Abstract: The relationship between democracy and economic growth is of long-standing interest. We revisit the panel data analysis of this relationship by Acemoglu, Naidu, Restrepo and Robinson (forthcoming) using state of the art econometric methods. We argue that this and lots of other panel data settings in economics are in fact high-dimensional, resulting in principal estimators -- the fixed effects (FE) and Arellano-Bond (AB) estimators -- to be biased to the degree that invalidates statistical inference. We can however remove these biases by using simple analytical and sample-splitting methods, and thereby restore valid statistical inference. We find that the debiased FE and AB estimators produce substantially higher estimates of the long-run effect of democracy on growth, providing even stronger support for the key hypothesis in Acemoglu, Naidu, Restrepo and Robinson (forthcoming). Given the ubiquitous nature of panel data, we conclude that the use of debiased panel data estimators should substantially improve the quality of empirical inference in economics.
The Challenge of High-Dimensional Panel Data
Economists frequently use panel data to understand causal relationships, such as how a change in a country's political system affects its wealth. However, these datasets are inherently "high-dimensional." Even if the primary focus is a single variable like democracy, researchers must account for "fixed effects"—the unique, unobserved characteristics of every country and every year in the sample.
As the number of units ($N$) or time periods ($T$) grows, the number of these "nuisance parameters" also increases. This creates a mathematical bottleneck known as incidental parameter bias. In standard models like Fixed Effects (FE) or Arellano-Bond (AB), this bias is often large enough to shift results significantly and invalidate standard statistical tests. Many foundational findings in empirical economics may be built on these biased foundations, where the bias persists even as more data is added.
Debiasing Strategies
To address the bias inherent in standard estimators, the authors revisit the dynamic linear panel data model and employ two main strategies to restore valid statistical inference:
- Analytical Bias Correction (DFE-A): This method uses complex mathematical derivations to estimate the expected size of the bias based on the model's structure and then subtracts it from the initial estimate.
- Split-Sample Bias Correction (DFE-SS & DAB-SS): This is an intuitive "jackknife" approach. For Fixed Effects, the data is split along the time dimension; for Arellano-Bond, it is split by country. By comparing estimates from the full sample to those from the half-samples, researchers can isolate and remove the bias term.
Data and Summary Statistics
The study utilizes a balanced sub-panel extracted from a landmark dataset by Acemoglu et al. (2019). The sample covers 147 countries over the period from 1987 through 2009. The outcome variable is the logarithm of GDP per capita (in 2000 USD), and the treatment variable is a binary democracy indicator that identifies countries with free elections and inclusive political processes.
Empirical Results
The application of debiasing techniques reveals that standard "off-the-shelf" methods significantly underestimate the impact of democracy. Standard FE estimated that a transition to democracy increases growth by 1.89% in the first year. However, the debiased FE (DFE-A) pushed this up to 2.27%, and the debiased AB reached as high as 5.22%.
The most striking difference appears in the long-run estimates. While standard FE suggested a long-run GDP increase of 16.05%, the debiased estimators (both FE and AB) converged on a much higher figure: approximately 25-26%. These results provide even stronger support for the hypothesis that democracy causes growth, with the findings remaining highly significant even after correcting for the high-dimensional nature of the data.
Conclusion
While debiasing improves the accuracy of point estimates, the authors note a discrepancy between analytical standard errors and those generated by bootstrapping, suggesting that analytical corrections might be slightly over-optimistic about precision in smaller panels. Nevertheless, given the ubiquity of panel data in social science, adopting these debiased estimators should be a priority for improving the quality of empirical research across the field.
Summary updated