited visitors to the site to submit their predictions of the percentage results gained by each candidate/party; this time only five people did so, with my own guess making a sixth. The most accurate entry was submitted by Tom McVey (which made up for his having submitted the least accurate prediction in 1998); his average variance from the eight candidates' real results was 1.54%. Patrick Matthews, who uniquely submitted predictions to an extra decimal place, got closer than anyone else to three of the eight results. Their entries were:

Party	DUP	UUP	PUP	UKUP	Alliance	Nat Law	SDLP	Sinn Fein
Result	28.4%	17.6%	3.3%	3.0%	2.1%	0.14%	28.1%	17.3%
Tom McVey	25%	19%	2%	5%	4%	0%	29%	16%
Patrick Matthews	25.4%	16.8%	3.8%	6.7%	4.5%	0.2%	28.6%	14.0%
Average of all entries	24.9%	17.3%	3.8%	6.5%	5.1%	0.0%	27.4%	15.0%

2000 contest

The next election, caused by the unexpected death of Clifford Forsythe, was the parliamentary by-election for South Antrim in 2000. This time there were twenty-four entries in my prediction contest, which perhaps reflects the growth of the internet and the extra interest in the election, which it was anticipated would be a tight run between the DUP's William McCrea and the UUP's David Burnside, with four other candidates making up the rest of the field. The best prediction, with an average variance of 1.74% and within 1% of the result for four of the six parties, was David Boothroyd's; Noel Horgan came second with an average variance of 2.15%. Only six of the twenty-four entrants anticipated McCrea's victory, and most of them expected a much lower Unionist vote share.

Party	DUP	UUP	Alliance	Nat Law	SDLP	SF
Result	37.95%	35.26%	6.64%	0.16%	11.44%	8.54%
David Boothroyd	33%	36%	7%	0%	15%	9%
Noel Horgan	35%	35%	5%	0%	10%	15%
Average of all entries	30.9%	34.1%	8.0%	0.2%	15.4%	10.9%

2001 contest

The most complex competition so far was for the two sets of elections held simultaneously in 2001 for the 18 Westminster seats and the 582 local council seats. I asked entrants to predict, seat by seat, the 18 winners of the Westminster election, and also to predict the number of local council seats won by the five largest parties. I received 131 predictions of the Westminster outcomes, and 115 for the local council elections. Astonishingly, 4 of the contestants accurately predicted all 18 Westminster seats: Cai Larsen, Seth Speirs, Killian Forde and Tom Nolan.

All 18 seats right: Cai Larsen, Seth Speirs, Killian Forde and Tom Nolan

Seats for which most entrants predicted the right result: DUP to hold North Antrim, East Belfast, gain North Belfast; UUP to hold Lagan Valley, Upper Bann, South Belfast, gain North Down; SF to hold West Belfast, Mid Ulster; SDLP to hold South Down, Foyle, Newry and Armagh

Seats for which most entrants predicted the wrong result: DUP gains in Strangford, East Londonderry (UUP hold was expected); SF gain in Fermanagh and South Tyrone (UUP hold was expected); SF gain in West Tyrone (SDLP gain from UUP was expected); UUP regain byelection loss in South Antrim (DUP hold was expected)

Of the four who got all 18 Westminster seats right, Cai Larsen had the best prediction for the local elections. Conal Gillespie's whose prediction of the local election results was eerily accurate, and he won a cash prize that had been put forward by Paul Linehan (Cai Larsen won a book from me). The below table shows his and Cai's predictions for that contest:

Party	DUP	UUP	Alliance	SDLP	SF
Result	131	154	28	117	108
Conal Gillespie	131	156	30	110	104
Cai Larsen	110	155	35	115	95
Average of all entries	107	163	35	121	94

How to calculate the "winner"

This last set of elections provoked some discussion between me and other participants who felt that rather than measure the absolute variance between a prediction and the result, one should weight the results by the electoral strength of each party - or perhaps square the variance as well. I am not inclined to follow either of these routes. If one tries weighting by the strength of each party, then in a situation where Alliance got 5% and the SDLP 20%, a prediction of 15% for Alliance (10% error) would be penalised less severely than a prediction of 17% for the SDLP (3% error), which cannot be right. As for squaring the differences, they mount up depressingly quickly in any case.

However, I take the point that for instance in the South Antrim case, it was more important to get the DUP/UUP vote shares right than the others, and for the European elections the results of the top four parties mattered more than the rest. Therefore for future contests where the aim is prediction of vote share, I shall double the weighting of winner(s) and runner-up. However for predictions of seats won I shall stick to the mean variance counting all parties equally.

Lessons learned

The most striking thing about these prediction contests, for me, is that so many of the predictions made have been so very wide of the mark. Even the best entries in 1998 failed to pick up on how many seats would go to the the UKUP and other Unionists, or to anticipate the UUP's low total. In 1999 the DUP and Sinn Fein surprised with their good result, and Alliance and the UKUP did far worse than anyone anticipated. The predictions overall certainly match people's expectations of the results - for instance, Iris Robinson's victory in Strangford in 2001 had been so widely anticipated that one was tempted to ask if she was not disappointed that she did not do better. However they don't match the results themselves particularly well.

Only three people have participated in all five prediction contests to date. Their record does not suggest a strong learning process. Keith Mills, as already mentioned, had the best result of those who predicted each constituency in 1998. In 1999 however he came fourth out of six. In 2000 he was eighth of 24; and while he scored well in the Westminster contest in 2001, getting only two seats wrong, he came 85th of 115 in the local government part of the contest. Ciaran Quinn has come in the bottom half of each contest so far - which I am pleased to see has not deterred him from setting up a predictions page for the 2002 Dail election. I myself have managed to finish in the top half of the screen only once, for the 2001 local elections. 13 people have entered three contests only; of them, Sean Fear has finished in the top 30% of the contest, and D.J. Moore in the top 40%, on all three occasions that they have entered.

The Future

The next scheduled election in Northern Ireland are the Assembly elections in 2003. I anticipate running a constituency-by-constituency contest then as I did in 1998, along with perhaps more interaction than before, depending on available resources.

The Prediction Contests: 2004 | 2001 | 2000 | 1999 | 1998

Other sites based at ARK: ORB (Online Research Bank) | CAIN (Conflict Archive on the INternet) | Northern Ireland Life and Times Survey

Front page | Site Map | Notes and Queries | Updates Mailing List | About this site

Your comments, please! Send an email to me at nicholas.whyte@gmail.com.

Nicholas Whyte, 17 March 2002

The Author | What's New | Your Comments