Latest Publications

Election Day: Return Trends 2

Trends have stabilized allowing a range of confidence for final results.  The chart below shows trendlines (solid) beginning almost from the beginning, and shorter trendlines (dotted) beginning from about 50% reporting.

These results would suggest the following government possibilities:

SNS 20
Smer 50
SMK 20
KDH 14
SF 0
Smer+ HZDS+ SNS 85
Smer+ KDH+ SMK 84
Smer+ SDKU 81
Smer+ SDKU+ KDH 81
Smer+ SDKU+ SMK 81

This is not too different from what we’ve seen elsewhere but secures at least the mathematical possibility of Smer+KDH.  From the looks of it, bargaining will determine the coalition.  Fico has three distinct possibilities: HZDS/SNS, KDH/SMK or SDKU.  Since he has resolutely refused to suggest a coalition direction, it may come down to the best offer.

Election Day: Return Trends

Sme’s retransmission of the election numbers as they appear offers an interesting opportunity to collect and process potentially useless data.  I’ve been downloading the results at roughly 15-20 minute intervals and the numbers can fortunately be dropped immediately into Excel.  As might be expected, the differences in the type of party demographic and tendency of certain types of precincts to report early or late produces what may appear to be a linear relationship over time that could (in theory, if all else is equal) be used to predict final outcomes.  I do not necessarily believe that this is the case, but I present, for your entertainment, what that would look like:

What does this (meaninglessly) predict:
Smer: 28.5%
SDKU: 24%
SMK: 12.5%
SNS: 11%
KDH: 8%
HZDS: 6%
SF: 4.5%
KSS: 3%

‘Twould be an interesting world if that were the case…

Election Day: Coalition Scenarios

It is still too early to tell whether either the OMV numbers accurately reflect the outcome (STV has been using it as its baseline for TV coverage, even though commentator Darina Malova has been extremely adept at suggesting caution) and whether the surprising MVK exit poll results are any better (again, some of the results are surprising), but it is possible from this vantage at least to look at what might or might not be possible.  This is just "wheel-spinning" since we will need a better look at results, but I have a few minutes, so why not.

First, though, I am struck that my calculations show something very different from the way that Pravda has translated the Markiza/MVK numbers.
I cannot be 100% certain that these are wrong, but I’ve reworked my own calculations several ways using Markiza/MVK numbers and I get something different:

                       Share       Seats
                                               Their #        My #
Smer         27.2%           45                48
SDKU        19.0%           32                34
SMK            11.8%            21                21
SNS             9.6%              18                17
HZDS          8.6%              17                15
KDH            8.6%              17                15
KSS               4.7%              0                    0
SF                    3.8%              0                    0

If true, my numbers suggest a significantly higher percentage to the two largest parties.  I am not sure how we arrived at such different numbers, but if my method is wrong, I would very much like to hear from somebody why.

In any case, I present below a number of possible seat distributions depending on polls (and, apparently, methods of calculation)

  OMV/STV Markiza/MVK Markiza/MVK Markiza/MVK + KSS (5.0) Markiza/MVK + KSS (5.2)
HZDS 18 17 15 14 14
SNS 20 18 17 16 16
Smer 46 45 48 46 45
SMK 17 21 21 20 20
KDH 16 17 15 14 14
SDKU 23 32 34 32 32
ANO 0 0 0 0 0
HZD 0 0 0 0 0
KSS 10 0 0 8 9
SF 0 0 0 0 0
Smer+ HZDS+ SNS 84 80 80 76 75
Smer+ KDH+ SMK 79 83 84 80 79
Smer+ SDKU 69 77 82 78 77
Smer+ SDKU+ KDH 85 77 82 78 77
Smer+ SDKU+ SMK 86 77 82 78 77
SDKU+ SMK+ KDH 56 70 70 66 66

Of main note here are several points:

  • Smer+HZDS+SNS is viable in every circumstance (unless KSS manages to act as a spoiler and gets slightly above the 5% line, thus getting 9 seats instead of the minimal 8)
  • Smer+KDH+SMK is viable in every circumstance, as are all other instances of Smer+2 of the three government parties
  • Smer+SDKU is viable according to MVK/Markiza (thanks to SDKU’s strong showing there) but not in OMV
  • SDKU+SMK+KDH is not viable in any circumstance though if both SF and KSS fail to get into parliament it may only have to "acquire" 6 votes from elsewhere.
  • Even should SF exceed the threshold (a separate set of scenarios not shown here), it would still give a right wing coalition only 74 or 75 seats under the MVK numbers and fewer under the OMV numbers.

