[Paper Review] Non-extensive diffusion entropy analysis: non-stationarity in teen birth phenomena

Nicola Scafetta, Paolo Grigolini+2 more|arXiv (Cornell University)|May 24, 2002

Statistical Mechanics and Entropy4 references3 citations

TL;DR

This paper introduces a non-extensive diffusion entropy analysis (DEA) using Tsallis q-entropy to detect non-stationarity in time series, applying it to teen birth data in Texas (1994–1998). It finds that unmarried teen births exhibit strong memory effects (q = 1.257), indicating social process dependence, while married teen births are closer to random (q ≈ 1), with wavelet analysis linking non-stationarity to school calendar effects.

ABSTRACT

A complex process is often a balance between non-stationary and stationary components. We show how the non-extensive Tsallis q-entropy indicator may be interpreted as a measure of non-stationarity in time series. This is done by applying the non-extensive entropy formalism to the Diffusion Entropy Analysis (DEA). We apply the analysis to the study of the teen birth phenomenon. We find that the unmarried teen births are strongly influenced by social processes with memory. This memory is related to the strength of the non-stationary component of the signal and is more intense than that in the married teen time series. By using the wavelet multiresolution analysis we attempt to give a social interpretation of this effect.

Motivation & Objective

• To develop a method to quantify non-stationarity in complex time series using non-extensive entropy formalism.
• To apply this method to teen birth data to detect memory effects in unmarried versus married teens.
• To interpret the origin of non-stationarity through wavelet multiresolution analysis.
• To challenge the assumption of stationarity in social science time series analysis.

Proposed method

• Adopt non-extensive Tsallis q-entropy as a measure of non-stationarity in diffusion processes.
• Apply diffusion entropy analysis (DEA) to time series by generating diffusion trajectories x^(z)(t) = Σξi+z over time.
• Use the scaling property p(x,t) = t^(-δ(t)) F(x/t^δ(t)) to model non-stationary behavior with time-varying δ(t) = δ₀ + η ln(t).
• Evaluate the Tsallis entropy S_q(t) = [1 - Σp_i(t)^q]/(q-1) and identify the 'magic' q = Q that linearizes S_q(t) vs. ln(t).
• Detrend the data using a fit Ξ(t) = A + Bt + C cos(ωt) + D sin(ωt) to remove linear and annual cycles.
• Use wavelet multiresolution analysis to estimate conception rates and link non-stationarity to social structures like school calendars.

Experimental results

Research questions

• RQ1 How can the Tsallis q-entropy be interpreted as a measure of non-stationarity in time series?
• RQ2 What is the strength and nature of memory in unmarried versus married teen birth time series?
• RQ3 How do social factors such as school holidays influence conception patterns in unmarried teens?
• RQ4 To what extent do linear and cyclical trends mask underlying non-stationary dynamics in teen birth data?
• RQ5 Can wavelet analysis reveal the social mechanisms behind non-stationarity detected by non-extensive DEA?

Key findings

• The unmarried teen birth time series exhibits a magic q-value of Q = 1.257, indicating strong non-stationarity and memory effects.
• The married teen birth series has Q ≈ 1, suggesting it behaves like a stationary, memoryless process.
• Wavelet analysis reveals that conception rates in unmarried teens drop sharply during summer months, likely due to school closures.
• The non-stationarity in unmarried teen data is primarily linked to social calendar effects, such as school holidays.
• The non-extensive DEA framework successfully detects memory strength via deviation of q from 1, offering a new tool for analyzing complex, non-equilibrium systems.
• Detrending with linear and annual components (A, B, C, D) effectively isolates residual memory effects for analysis.