Election Prediction Using Google Trends

In this chapter, we analyzed various elections using Google Trends, and before we explain the process, let’s first find out what Google Trend is 먹튀검증.

Google Trends

Google Trendswww.google.com/trends is a search service provided by Google that is accessible to all Internet users around the world. The Google trend provides a graph of the amount of searches for a particular search term, not only in individual countries, but also in certain areas or cities of the country. It also has the advantage of separating and searching individual searches on the Web, image search, Google Shopping and YouTube, along with information on various categories such as health, games, science, finance, news, books and literature, beauty and fitness. The term of the search term provides data on a weekly basis from January 2004 to the present, and because the surveyor can arbitrarily adjust the period, the search volume by date can also be measured over a certain period of time. In addition, the related words of that particular search term can be inferred from the fact that it is linked to another search term.

Google Trends provides search volume data in CSVComma Separated Value file format, so it has the advantage of using this data to perform various analyses. However, the disadvantage of caution is that the number of searches will vary as the search period changes, since it assumes that the highest number of searches is 100 after the number of searches is standardized, and the remaining searches are provided on a 0 to 100 scale basis, rather than the actual number of searches over a certain period of time period. For example, if you assume that a particular search term has a daily search rate of up to 50,000 times in a year and at least 500 times, Google Trends shows a maximum search rate of 50,000 as 100 and a minimum search rate of 500 as 1. If there is no search, it is indicated as zero. Since the data is CSV format, it has disadvantages of breaking Hangul. When used in Excel, it is convenient to save the data as ‘Excel Integrated Document’.

Google Trends to Change the Framework of Election Forecasts

So how can we predict elections using Google Trends? And how will the frequency of searches for a particular candidate lead to the number of votes cast in an election? The author believes that there is a very close relationship between the two. Based on existing polling methods, it is impossible to examine the voting propensity of all voters, so they go through a sampling process of choosing samples from the population. However, various problems arise in the sampling process, which leads to incorrect predictions of the election results. So, on what basis can big data provided by Google Trend change the framework of election forecasting? Above all, analysis using big data makes it unnecessary to have algorithms that estimate populations with existing samples, because big data itself can represent populations.

Although exact figures are not known, the fast spread of mobile phones is thought to be causing most people to surf the Internet via mobile or do various social networking sites such as Facebook, Twitter and Instagram. These days, even for those who are a little older, you can often see the sights of using mobile phones to access various information or news. However, if you carefully analyze what Internet or mobile phone users posted on their search lists or social networking sites, you will be able to infer the political tendencies they have hidden in their minds, whether they intended or not. In other words, if the traditional method of opinion polls is to ask the selected sample group directly by telephone which candidate the surveyor will choose, then the big data method is to predict their votes backwards by analyzing the enormous amount of evidence left on the Internet or SNS by potential voters close to the population.

So, is it possible to read voters’ votes with just the amount of search provided by Google Trends? Let’s assume this: If a voter is interested in a particular candidate, he or she will naturally search the candidate first on the Internet or on mobile phones, and there is a high chance that he or she will vote for a favorite candidate after forming likes or dislikes through information about the candidate. Unfortunately, Google Trends doesn’t even provide information on likes or dislikes, so it’s hard to know the exact percentage of them, but considering the fact that they typically search more candidates of interest, a large search volume can soon be thought of as a high percentage of likes.

The criteria for determining favor and dislike for a candidate in Google search can be predicted with search patterns, although positive and negative words are sometimes analyzed individually. In the case of disfavor, there is generally no search volume at all, but soon fade away after a period of rapid growth. If a bad issue occurs because it was not good enough to begin with, the peak will suddenly occur and disappear. For the most part, there are not many voters who constantly search for unfriendly candidates. In fact, a worse case than an unfavorable one in an election is to show no interest in the candidate itself, and the chances of winning votes from voters are significantly reduced because there is no search itself, regardless of good or bad issues. When the author mentioned in the lecture that Trump’s search volume was unusual since early 2016, some students asked if the search volume had increased due to unfriendly interests such as Trump’s misbehavior and remarks. However, he replied, “As Trump’s pattern has shown a steady trend in the long run, it seems that it is not just a temporary popularity that is searched in an unfavorable.” In the case of Trump, he showed a favorable-looking pattern of responding to certain issues, whether positive or negative, at a time when there is a steady amount of search.

The preceding graph shows that there has already been a search fixture in support of Trump’s candidate since January, but the search rate below the red dotted line is not falling, and the dotted line does not appear in the Google and is the line suggested by the author. And while it’s unclear whether it’s likeable or unfriendly, it’s starting to show the peak red arrows that occur on a particular issue. In other words, if Trump had been an unfriendly candidate from the start, he should have shown a sudden peak and disappeared without the search fixed base itself.

The next point to note is that the use of big data like Google Trends can reduce respondents’ bias toward social desirability. In general, people tend not to want to show their innermost feelings to others. Even if you answer, you often try to wrap yourself in a socially desirable situation. In the election, for example, Trump has been embroiled in a controversy over racist speeches and a scandal involving women’s disparaging and sexual harassment, which is expected to have led many voters who support him inwardly but do not support him openly. It would have been all the more so if there were friends or co-workers around, women who hated Trump for his immigration-turned-Hispanic or sex scandal. This is because you can be branded as a racist or a person who demeans women. In the 2016 U.S. presidential election, voters who do not express their opinions outside of the public were named “Shy Trump,” who created a huge upset in the presidential election. In other words, the Bradley effect of expressing one’s true feelings in the actual voting room, although he did not show his own vote outside.