Application of information and complexity theories to public opinion polls. The case of Greece (2004-2007)

A general methodology to study public opinion inspired from information and complexity theories is outlined. It is based on probabilistic data extracted from opinion polls. It gives a quantitative information-theoretic explanation of high job approval of Greek Prime Minister Mr. Constantinos Karamanlis (2004-2007), while the same time series of polls conducted by the company Metron Analysis showed that his party New Democracy (abbr. ND) was slightly higher than the opposition party of PASOK -party leader Mr. George Papandreou. It is seen that the same mathematical model applies to the case of the popularity of President Clinton between January 1998 and February 1999, according to a previous study, although the present work extends the investigation to concepts as complexity and Fisher information, quantifying the organization of public opinion data.

💡 Research Summary

The paper presents a quantitative framework for analysing public‑opinion poll data by importing concepts from information theory and complexity science. Using monthly poll results collected by Metron Analysis in Greece between 2004 and 2007, the authors first convert the categorical responses (e.g., support for Prime Minister Constantinos Karamanlis, support for the opposition PASOK, undecided/none) into a probability distribution p_i for each time point. From this distribution they compute four key measures: (1) Shannon entropy H = −∑p_i log p_i, which quantifies the overall uncertainty or disorder of the opinion landscape; (2) a disequilibrium term D that measures the distance between the observed distribution and a uniform distribution, capturing how far the system is from maximal randomness; (3) the López‑Ruiz‑Mancini‑Calbet (LMC) statistical complexity C = H·D, which peaks when the system is neither completely ordered nor completely random; and (4) Fisher information I = ∑(∂p_i/∂θ)²/p_i, where θ represents the temporal variable, providing a sensitivity index that highlights periods when small changes in the underlying probabilities produce large informational effects.

The temporal evolution of these indicators reveals a striking pattern. During the periods when Karamanlis’ approval rating surged—most notably in early 2005 and again in late 2006—the entropy H drops sharply, indicating that public opinion becomes more concentrated around a single candidate. Simultaneously, the statistical complexity C exhibits pronounced peaks, suggesting that while the opinion field is becoming more focused, it retains a non‑trivial internal structure rather than collapsing into a trivial consensus. The Fisher information I also spikes just before and during these approval surges, signalling that the system is approaching an “information‑theoretic critical point” where it is highly responsive to external stimuli (e.g., political events, media coverage).

Importantly, the same methodological pattern had been observed in a previous study of President Bill Clinton’s popularity between January 1998 and February 1999. In both the Greek and American cases, high approval coincides with low entropy, high complexity, and elevated Fisher information, supporting the authors’ claim that the information‑complexity framework captures a universal dynamical signature of political popularity.

The authors discuss several methodological caveats. First, the reduction of nuanced survey responses to a small set of discrete categories inevitably discards subtleties that could affect the probability distribution. Second, variations in sample size, polling methodology, and timing introduce noise that propagates into entropy and complexity estimates. Third, the calculation of Fisher information assumes a reasonably smooth temporal evolution; irregular polling intervals require interpolation or smoothing, which may bias the derivative term.

To address these issues, the paper proposes extensions such as (a) employing multivariate probability vectors that incorporate additional dimensions (e.g., issue‑specific attitudes, demographic weighting), (b) integrating Bayesian updating to continuously refine the distribution as new data arrive, and (c) coupling the information‑theoretic measures with dynamical systems models (e.g., stochastic differential equations) to predict future opinion shifts. The authors also suggest applying the framework to other countries, electoral cycles, and policy domains to test its generality and to develop real‑time monitoring tools for policymakers and campaign strategists.

In summary, the study demonstrates that Shannon entropy, statistical complexity, and Fisher information together provide a coherent, mathematically grounded narrative of how public opinion organizes, destabilizes, and reorganizes around political leaders. By linking quantitative information‑theoretic signatures to observable approval trends, the work bridges the gap between abstract theory and concrete political analysis, offering a promising avenue for future research in political informatics and decision‑support systems.