Foundations of Event History Analysis in Economic History

Event history analysis (EHA) is a statistical framework originally developed in biostatistics and engineering to model the time until an event of interest occurs. In the context of historical economic event studies, EHA provides historians and economists with a rigorous set of tools to examine the timing, duration, and determinants of phenomena such as financial panics, sovereign defaults, trade regime changes, or the adoption of new technologies. Unlike traditional regression approaches that focus on whether an event happened, EHA centers on when it happened and why that timing varies across cases. This temporal focus is critical for understanding the dynamic processes that drive economic change over historical periods.

The core elements of any event history analysis include the hazard rate—the instantaneous risk of experiencing the event at a given moment, conditional on survival up to that moment—and the survival function, which describes the probability that an individual or entity has not yet experienced the event by a particular time. These functions are estimated using specialized models such as the Cox proportional hazards model, parametric survival models (e.g., Weibull, exponential), or discrete-time logit models when events are measured at intervals. In historical economic studies, data often come in the form of annual, quarterly, or even daily observations of countries, banks, markets, or institutions, making these models particularly valuable.

A key concept that distinguishes EHA from other methods is censoring. In historical data, many observations may not have experienced the event by the end of the study period, or they may be lost to follow-up. EHA explicitly accounts for right-censored observations, preventing the biases that would arise from excluding them or treating them as non-events. For example, when studying the duration of economic expansions in the 19th century, a country that never experienced a recession during the observed timeframe contributes valuable information about survival, and EHA incorporates that information correctly. Similarly, time-varying covariates allow researchers to model how changing economic conditions—such as shifts in monetary policy, tariff levels, or commodity prices—affect the risk of an event over the observation window. This dynamic feature is essential for historical analyses where conditions evolve constantly.

Key Applications of Event History Analysis in Economic History

Financial Crises and Banking Panics

One of the most fruitful applications of EHA in historical economics is the study of financial crises. Researchers have long sought to understand what triggered the banking panics of the National Banking Era in the United States (1863–1914) or the wave of sovereign defaults that swept through Latin America in the 19th and 20th centuries. By applying Cox regression models to panel data of banks or countries, analysts can estimate the impact of reserve ratios, gold flows, harvest shocks, or political instability on the hazard of a panic or default. For instance, a well-known study by Calomiris and Gorton (1991) used discrete-time hazard models to analyze the seasonal patterns of banking panics, revealing that agricultural cycles and information asymmetries played a decisive role. These models allow the inclusion of both time-invariant factors, such as the structure of a banking system, and time-varying indicators like monthly reserve levels.

Business Cycle Duration and Turning Points

Event history analysis is also central to understanding the duration and recurrence of business cycles. Historians of economic fluctuations have used EHA to examine whether recessions in the 19th century were shorter or longer than those in the 20th century, and whether the probability of a recession ending (or beginning) depends on previous cycle characteristics. By modeling the time between troughs or peaks, researchers can test hypotheses about the stabilizing role of central banks, the impact of wars, or the effects of technological innovation on economic volatility. For example, Diebold and Rudebusch (1990) applied duration models to US business cycle data and found evidence of duration dependence—that is, the longer an expansion lasted, the higher the immediate risk of a recession. Such findings have direct implications for understanding historical policy responses and institutional learning.

Technological Adoption and Diffusion

Economic historians have also used EHA to study the spread of innovations. Whether examining the adoption of the steam engine across British textile mills, the diffusion of the automobile in early 20th-century America, or the spread of banking technologies like the automated teller machine, event history models allow researchers to identify the factors that accelerated or delayed adoption among firms, regions, or countries. The dependent variable is the time until adoption, while covariates might include market size, urbanization, access to capital, or prior experience with related technologies. For instance, a study by Kim (2005) used a Cox proportional hazards model to analyze the adoption of the Bessemer process in the American steel industry, finding that patent protection and prior experimentation with alternative methods significantly influenced the timing of adoption.

Institutional Change and Regime Transitions

EHA is well-suited for studying major institutional shifts—such as the adoption of free trade, the introduction of central banking, or the transition from fixed to floating exchange rate regimes. Historical events like the Corn Law repeal in Britain (1846) or the establishment of the Federal Reserve (1913) can be modeled as events whose timing depends on economic and political pressures. Time-series cross-sectional hazard models can incorporate variables such as fiscal deficits, trade balances, or legislative composition to test theories about why reforms happen at certain moments. For example, scholars have applied EHA to analyze the duration of colonial regimes or the timing of decolonization, linking the hazard of independence to economic exploitation, international pressure, and nationalist movements.

Methodological Advantages of Event History Analysis

Beyond its basic suitability for timing questions, EHA offers several specific advantages that make it indispensable for historical economic research.

  • Handling of Censored and Truncated Data: Historical datasets are often incomplete—some episodes end before the study period closes, or begin before it starts. EHA’s ability to incorporate right-censored and left-truncated observations ensures that every piece of available data contributes to the likelihood, maximizing statistical power and reducing bias.
  • Dynamic Modeling of Covariates: Because economic conditions change over time, a static covariate measured only at the beginning of an observation period may miss crucial variation. EHA can incorporate time-varying covariates measured at multiple intervals, allowing researchers to model how a change in interest rates, harvest yields, or political leadership immediately alters the hazard of a financial crisis or an institutional reform.
  • Flexible Functional Forms: Researchers can choose among parametric, semi-parametric, and non-parametric models based on the shape of the baseline hazard. If theory suggests that the risk of an event increases over time (positive duration dependence) or decreases (negative duration dependence), parametric models like the Weibull can test these hypotheses directly.
  • Comparison of Hazard Rates Across Groups: EHA makes it straightforward to compare survival curves across different historical periods, regions, or regime types. Log-rank tests and stratified models allow researchers to ask whether the timing of recessions in gold-standard versus fiat-money eras was statistically different.
  • Integration with Competing Risks: In many historical settings, multiple types of events can occur—for example, a country might default, reschedule, or receive a bailout. Competing risks extensions of EHA enable the analysis of these separate outcomes within a unified framework, giving a richer view of economic decision-making under duress.

