Detecting institutional change as it unfolds: Evidence from 120 years of reporting

Explaining comparative economic development requires identifying the causes of institutional change. This column uses AI to analyse 120 years of newspaper coverage and produce a new index of uncertainty about a country’s core institutions. In the countries studied, nearly every major shift in democracy happened when this index was high. Because it detects slow-building uncertainty that other risk measures miss, the index could serve as an early warning of institutional fragility.

Institutions – the rules of the game – are central to understanding why some countries are rich and others remain poor (North et al. 2009, Acemoglu and Robinson 2012). Explaining comparative economic development therefore requires identifying the causes of institutional change. Social scientists have long pointed to ‘critical junctures’ – moments when a country’s institutional path can shift dramatically (Collier and Collier 1991, Capoccia and Kelemen 2007) – as points at which institutions, and so economic trajectories, may potentially diverge.

Yet we lack a reliable means of empirically identifying critical junctures. Our research (Callen et al. 2026) seeks to make progress by measuring uncertainty about the future of a country’s core institutions. Such uncertainty should be reflected both in people’s beliefs about future institutions – for example, whether legal constraints on actions taken by the head of state will be enforced – and in contemporary news reporting. Using a large language model to read 120 years of the Times of London, we turn that reporting into a Critical Juncture index: a systematic measure of institutional uncertainty.

Such a measure also has policy relevance. Governments and international organisations devote enormous effort to anticipating and understanding instability, but existing measures of uncertainty focus primarily on economic policy and geopolitical risk rather than the possibility that a country’s core institutions may change. Our Critical Juncture index may provide an earlier signal that a country’s core institutions could change. Many plausible sources of institutional uncertainty today – the accelerating climate crisis, rapid rise of artificial intelligence, and shifting global power – do not map neatly onto any single crisis event, which is why a measure that can pick up on slow-building uncertainty is useful. This uncertainty also matters for behaviour: evidence from Afghanistan, for example, shows that take-up of a major state-building reform was substantially higher where people believed the government would endure (Blumenstock et al. forthcoming).

Newspapers capture institutional uncertainty in real time

We use archival newspaper text to capture how uncertainty about a country’s institutional future is expressed in contemporary discourse, before anyone knows how events will resolve. An advantage of large language models over traditional text analysis is their ability to interpret context and infer whether a particular piece of discourse reflects uncertainty about core institutions, rather than simply identifying references to political events or uncertainty.

To operationalise this idea, we fed a large language model millions of newspaper articles from the Times of London’s full digital archive, spanning 1800–2019, with a focus on 1900–2019. We developed an optimised classification prompt using examples drawn from historical events covering various types of critical junctures and evaluated it against a benchmark spanning 30 events. This set included ‘near misses’ and high-profile events that did not ultimately change core institutions. In validation analysis, the classifier agrees with human labels 85.6% of the time on our benchmark dataset and 69.7% of the time on a larger sample of nearly 2,000 Times of London articles.

Our index measures the share of articles classified as critical juncture articles over time. To identify unusually large increases in institutional uncertainty, we also compare each year’s share with the level predicted by the index’s values over the preceding three years. The 15 largest positive deviations for each country are identified as critical juncture event years and marked by vertical dashed lines in Figures 2 and 3.

We chose the Times of London for practical and substantive reasons. The newspaper was privately owned over the course of the study and has been published almost without interruption for well over a century. Its reporting extended far beyond British borders, reflecting Britain’s historical global reach, and its digitisation quality was strong enough to support analysis at this scale. Just as importantly, its core readership during the study period consisted of political and economic elites whose beliefs the index aims to capture. Because coverage is uneven across countries, we focus on the four the Times covers most: the UK, the US, France, and India.

The AI classifier we use here is a first attempt, not a finished instrument. As language models and classification methods improve, so should the measure of institutional uncertainty.

Measuring the possibility of institutional change, not change itself

Scholars often identify critical junctures retrospectively by looking for periods when institutions actually change (Collier and Collier 1991, Pierson 2000). We argue this methodology misses crucial ‘near misses’, where institutional change was possible but the status quo was maintained (Capoccia and Kelemen 2007, King et al. 2021). Without also examining these moments, we risk drawing the wrong conclusions about the underlying drivers of institutional change. A retrospective search also precludes studying critical junctures in real time.