Polling results have started to come in, though it’s going to be hard to make guesses until the share of votes counted gets high.  As some have noted, the returns will come in last in the cities and that might show a disadvantage for SDKU/SF until the final hours.  We shall see.

Election Day: Markiza’s Numbers

SME and Pravda have both published Markiza’s guess based on MVK numbers (the Markiza website’s Sprava page focuses instead on Siamese Twins in Los Angeles!).

Markiza Difference from average latest polls
Smer 27.2 -0.6
SDKU 19 +6.1
SNS 9.6 -0.3
HZDS 8.6 -3.0
SMK 11.8 +1.8
KDH 8.6 -0.7
KSS 4.7 -1.2
SF 3.8 -1.5
ANO 2 +0.4
HZD 1.3 -1.7

For Smer, KDH and SNS, this differs little from the previous month’s polling. For SDKU and SMK this result would represent a significant increase over polls (even given SDKU’s final-month increases). For HZDS this would represent a significant drop (and a particular humiliation were the party receive fewer voters than SNS). For KSS it would represent a smaller drop but a much
more significant one since it would put them below the 5% threshold.

I need to double check some calculations regarding the translation of seats into votes (Markiza’s appears to be wrong by several seats, but I need to confirm that) and will post parliament graphs and coalition possibilities when I’ve done the calculations.

Election Day: OMV Numbers

Polls have closed and Pravda is promising preliminary results in a few minutes, but all we’ve got for the moment is their republication of the OMV results based on an enormous (8,000 person) sample.  The results of that survey are not particularly surprising given what we’ve seen elsewhere (see below).  The interesting thing is how well this will correspond to actual results.  Even with the best, latest results possible, some other kinds of calculations may still be necessary.  That is really the point of this blog and the results of this election will provide us with a lot more data for understanding the differences between polls and results.  More on that soon.


































  Percent Change from last
SMER-SD 26.70 (1.60)
ĽS-HZDS 10.30 (0.80)
SMK 10.00 (0.60)
KDH 9.60 (0.20)
SNS 11.80 2.00
SDKÚ 13.60 1.50
SF 4.30 (0.70)
KSS 5.90 1.10
HZD 1.50 0.20
ANO 2.70 (0.10)

Party Preferences and Trends: All the Numbers

For those who did not receive it by email, I am posting here the full set of poll results in graphic form including results for individual parties and individual polling firms along with recent trendlines.  These include all available final polls (but we have no word yet from OMV, FOCUS or Median).  I’ll update and repost if new polls produce major differences, but that’s rather unlikely.

Download slovak_election_polls_final_week.pdf

Two Days to Go: What the Polls Show Now

Final-week polls are in from three of the six firms with
regular political polls (including two of the "big three": UVVM, FOCUS and MVK). 
We find little difference in the relative positions of parties and actually
relatively little actual change (the differences between May and June reflect a
change in only 1 in 15 of all stated preferences).   More interesting
is that the trends of the last month correspond quite closely the predictions
that the turnout model makes for party support.  Specifically, those
parties that have traditionally lost support between polls and actual voting
(especially Smer and SF) have lost support in the last two weeks while those
that traditionally gain more support than polls would indicate (HZDS, KDH, KSS
and especially SDKU) have gained support in the last two weeks.  Only SMK
and SNS do not fit the expected plan, but they are not major exceptions
either, since their changes were among the smallest recorded here, though by no
means insignificant for SNS.

Party Poll Average 1-Month trend, by poll # of Polls showing
UVVM MVK Dicio Avg. Upward Trend Downward Trend
Smer 28.2 -3.5 -3.2 -7.8 -4.8 0 3
HZDS 12.0 2 1.8 1.4 1.7 3 0
SMK 10.0 -0.5 -0.6 0.9 -0.1 1 2
KDH 9.2 -1.3 1.6 0.3 0.2 2 1
SNS 9.3 0.5 0.1 1.2 0.6 3 0
SDKU 12.6 1.3 3.6 2.7 2.5 3 0
SF 6.4 -1.6 -0.8 -1.3 -1.2 0 3
KSS 5.1 1.5 -1.5 1.9 0.6 2 1

The open question for me is whether the change we are
seeing is part of the same process measured in the turnout model (which
would then need some renaming and recalculation) or whether it is something
different.  The turnout model relies on the difference between parties’
final poll and the election results, but one thing has changed: never before
have we received the results of polls taken so close to election day.  In
this small way, we are in a new world.  The turnout model assumes that the
increases and decreases are the result of differences in the degree to which
party voters are committed to going out to vote, but the evidence above might
indicate that an almost identical process has begun for some parties even
without the election-day test of party commitment.  I am therefore not at
all certain whether to apply the ratios used in the turnout model to such late
data since it might produce a double effect.  The turnout model, for
example, assumes a significant drop between Smer polls and Smer vote, but if the
turnout model is actually measuring some of the final week preference shift
rather than just the turnout effect, then applying it to the numbers from the
final week will produce an over-estimation, and I should stick with the results
from the pre-final polls which correspond better to the previous elections from
which the models were derived.  It seems a shame, however, to waste great
data from the final week (which is so far remarkably consistent from one poll to
the next), so I include below the (over)estimate derived from the turnout model,
one with SF and one without.



