Abstract
Sometimes the normal course of events is disrupted by a particularly swift and profound change. Historians have often referred to such changes as “revolutions”, and, though they have identified many of them, they have rarely supported their claims with statistical evidence. Here, we present a method to identify revolutions based on a measure of multivariate rate of change called Foote novelty. We define revolutions as those periods of time when the value of this measure is, by a non-parametric test, shown to significantly exceed the background rate. Our method also identifies conservative periods when the rate of change is unusually low. We apply it to several quantitative data sets that capture long-term political, social and cultural changes and, in some of them, identify revolutions — both well known and not. Our method is general and can be applied to any phenomenon captured by multivariate time series data of sufficient quality.
Introduction
The term "revolution" is ubiquitous in history, science, and sociology—from the Industrial Revolution to Kuhnian paradigm shifts. However, these labels are often applied subjectively, leading to debates where one scholar's "revolution" is another's "gradual transition." To move beyond qualitative assessment, there is a need for a quantitative baseline to measure multivariate rates of change and determine if a specific window of time is statistically distinct from its surroundings.
Quantifying Change: The Foote Novelty Method
The proposed "revolution detector" adapts Foote novelty, a metric originally designed for segmenting audio files, to multivariate time series data. The process involves three primary steps:
- Temporal Distance Matrix: For a multivariate time series, a distance matrix is calculated between every pair of time points, creating a heatmap of similarity over time.
- Checkerboard Kernel: A "kernel"—a small square matrix with a checkerboard pattern of +1 and -1 values—is slid along the diagonal of the distance matrix. This kernel compares the variability within two adjacent time blocks to the variability between them.
- Novelty Score ($F_t^k$): A high score indicates a "checkerboard" pattern, signifying a period where the system was stable, changed rapidly, and then stabilized in a new state.
To ensure these peaks are not random fluctuations, a non-parametric permutation test is used to identify periods where the rate of change significantly exceeds the background rate.
The "Microscope" Effect and Scale
A key insight of this framework is the role of the kernel half-width ($k$). The choice of $k$ acts like a lens: small values detect short-term, "jittery" revolutions, while larger values reveal long-term, structural shifts. This allows researchers to identify revolutions at different temporal scales.
The method also identifies "conservative" periods—times of unusually low change or stasis. Furthermore, the authors emphasize that for "random walk" data, using first-differences (the change from $t$ to $t-1$) is crucial to avoid false positives from long-term drifting trends.
Empirical Results across Domains
The versatility of the method was demonstrated across several diverse datasets:
- Democracy (V-Dem): Identified four distinct waves of global democratic change, including the post-WWII era (1944–1949) and the fall of the Iron Curtain (1989–1996).
- Music (Billboard Hot 100): Confirmed major revolutions in American pop: the rise of rock (1968), the synth era (1982), and the emergence of hip-hop (1988–1991).
- Crime Rates: Pinpointed "The Great Disruption" (1965–1978) and the subsequent "crime drop" of the 1990s.
- Cultural Trends: Found that trends in US baby names move in revolutionary bursts rather than smooth transitions.
Conclusion and Limitations
While powerful, the method requires high-quality, sufficiently long multivariate time series. It distinguishes between revolutions that change the mean of variables and those that change their relationships (covariance), though further work is needed to automate this distinction. Crucially, the tool identifies when a revolution occurred, serving as a diagnostic instrument for historians to investigate the underlying why.
Summary updated