By focusing on uncertainty about future institutions, our measure makes progress on both issues. Because the uncertainty we capture does not depend on whether institutions ultimately change, we can identify critical junctures without selecting episodes based on their outcomes. Moreover, we can study the evolution of institutional uncertainty in real time, whether through newspaper data or direct surveys of beliefs embedded in experiments (Callen et al. 2024).

Does the Critical Juncture index identify new forms of uncertainty?

An important test of a new measure is whether it tells policymakers something the tools they already have do not. Comparing our index to both the Economic Policy Uncertainty index and the Geopolitical Risk index (Baker et al. 2016, Caldara and Iacoviello 2022) – both valuable and widely used tools in their own right – we find that all three indices align during major global crises but diverge when institutional uncertainty builds in the absence of overt crisis.

As Figure 1 shows, the Critical Juncture index is elevated in the years leading up to WWI, whereas the Geopolitical Risk index does not rise until the assassination of Archduke Franz Ferdinand. This suggests the Critical Juncture index picked up tensions already present in contemporaneous discourse, signalling the potential for a major rupture before it came.

Figure 1 Institutional uncertainty (Critical Juncture index) versus geopolitical risk, UK, 1900–2019

Notes: Both series smoothed (12-month moving average) and normalised to [0,1].

Moreover, after the mid-1960s, and especially after 2008, the Critical Juncture index diverges from other indices, rising while the Geopolitical Risk index stays broadly flat. By 2019, the end of our sample, the share of UK news coverage reflecting institutional uncertainty was as high as in the early 1940s – a signal other leading risk measures miss. Some of this recent rise may reflect changes in news coverage, which our current measure cannot fully separate from changes in institutional uncertainty. Yet the index also exhibits several sharp increases relative to the level predicted by its behaviour over the preceding three years, which are difficult to explain by gradual changes in media coverage and provides further evidence of heightened contemporary institutional uncertainty.

How do critical junctures relate to institutional change and economic growth?

The Critical Juncture index makes it possible to look systematically at whether historical episodes fit the theory that critical junctures shape a country’s institutional path (North et al. 2009, Acemoglu and Robinson 2012). Using the V-Dem polyarchy index (Coppedge et al. 2023), a well-established measure of electoral democracy, we find a suggestive pattern across our four countries (Figure 2). Many periods of high institutional uncertainty pass without any change to a country’s democratic institutions at all, but nearly every major shift in a country’s democracy score occurs around a time the index flags as a critical juncture. In other words, critical junctures do not necessitate change, but in our data, major change rarely occurs outside of them.

Figure 2 V-Dem polyarchy scores for India, the UK, the US, and France, with critical juncture event years (dashed)

Furthermore, the idea that institutions are a central driver of economic growth is well established in development economics, but testing the basic proposition that critical junctures lead to changes in growth trajectories requires measuring critical junctures. We take a preliminarily look at this by tracking how critical junctures align with shifts in a country’s long-run growth trajectory, following GDP per capita over time alongside the years the Critical Juncture index flags as having the largest increases in institutional uncertainty.

As shown in Figure 3, several of the clearest breaks in growth trajectory – India at independence, and France and the US after World War II – align closely with flagged critical junctures. 

Figure 3 Log GDP per capita for India, the UK, the US, and France, with critical juncture event years (dashed vertical lines), 1900–2020

Beyond its potential as an early-warning system for emerging institutional fragility, the index is, in our view, a starting point for deeper conceptual and empirical work. For example, some critical junctures stem from unanticipated shocks, while others are visible to the actors involved as they unfold. In the latter case, beliefs may not only signal institutional change but also help cause it – a distinction that could shed light on the causal drivers of institutional change. Validating the index against survey data on beliefs is another natural next step, while extending beyond a single British newspaper would reduce bias towards events that mattered to British readers. Finally, such a measure could allow us to compare the importance of critical junctures with the slower-moving process of ‘institutional drift’ in shaping institutional development. More broadly, systematically measuring critical junctures may help advance empirical inquiry into a range of questions about when, why, and how institutions – and, in turn, economic trajectories – change.

Source : VOXeu