At one level, however, it is remarkable how little
difference this makes.  Whether with or without the turnout model, there is
little chance of a continuation of the current coalition (in its best case
scenario it gets 71 seats), and therefore little chance for a government that
does not include Smer.  It does make some significant differences in
coalition possibilities, however:  if the final results are as extreme as
the graphs above, Smer loses the ability to form a three party government except
with SDKU and another partner.  This would eliminate two of the
most-talked-about coalition possibilities: Smer-HZDS-SNS and Smer-KDH-SMK. 
Since I suspect a slightly more moderate result, I think that at least one of
these coalitions may yet be on the table, but that is a question for another

Update: Beating the Odds

What are the odds-makers thinking?

Working with the betting data from Tipos led me to think
that it was possible to "reverse engineer" the odds given by the odds-makers to
figure out the assumptions behind their choices.  Assuming that the were
working with the same data I was–recent public opinion polls–I attempted to
calculate the degree to which they expected parties to over- or under-perform
their polling numbers.  In practice, this means estimating the share of
opinion polls which show a party exceeding the threshold set for it by Tipos and
setting the threshold at the point where the share of polls in favor corresponds
with the degree of certainty required by the odds (as determined through the
risk-return ratio).  In other and less pointlessly complicated words, how
many more–or fewer–voters does Tipos think will turn out to allow it to make
money from the odds that it has set (these have changed slightly since I did the
calcuation, but in general the changes are strikingly small).

For the sake of simplicity, I have presented the results
of the reverse-engineering below.  They can be read this way:  For
SDKU the bet against ratio is 1.26 and the bet for ratio is 1.39.  This
means that if you expect fewer than 126 voters to turn out for every 100 who
state a preference for SDKU, it makes sense to bet against SDKU exceeding the
threshold of 13.5 set in the Tipos bet.  If, however, you expect more than
139 to turn out for every 100, then it makes sense to bet that SDKU will exceed
13.5.  It is notable that Tipos appears to be banking on positive ratios
for five parties.  It is also notable that these five parties also have
positive ratios in the Turnout Model.  (This is not surprising, since
despite its Capital Letters the Turnot Model not a very complicated model at
all, and one which relies on on easily discernible past performance.  It is
neither rocket science nor, in fact, much political science–though this does
not explain why the Slovak press continues to report poll results as if they
actually meant something in themselves.)  Only HZDS does not have a high
enough ratio in the Turnout Model to make it worth a positive bet.  In fact
the turnout model shows a ratio that is below the bet against threshold,
suggesting that either I or Tipos have made a big mistake.

Party Ratio of actual voters to poll supporters
party if you think
ratio is lower than:
Bet for
party if you think ratio is
higher than:
Ratio used in Turnout Model
KSS 0.93 1.03 1.38
SMK 1.03 1.09 1.08
KDH 1.02 1.04 1.13
HZDS 1.18 1.22 1.03
SDKU 1.26 1.39 1.42

At the same time, there are three parties that Tipos
appears to think will underperform.  These are to be read the same way:
Smer’s bet against threshold is .85 which means that it makes sense to vote
against Smer exceeding the 27% threshold if you think that only 85 voters will
actually turn out for each 100 who say they will vote for the party.  This
suggests bets for Smer and SNS, with SF’s ratio just on the line.  Here my
own numbers conform nicely with Tipos’s.

Party Ratio of actual voters to poll supporters
party if you think
ratio is lower than:
Bet for
party if you think ratio is
higher than:
Ratio used in Turnout Model
SF 0.74 0.80 0.80
Smer 0.85 0.87 0.80
SNS 0.91 0.97 0.82

This means that Tipos implicitly expects parliament
looks something like the following:

Party Vote Seats
Smer 27.4% 44
HZDS 14.5% 23
SDKU 13.5% 22
SMK 10.9% 18
KDH 9.6% 15
SNS 7.4% 12
KSS 5.1% 8
SF 5.1% 8


Party Vote Seats
Smer 27.4% 47
HZDS 14.5% 25
SDKU 13.5% 12
SMK 10.9% 23
KDH 9.6% 18
SNS 7.4% 16
KSS 5.1% 9
SF 4.9% 0

Why care what Tipos thinks?  No reason, really,
except that somebody there stands to lose a lot of money (and perhaps a job) if
the results are wrong.  If I had that burden, I might do a lot more to make
sure my answers were correct.  Or perhaps not.

