For once I’ll get the lead on the Slovak press and talk about a poll first, the results of which I have added to the Dashboard. Median published its January poll results today and they have their effects on the overall averages, pushing the MKP-SMK average below 5% for the first time in my dataset, while flattening out the sharp rises of Most-Hid and SaS and softening the drop of HZDS and SNS. We shall see how the press covers it (“CSAKY, BUGAR AND SULIK FALL SHORT!”) but in the context of other polls, its’ rather hard to accept. There are lots of ways to analyze this but for the sake of time and simplicity I’ll simply compare Median to the other three polls taken at approximately the same time (MVK, FOCUS, Polis). On the two biggest parties, Smer and SDKU, the Median result is right in line but on all the rest it stands out to a remarkable degree. Long-established Slovak parties (SNS, HZDS and KDH) score high in Median’s poll. Very high. Meanwhile newly established parties (and SMK) score low. Very low. In fact, for all six of these parties, the Median poll result is not only the outlier, but its addition more than doubles the range of poll values or more (x3 for HZDS, x4 for SAS and x6 for SNS). In other words Median results are more different from the nearest of the three other polls than any of those polls are from one another, as the graph below shows.
|Where Median polls high||Where Median polls low|
|Highest of other 3 polls||6.2||5.8||10||6.7||6.4||9.6|
|Lowest of other 3 polls||5.6||4.1||8.9||5.1||5.6||8.1|
|Largest difference between other 3 polls||0.6||1.7||1.1||1.6||0.8||1.5|
|Difference between Median and nearest of other 3 polls||3.4||3.5||1.3||1.8||1.3||5.4|
As I’ve discussed before, this may have a lot to do with Median’s unorthodox procedure of not providing a list of parties to choose from. But what to do with these results? I’m loathe to eliminate them as I suspect they capture some kind of political truth (some people when shown names of new parties in polls pick them but revert to long standing loyalties and remembered names in the voting booth) but they do ignore another (the inability of people to remember the name of the new party they’re interested in). And the firm’s past performance suggests that it’s take on the political truth does not much help us predict final results (http://www.pozorblog.com/?p=684). Median may thus act as a useful corrective but simply cannot be looked at in isolation.