Challenges and Limitations in Applying EHA to Historical Data

Despite its power, event history analysis is not without challenges when applied to historical economic data. Researchers must be aware of several common pitfalls.

Data Quality and Availability

Historical data are frequently sparse, measured with error, or available only at coarse time intervals. For example, GDP estimates in the 19th century might be calculated only annually or even decennially, while a banking panic could unfold over days or weeks. Such temporal mismatches can bias hazard estimates, especially if events are driven by short-term fluctuations that are not captured in the data. Similarly, missing covariates—perhaps due to destroyed records or inconsistent accounting standards—can introduce unobserved heterogeneity, potentially leading to omitted variable bias. Advanced techniques like multiple imputation or latent variable models can help, but they add complexity and require strong assumptions about the missing data mechanism.

Model Assumptions and Specification

Assumptions about the baseline hazard shape, proportional hazards, and functional form of covariate effects are often difficult to verify with historical data. The proportional hazards assumption—that the effect of a covariate is constant over time—is frequently violated when, for instance, the impact of a gold discovery on default risk changes as the international monetary system evolves. Researchers can relax this assumption by including interaction terms with time, using stratified models, or turning to flexible semi-parametric approaches, but these solutions demand careful diagnostic testing and theoretical justification.

Unobserved Heterogeneity

Even with broad sets of covariates, unobserved differences across historical units (banks, countries, firms) can bias hazard estimates toward negative duration dependence—making events appear less likely over time simply because the most “frail” units have already failed. Standard survival models can account for unobserved heterogeneity through frailty terms, but identifying such models requires multiple observations per unit or strong distributional assumptions. In historical contexts where each unit may be observed only once (e.g., a single default event per country), frailty models become difficult to estimate reliably.

Interpretation in Historical Context

Statistical results from EHA must always be interpreted within the rich context of historical events. A hazard ratio showing that tariff increases double the risk of a financial crisis does not by itself explain the political economy that led to those tariff increases. The historical narrative, qualitative evidence, and institutional details remain essential for making sense of the numbers. EHA is a tool for description and causal inference under specific conditions, not a substitute for careful historical reasoning.

Case Study: Timing of Sovereign Defaults During the Classical Gold Standard, 1870–1914

To illustrate the power of EHA in historical economic research, consider the analysis of sovereign defaults in the era of the classical gold standard (1870–1914). During this period, many countries adopted gold convertibility, which was supposed to enforce fiscal discipline. Using an event history approach, researchers can model the time from a country’s entry into the gold standard until its first default, or the time between successive defaults.

A typical study might collect data on 30 countries over 44 years, recording the year of each default and a set of covariates: the ratio of public debt to GDP, the level of gold reserves, the openness of the economy (trade-to-GDP ratio), political regime type (democracy vs. autocracy), and the occurrence of wars. Because many countries never defaulted during the period, their observations are right-censored at 1914. A Cox regression estimates the hazard of default, with results showing that higher debt ratios and lower gold reserves significantly increased the hazard, while democracies had a lower hazard of default than autocracies, possibly due to stronger institutions and greater credibility.

Further analysis might test for duration dependence by fitting a Weibull model. If the shape parameter is greater than 1, it indicates that the longer a country maintained its solvency, the higher the immediate risk of default—consistent with a “debt fatigue” argument. Alternatively, a negative duration dependence (shape < 1) might suggest that countries that survived the first few years of gold standard membership became increasingly stable. The EHA framework allows the researcher to disentangle these dynamics and to examine how covariates like a sudden decline in export revenues (a time-varying covariate) affected the hazard in the years leading up to a crisis.

This case demonstrates how EHA transforms historical data into analytically rigorous evidence about the causes of economic events, while still respecting the chronological nature of historical processes.

Conclusion

Event history analysis provides historical economists with a powerful and flexible toolkit for studying the timing and causes of economic events. By focusing on duration and risk, EHA goes beyond binary outcomes and sheds light on the continuous unfolding of economic change. Its ability to incorporate censored data, time-varying covariates, and multiple event types makes it especially valuable for the kinds of datasets that historians work with—incomplete, messy, but rich in temporal variation.

Nevertheless, the method must be applied with care. The quality of historical data, the validity of model assumptions, and the need for contextual interpretation all impose constraints. Used thoughtfully, EHA can complement traditional narrative history and other quantitative methods, such as difference-in-differences or instrumental variables, offering insights that neither approach alone can achieve. As historical datasets grow more detailed and computational tools more accessible, event history analysis will undoubtedly play an even larger role in our understanding of how economies have evolved over time.

For further reading on the methodology, consult this overview of survival analysis or Allison’s foundational text on event history analysis. A classic empirical application can be found in Calomiris and Gorton’s study of banking panics, which demonstrates many of the techniques discussed here.