The recently revamped governing coalition — now minus the Social Democrats — has set its sights and goals. At the end of May, the Reform Party and Eesti 200 unveiled a coalition agreement that is noteworthy in several respects. To begin with, the agreement in its final form was concluded after the coalition had already begun work, and its length and drafting process are also significant.

In terms of content, the document resembles an election program. It is essentially a long list of highly specific but often vaguely worded goals and actions. So why not analyze it the same way we would an election platform?

One standout feature of the coalition agreement is its emphasis on automated decision-making mechanisms and artificial intelligence. In both governance and the study of governance, there is no avoiding AI and machine learning in the future. In fact, even today, these technologies can assist in analyzing the political content of a document — like this very coalition agreement.

The most well-known and widely used election program dataset in political science contains nearly 5,000 election platforms from over 1,400 parties. Researchers have classified more than three million specific political statements from these programs into 56 policy categories. This is the result of decades of work by hundreds of scholars.

In the age of artificial intelligence, efforts have also been made to train machines to classify texts into these policy categories based on the already-categorized data from election programs. One such model, "xlm-roberta-large," has been used to create a classification model called "manifestoberta." Among other capabilities, this model can classify any political statement in Estonian into one of those 56 categories. Naturally, its accuracy is far from perfect, but it's sufficient to provide analytical value — especially in forming a general overview.

So what does this model say about our government's new coalition agreement?

Before feeding all the positions from the agreement into the model, the raw text needs to be cleaned up a bit. For example, all deadlines associated with specific actions should be excluded from the analysis. By removing sentences shorter than 35 characters, we arrive at a final dataset of around 800 sentences containing various political goals and positions.

What policy areas stand out? And where might this agreement fall on the left-right political spectrum?

Let's start with the broadest view. In the classification scheme used for election programs, 13 of the 56 policy areas are designated as right-leaning and 13 as left-leaning. The difference in overall proportions between these two groups is commonly used to place a party's program on the left-right spectrum.

So, how does this new coalition agreement measure up? About 32 percent of the agreement's content consists of positions classified as right-leaning, while 29 percent fall on the left. More than half of the agreement lies within left-right domains, and they are quite evenly balanced. Therefore, it cannot be said to be a clearly right-leaning or left-leaning agreement. Rather, it is a mixture of policies typically associated with both sides.

What exactly makes up these poles? On the right, three policy areas stand out in particular.

The most frequent category is the positive mention of military-related matters, which is the second most frequently cited policy area overall in the agreement. This category accounts for 11.7 percent.

Next is the emphasis on maintaining domestic order, the second most common right-leaning policy area at 7.5 percent. Third are positions related to state support for businesses (6.5 percent). In short, this government's right-leaning agenda is defined primarily through issues of internal and external security. Should security — especially external security — be considered a right-wing issue in the Estonian context? That's a separate question entirely. For this analysis, we stick with how the topic is categorized in the methodology used to classify election programs.

And what makes up the agreement's left-wing pole? The most frequent category here is the expansion and development of the education sector by the state, which is also the most cited policy area in the entire agreement.

This category accounts for 12.7 percent of the agreement. The next dominant left-wing policy area is the expansion of the welfare state, covering just over 10 percent. Other left-leaning policies — including those related to the economy or entrepreneurship — do not appear in the agreement, at least according to the classification model.

So, what can artificial intelligence tell us about this coalition agreement at the most general level?

Perhaps most importantly, it shows that this is clearly a state-centric policy platform. Regardless of the ideological direction — and even setting aside the debatable left-right nature of some policies — the agreement underscores the direct role and importance of the state across various sectors, whether in ensuring security, providing education and welfare services or supporting businesses.

In addition to the aforementioned, another frequently occurring policy area is the development of technology and infrastructure (11.4 percent), which isn't clearly left- or right-wing. The same goes for policies aimed at improving governance efficiency (7.4 percent). The role and responsibility of free citizens, as well as the principle of a limited state, are virtually absent from the agreement, according to the model. And economic themes tend to appear primarily in the context of state assistance, not the invisible hand of the free market.

For that reason, this coalition agreement can be described as state-centered. Against this backdrop, it's no surprise that even in a so-called era of austerity, expenses are more likely to grow than shrink and the administrative apparatus struggles to downsize. The state, it seems, looks first and foremost toward itself — and more generously than critically.

--

Follow ERR News on Facebook, Bluesky and X and never miss an update!