[Paper Review] Power law distribution in High School Education: Effect of Economical, Teaching and study conditions

Hari M. Gupta, José R. Campanha+1 more|arXiv (Cornell University)|Jan 27, 2003

Education Methods and Practices2 references3 citations

TL;DR

This study analyzes university entrance exam scores in Brazil's UNESP to show that physical and biological sciences exhibit long power-law tails in student performance, while humanities follow a normal distribution. Using gradually truncated power-law distributions, the authors attribute the power-law behavior to subject interdependence and find that study conditions, teaching quality, and income level significantly boost performance—especially in science, where better conditions yield up to 140% higher average scores.

ABSTRACT

We studied the statistical distribution of student's performance, which is measured through their marks, in university entrance examination (Vestibular) of UNESP (Universidade Estadual Paulista) with respect to (i) period of study - day vs. night period (ii) teaching conditions - private vs. public school (iii) economical conditions - high vs. low family income. This examination reflect quality of high schools education. We observed long ubiquitous power law tails in Physical and Biological sciences in all cases. The mean value increases with better study conditions followed by better teaching and economical conditions. In humanities, the distribution is close to normal distribution with very small tail. This indicate that these power law tail in science subjects are due to the nature of the subject itself. Further better study, teaching and economical conditions are more important for physical and biological sciences in comparison to humanities. We explain these statistical distributions through Gradually Truncated Power law distributions. We discuss the possible reason for this peculiar behaviour and make suggestions to improve science education at high school level.

Motivation & Objective

• To investigate how socioeconomic, educational, and study conditions affect student performance in university entrance exams.
• To determine whether the observed power-law distribution in science subjects is due to subject nature or external factors.
• To compare the relative impact of study time, school type (private/public), and family income on academic performance.
• To model performance distributions using gradually truncated power-law functions to capture long-tailed behavior in science subjects.
• To provide data-driven recommendations for improving science education at the high school level.

Proposed method

• Analyzes marks from UNESP's university entrance examination (Vestibular) across 1998–2000 for three groups: physical sciences, biological sciences, and humanities.
• Applies a gradually truncated power-law distribution model defined by P(x) = c₁ / [c₂ + |x−xₘ|¹⁺ᵅ] × f(x), where f(x) introduces a smooth cutoff.
• Uses the normalization condition and a cutoff function f(x) that transitions from power law to exponential decay at x_c.
• Applies β = 2 − α to ensure large-scale convergence to a normal distribution.
• Compares empirical data across subgroups: day vs. night study, private vs. public schools, high vs. low income.
• Fits model parameters (α, xₘ, x_c, k, mean) to observed distributions and validates fit through curve comparison.

Experimental results

Research questions

• RQ1 Do power-law tails in student performance marks in science subjects arise from subject structure or external conditions?
• RQ2 How do study conditions (day vs. night), school type (private vs. public), and family income affect performance in science and humanities?
• RQ3 Why is the performance distribution in humanities approximately normal, while science subjects show long power-law tails?
• RQ4 To what extent do better study, teaching, and economic conditions amplify performance differences in science versus humanities?
• RQ5 Can gradually truncated power-law distributions accurately model educational performance data across different socioeconomic and educational subgroups?

Key findings

• Physical and biological sciences consistently exhibit long power-law tails in performance distributions across all subgroups (day/night, private/public, high/low income), indicating a systemic pattern.
• The mean score for day-time students is 140% higher than for night-time students in physical sciences, 157% higher in biological sciences, and 37% higher in humanities.
• Private school students outperform public school students by 117% in physical sciences, 126% in biological sciences, and 36% in humanities.
• High-income students score 81% higher than low-income students in physical sciences, 67% higher in biological sciences, and 27% higher in humanities.
• The power-law behavior is most pronounced in science subjects due to their interdependent, cumulative nature, which fosters long-term memory effects.
• The gradually truncated power-law model fits empirical data well across all groups, with parameters like α, xₘ, x_c, and k varying systematically with conditions.