In September we will be allowed to vote again. Will the upcoming election campaign finally go down in history as the campaign in which social media finally broke through? In this fifth part of the series on the role of social media in the upcoming elections, the answer to the question: can Twitter predict the election results?
Opinion polls
It's election time, so the media are once again flooding us with opinion polls. Every week or even every day, Maurice de Hond , Ipsos Synovate , TNS NIPO and EenVandaag come up with their own polls on the current political preferences of the Dutch. The opinion polls provide an image of the shifts in voter preferences, but in absolute terms they are often off, by as much as 15 seats.
There can also be major differences between the polls conducted by the agencies: for example, at the end of June, the VVD was the largest party according to Maurice de Hond, while the SP was clearly number one according to TNS Nipo.
Market research agencies work with panels of several thousand members. They try to involve as representative a sample of society as possible, but they only succeed to a certain extent. The large number of floating voters does not make polling any easier. The question is whether it can be done differently.
In recent years, social media have become a prominent place for discussing kuwait telegram data political issues and for politically interested people to meet. This produces enormous amounts of data, which are public and searchable. Previously, analyses of large amounts of Twitter messages have been used to predict stock prices, flu outbreaks and box office hits. Is it possible to tap into the same 'wisdom of crowds' to predict election results?
Both Frankwatching (the Twitter Monitor in collaboration with DamKam) and Marketingfacts (in collaboration with Coosto) are investigating political developments through analyses of social media in the run-up to the elections. Some results are encouraging. For example, Marketingfacts predicted the outcome of the election for the CDA party leader in May this year quite accurately: Sybrand van Haersma Buma would win, with a majority of 50%, followed by Mona Keijzer. This prediction came true. In 2010, the results of the midterm elections in America were accurately predicted using analyses based on Facebook and Twitter.
If we look at the experimental polls of Buzzpeil , we see small differences with the poll of Maurice de Hond , which was released on the same day. According to Buzzpeil, the VVD is at 33 seats, according to De Hond at 29. Buzzpeil puts the SP at 32 seats, De Hond at one seat less. According to Buzzpeil, the PVV is at 18 seats, according to De Hond at 20. Buzzpeil puts the PvdA at 20 seats, De Hond thinks one seat less. The CDA is at 14 seats in both polls. The differences are therefore relatively small, perhaps with the exception of the VVD. Does this mean that we can rely on social media?
Debate about reliability
Several scholars have been investigating for some time whether social media data can be used for political polls. Researchers from the Technical University of Munich, Andranik Tumasjan and others, conducted research into the 2009 German parliamentary elections and analyzed more than 100,000 messages on Twitter. The question was what the relationship was between tweets about political parties and the final election results. Their results suggest that social media can be an important predictor of election results.
The percentage of tweets mentioning a political party appears to be very similar to the percentage of votes the party ultimately received. 30.1% of tweets were about Angela Merkel's CDU, while the party received 29.0% of the votes. The liberal FDP received 17.3% of tweets and 15.5% of the votes. The largest difference between tweets and votes was for the Greens, who received 3.3% more votes than tweets. These scores are comparable to those of regular opinion polls, which are always slightly off the real results.
However, this does not yet provide a convincing instrument that can truly predict the elections. The German researchers show that there are indeed similarities between social media and election results, but the question is over which period the tweets should be examined: how long before the elections should this period start and which parties should be included in the analysis?
Jungherr and his colleagues also show that tweets about the German elections can also give completely different results. If the relatively successful Pirate Party in Germany is added to the analysis, it turns out that the election results are no longer predicted correctly at all. But after adding the Pirate Party to the analysis, 34.8% of the tweets are suddenly about this party, which has no relation to the election results. Such a large amount of messages from one party can strongly distort the results of the analyses. So there are still some snags in the social media analyses when it comes to predicting the election results that make social media perhaps not a good way to predict election results .
A new instrument?
Social media analyses also suffer from the fact that there is no representative sample of the population . Highly educated and politically interested men are overrepresented. Opinion polls by market research agencies are more representative due to careful composition of the panels. This is consciously taken into account when weighing the results.
However, social media do seem to be good predictors in certain cases. Parties or candidates that are often mentioned in social media usually also get more votes. This comparison is dangerous, because hypes can have a big influence on the results in social media, while they do not have to be related to voter preferences. The example of the German Pirate Party clearly shows this.
So there is still plenty to research about analyses via social media and the possibilities to predict with them. Becoming the largest party in September, that is of course what it is all about now. But let us not forget that social media contain a lot of qualitative information about how people in society think about political proposals and social themes. And that is at least as valuable.