In a few hours (22:00 CET, 4pm  ET) the papers and TV stations will fall all over themselves to present early results based on exit polls (unless they ignored the lesson that STV learned the hard way in 2006: even the most elaborate large-sample pre-election survey is not the same as an exit poll).  About an hour later, results will begin to trickle in and then turn into a torrent.  Both of these allow just enough data to make a prediction about the final result.  Those with any common sense will go to a movie or find something else to do until about 0:30 CET/6:30 PM ET when there may be enough data to make a final call (though in an election as close as this one, that may not be enough time), but there is probably not anybody that sensible still reading this blog.  So if you want to make a guess about the final from the exit polls or from the early results, here is what to do:

How to guess  from the exit polls:

Don’t, especially for close races (like whether Most-Hid and HZDS are above the threshold).  Exit polls are better than other kinds of polls but they are far from perfect.  Below you can find two charts with exit poll results and actual results, one for Slovakia in 2006 and one for the Czech Republic just two weeks ago in May 2010.  They are remarkably similar in their overall results:  the average difference between exit polls and actual results for the eight top parties in each is about 0.7 or 0.8 and as high as 2.0 even for medium-sized parties.  This translates into differences up or down of as much as 20%, especially (but not exclusively) for the smaller parties.  If we could assume that the 2010 difference would resemble that of 2010, then I think we could make a better prediction from exit polls, but except for SNS (whose voters might not be able to quite admit their choice to a bunch of young exit pollsters), I am not sure how this year’s exit polls will differ from results.  Of course if HZDS scores 7.2% or SMK-MKP scores 3% we can be fairly sure of those parties final position vis a vis the exit polls, but I don’t expect this.

Country	Party	Exit Poll	Results	Exit Poll Raw Error	Exit Poll Percentage Error
Slovakia 2006	Smer	27.2	29.1	-1.9	-7%
	SDKU	19	18.4	+0.6	+4%
	SNS	9.6	11.7	-2.1	-18%
	HZDS	8.6	8.8	-0.2	-2%
	SMK	11.8	11.7	+0.1	+1%
	KDH	8.6	8.3	+0.3	+3%
	KSS	4.7	3.9	+0.8	+21%
	SF	3.8	3.5	+0.3	+10%
	Average (absolute value)			0.8	8%
Country	Party	Exit Poll	Results	Exit Poll Raw Error	Exit Poll Percentage Error
Czech Republic 2010	Average (absolute value)			0.8	8%
	CSSD	19.5	22.08	-2.6	-12%
	ODS	20	20.22	-0.2	-1%
	TOP09	17.5	16.7	+0.8	+5%
	KSCM	10.5	11.27	-0.8	-7%
	VV	11	10.88	+0.1	+1%
	KDU	4.5	4.39	+0.1	+3%
	SZ	3	2.44	+0.6	+23%
	Suverenita	3	3.67	-0.7	-18%
	Average (absolute value)			0.7	9%

So take a sip of  the exit polls, roll them around your mouth, and spit them back and wait for the full glass.

How to guess from early results

I was surprised not to see this done in 2006 (or in the Czech Republic two weeks ago), but maybe I missed it.  It should be possible to use the patterns of voting returns from 2006 to help make predictions from early results.  In my 2006 live-blogging of the election I actually took snapshots of the results as they came back over time, and I hope to use these tonight to make a better guess.  Because the speed of election returns has to do with the size and rurality of precincts, some parties early returns were higher or lower than the final by a significant amount, as the graph from 2006 shows:

Parties with more rural electorates–KDH in light blue, HZDS in brown–tended to decline as the larger urban precincts began to report later in the process (Smer declined as well, though its urban-rural share was about average).  More urban parties–SDKU and SF in particular–tended to increase.  SNS and MKP-SMK, with concentrations in middle-sized towns did not change much (though this year things will be different at least for MKP-SMK which has lost much of its urban electorate to Most-Hid and should more closely resemble KDH and HZDS, with a declining trendline).

Because parties characteristics with regard to such factors changes quite slowly, this should actually provide a fairly stable source of data that would allow us to use the 2006 data to adjust the 2010 early returns (though I will also be testing the trends from 2010 against those of 2006 as they happen to see if they truly are consistent.)  In any case, if 2006 serves as a good guide, here is a matrix to calculate the “actual result.”  Search for party and the number of precincts returned at any given time and multiply the result of your party by the percentage listed.

Party	Adjustment factor: Multiply party score by percentage below to get better predictions of actual results
	Number of precincts reporting
	1000	2000	3000	4000	5000	5800	5900
Smer	92%	95%	96%	97%	98%	100%	100%
SDKU	123%	109%	105%	105%	104%	101%	100%
SNS	100%	100%	100%	101%	101%	100%	100%
MKP	109%	109%	108%	100%	96%	99%	100%
HZDS	92%	99%	101%	101%	102%	101%	100%
KDH	94%	96%	96%	98%	101%	100%	100%
KSS	89%	94%	97%	98%	99%	100%	100%
SF	118%	105%	102%	103%	102%	100%	100%

I will try to do this on the blog, but feel free to try it at home.  It would not surprise me if the Slovak press has something like this in store  (though I have occasionally criticized them for their use of polls, Slovakia’s journalists have been fairly good at adopting new methods if they will give them a leg up on the competition.

I make no promises for this model any more than for the exit poll adjustment, but  those who have not done the sensible thing and simply ignored the whole thing until final results are in will probably appreciate the entertainment value.