Bureau of Meaningless Statistics: The (Non)Effect of Undecided Voters

There is an old saying that “figures don’t lie, but liars do figure” (which I’m sure has some equivalent in almost every language) and there is a wonderful book written in 1954 provocatively entitled “How to Lie with Statistics.”  In Slovakia’s election coverage in 2010 the challenge is not so much statistical lies as lazyness.  Figures appear from various polling firms and they are duly published by newspapers that what people to pay attention whether they have solid basis in fact (nobody’s lying, per se, but they also have no way of knowing whether they are telling the truth) or whether they have any impact.  As a case in point take today’s article in SME, “Smer and SaS can score among undecideds.” (HN does the same)  It is, perhaps, interesting that this is the case, but the article makes little effort to deal with the two real underlying questions:

  • First, is this a useful way of adjusting polling numbers? I don’t know, but neither does SME.  I don’t have any evidence imediately at hand, though I will look to see if I have any precedents from 2006.
  • Second, if this were useful in adjusting numbers, would it have any effect on the overall outcome.  Here the calculation is the work of about 5 minutes at a spreadsheet (use focus data to figure out the overall share of undecideds, multiply this by the percentages printed in the article, add this to the original percentage gained by the party, recalculate to equal 100).  The results are in the table below.  And the answer is “not much”
Party May Poll Share Share among undecideds Contribution of undecideds Revised preferences
(sums to more than 100)
Revised total share
Smer 35.3 16.9 2.6 37.9 34.6
SDKU 14.0 6.6 1.0 15.0 13.7
SaS 13.3 12.8 1.9 15.2 13.9
KDH 8.3 5.8 0.9 9.2 8.4
SNS 6.1 3.5 0.5 6.6 6.1
MK 5.9 2.8 0.4 6.3 5.8
Most-Hid 5.6 4.6 0.7 6.3 5.8
HZDS 5.1 3.5 0.5 5.6 5.1
KSS 1.6 1.6 1.5
Unie 0.8 0.8 0.7
SDL 0.8 0.8 0.7
Paliho Kapurkova 0.7 4.0 0.9 1.6 1.4
ND 0.7 0.7 0.6
EDS 0.7 0.7 0.6
ZRS 0.6 0.6 0.5
Nase Slovensko 0.4 0.4 0.4
SRK 0.1 0.1 0.1
Azen 0.0 0.0 0.0
100.0 109.4 100.0
Won’t vote 16.1
Undecided 15.1

No party shifts its share by more than 0.7 percentage points, no party drops below the threshold, and the only shift in relative ranking is that SaS slightly overtakes SDKU (and Most-Hid ties MKP-SMK).  And what effect would this have on overall parliamentary outcomes?  Well almost nothing. As the graph below shows, run these percentages through the seat calculator and you get the following results: Current coalition minus 1, current opposition plus 1.

May Parliamentary Seats Revised Parliamentary Seats Seat Change
Smer 57 56 -1
SDKU 23 22 -1
SaS 21 22 +1
KDH 13 14 +1
SNS 10 10 0
MK 9 9 0
Most-Hid 9 9 0
HZDS 8 8 0
150 150

Buried in these results is actually a strong incentive for papers to do the deeper (which is to say not very deep at all, but at least not utterly superficial) calculation.  The coalition v. opposition numbers for the original FOCUS poll (without undecideds) is 75:75.  Add in the undecideds and we get a new parliamentary balance: Coalition 74, Opposition 76.  Had SME only run the numbers, could have run the equally meaningless but far sexier headline, “Undecideds give opposition majority in parliament.”  Maybe it’s a good thing that the busy reporters at Slovakia’s papers don’t have time to do the extra work.

1 thought on Bureau of Meaningless Statistics: The (Non)Effect of Undecided Voters

  1. One big difference would be that if SaS get more seats than SDKU, and the opposition get a majority, then SaS would presumably have the right to provide the prime minister.

Leave a Reply