Finally, it has been noted that I do not deal with Tipos
bets regarding turnout or the future prime minister.  In the case of
turnout, I have so little data that I hesitate to make a call (though
interestingly the assumptions of the turnout model do themselves incorporate a
turnout figure of 65%.)  I suspect that the turnout will not be below the
57% threshold set by Tipos, but I have no reason to offer the 61% degree of
certainty that it would require to make a bet. 

As for prime minister, this
is a question that takes the speculation beyond what I have described as level 1
(correct assessment of poll numbers), level 2 (correct translation of polls into
seats), level 3 (correct prediction how seats translate into coalitions) to a
level 4 which involves the correct prediction of whom coalitions choose as their
leader.  If my level 2 analysis is right, there is strong reason to think
that the next coalition must include Smer but I have not figured out a good way
of calculating odds for this.  At a gut
level, I would put it above 75%.  The question, then, is whether Fico would
allow somebody else to be Prime Minister?  Here my own bias may show
through:  regardless of any worthy programmatic goals of the party, Fico’s
enterprise in Smer has seemed entirely designed to elevate him to the seat of
power.  Since his dominance within the party is, for the moment, extremely
high, I cannot imagine that he would surrender the opportunity in order to make
a coalition or that others in his party would have the power to force a
compromise candidate for the premiership.  I hope I am proven wrong on
this, but here my doubt goes well beyond the standard required by Tipos. 
On the basis of likely coalitions, it is worth betting that Fico will be the
next prime minister, though I wouldn’t necessarily bet that he will still be
prime minister four years from now.

Four Days To Go: Beating the Odds

Five Days to Go: The MVK Numbers

I will try to deal with polls as they come in.  Smer suggests that the major numbers may come as soon as Thursday but that some may come on Friday.  We should have several data sources to work with by that point.

Until that point, all we have is the latest MVK, conducted between June 2 and June 8 (last Thursday).   The table below compares it to the most MVK poll and the averages of all polls from April and May:

Party Average of Polls MVK Polls
April May June Change from May
Smer 33.6 30.6 30.4 -0.8
HZDS 12.2 10.7 10.1 0.4
SMK 9.6 10.1 9.0 -1.6
KDH 9.0 9.6 10.2 1.3
SNS 8.0 9.0 8.3 -1.6
SDKU 9.3 9.9 11.4 2.5
SF 7.5 6.6 7.5 0.7
KSS 4.5 5.4 4.8 -1.7
HZD 1.9 2.3 1.6 -0.7
ANO 2.8 2.8 2.4 -0.4

The central story here is consistency:  the June numbers differ little from the previous month, either from MVK’s own polls or from the average of all polls.  Of the differences, many are the direction that the turnout-based model predicts: SDKU, KDH and HZDS show gain (greatest for SDKU) in the final weeks, Smer and SNS show a loss.  The other three parties do not follow the predicted pattern: SF increases slightly, KSS decreases slightly and SMK decreases by a somewhat larger amount.  This last change is an important sign that individual polls cannot be taken too seriously (even if they offer support to a model that the author wants to show is valid).  There is no obvious reason why SMK’s support would drop at all, much less by 1.6 percentage points, and so this is clearly an artifact of the polling methodology and the margin of error inherent in any sampling.  But if SMK’s poll numbers change despite fundamental stability of the party’s electoral base, then this could be true of any of the other numbers as well.  For now it’s best to wait for the other polls for a final average.

In the bigger picture, even if the changes can be taken as real, they have very little effect on the formation of a government.  The parliament that would emerge from the MVK numbers is so close to that which would emerge from the previous month’s average numbers as to make little difference.  As before, there almost is no chance for a SDKU-led coalition, and as before Smer will require three parties to form a government (and as before, my turnout model suggests that a Smer-SNS-HZDS government could not muster a majority).   The only real change suggested by the recent MVK numbers is the possible absence of KSS, but MVK chair Haulik immediately nuanced the numbers to suggest that KSS still had a good chance of making it through.

More numbers will probably tell the same story of stability, but I will nevertheless attempt to analyze them as they come in.