Uniform National Swing – the track record

I’ve put together a short series on UNS – what it is, what’s its track record, and what levels of distortion have occurred in recent elections. This is part two of three.

When did UNS break? It can’t simply be one of those myths that it’s fairly accurate – it must have been a useful forecasting tool once. I’m going to show tables for the 1987, 1992, 1997, 2001 and 2005 General Elections, with the UNS results from the real score, the average eve-of-election poll score, the Golden Rule poll and its opposite (a “Leaden Rule” – the poll that’s worst for the Conservatives). The UNS calculator is a simple 3-party one (SNP/Plaid scores held constant) and “eve-of-election” includes all polls from the last three days of the campaign.

1987 General Election	Con	Lab	LD	Majority
Actual Result	376	229	18	Con 101
UNS from actual score	381	230	23	Con 111

Wow. Not bad. Not bad at all, given the limitations of the simplifications. In 1987, the UNS worked pretty well. So, from all accounts, did the polls (I couldn’t find the pre-election polls readily online, I’m afraid). Every forecaster was left happy – the polls were fairly accurate, the UNS tool worked well – what could possibly go wrong?

Then came the debacle of 1992, and things were never glad confident morning again for UNS. Or the polls.

1992 General Election	Con	Lab	LD	Majority
Actual Result	335	271	20	Con 21
UNS from actual score	361	253	13	Con 71
UNS from average of polls	303	307	17	Hung, Lab 19 short
UNS from “Golden Rule”	313	297	17	Hung, Con 13 short
UNS from “Leaden Rule”	290	317	20	Hung, Lab 9 short

When the polling errors were removed, John Major should have only lost 15 seats from Margaret Thatcher’s 1987 landslide. He lost 41, his majority falling from a notional 71 (under UNS) to just 21. The simultaneous failure of the polls only added to the impression that flipping coins would be more accurate. The errors occurred in opposite directions, however, and the terrible performance of UNS was masked by the polling errors – had they occurred in the same direction UNS from the polls would have predicted a Conservative majority of 173 with Labour reduced to sub-Foot levels.

It’s notable that – on a straight UNS from 1987 – an equal score of 38/38/20 would have actually left the Tories as the largest party in a hung Parliament (they’d have been on 312 to Labour’s 298). The pro-Labour distortion with which we’re so familiar – so much so that it’s now an article of faith – did not exist prior to 1992.

In 1997, the polls failed once again (with the honourable exception of ICM). However, this was barely noticed at the time, for two reasons. One – the outcome was hardly in doubt. Two – the failure of the polls was countered largely by the failure of UNS.

1997 General Election	Con	Lab	LD	Majority
Actual Result	165	419	46	Lab 179
UNS from actual score	206	396	29	Lab 133
UNS from average of polls	179	424	29	Lab 189
UNS from “Golden Rule”	229	373	29	Lab 87
UNS from “Leaden Rule”	144	456	33	Lab 253

So whilst the polls were out and UNS were out, if you’d run UNS from the average of the polls, you’d have been not too far out. In reality, Blair should have had a majority 46 less than he did, and Major should only (!) have lost 137 seats from his notional 343, leaving him on about Michael Foot levels of representation – 41 seats stronger than the position in which the Conservative Party actually found itself.

In 2001, the polls and UNS again failed in opposite directions.

2001 General Election	Con	Lab	LD	Majority
Actual Result	166	413	52	Lab 167
UNS from actual score	181	403	47	Lab 147
UNS from average of polls	160	423	47	Lab 187
UNS from “Golden Rule”	183	406	41	Lab 153
UNS from “Leaden Rule”	145	435	51	Lab 211

Again, the polling average and UNS errors nearly cancelled out. Labour overstatement in the polls was matched by Labour comfortably exceeding UNS – losing only 6 seats when UNS had that figure nearly tripled. William Hague increased Conservative representation by only 1 seat, when he should have done precisely sixteen times as well. According to UNS.

However, in 2005, the polling error and UNS error were in the same direction. Labour were once again overstated, but this time, the Tories clawed back a little of their accumulated losses beyond UNS.

2005 General Election	Con	Lab	LD	Majority
Actual Result	198	356	63	Lab 66
UNS from actual score	185	372	60	Lab 98
UNS from average of polls	177	378	62	Lab 110
UNS from “Golden Rule”	185	372	60	Lab 98
UNS from “Leaden Rule”	172	387	58	Lab 128

Howard did better than he should have done – especially when looking at the average of the final polls. Blair should have retained a landslide.

So what have we learned? That UNS has been rather divergent from reality from 1992 onwards. If it so happens that the errors in UNS are opposed to the errors in the polls, then it’s not so noticeable. If, however, there’s an expected error in UNS in the same direction as an expected error in the polls, then it will go quite badly wrong. If we believe that (for example) the marginals will help the Conservatives (and hinder Labour) noticeably beyond UNS, tactical voting will unwind and the polls may well repeat the Labour overstatement – then the UNS-from-average-poll figure would really not be one on which I’d be willing to risk my hard-earned cash. I was looking at the archives the other day and – a week before polling day – noticed two adjacent threads. One noting the “Balance of Money” prediction of a Labour majority of 80. The other noting the UNS prediction of a Labour majority of 142 …

How would things have gone last time if UNS had been true all the way? Well, we’d have to assume 1992 boundaries throughout (which have been revised twice since, and both times to the benefit of the Tories), so this really is “just a bit of fun”:

1992: Con 361, Lab 253, LD 13. Conservative majority of 71
1997: Con 234, Lab 372, LD 20. Labour majority of 93
2001: Con 249, Lab 356, LD 21. Labour majority of 61
2005: Con 277, Lab 312, LD 37. Hung Parliament. Labour short by 14 seats