India’s Economic Growth: Statistical Models Under Scrutiny

A recent analysis by Vidhu Shekhar, an associate professor of finance at SPJIMR, challenges statistical studies attempting to undermine India’s economic growth narrative. His work meticulously exposes significant methodological flaws and dataset sensitivities within these models.

This critique comes after previous contested claims suggested India had overstated its GDP growth by approximately 2.5 percentage points annually. That earlier assertion was later found statistically weak due to changing indicators and insufficient statistical power.

Dissecting the ‘Synthetic Control’ Methodology

The current focus of Professor Shekhar’s examination is a study by Kevin and Robin Grier of Texas Tech University, which utilizes a ‘synthetic control’ method. This study controversially asserts that Indian incomes were roughly 10% lower by 2023 than they would have been without Narendra Modi’s leadership.

Shekhar thoroughly challenges the reliability of this methodology through a compelling back-test. When the exact same method is applied to historical data from 2004 to predict the subsequent decade, it similarly ‘discovers’ a 10% Indian economic shortfall by 2014.

This critical finding suggests a fundamental flaw in the model, as it identifies a ‘Modi growth catastrophe’ a full decade before Modi’s tenure began. The synthetic control method, therefore, appears to generate consistent errors regardless of the actual political leadership.

Key Figures from the Analysis

10%: Indian income shortfall by 2023, as claimed by the Grier study.

10%: Indian income shortfall by 2014, predicted by the same model in a back-test before Modi’s tenure.

2.5 percentage points: Annual GDP growth previously claimed to be overstated in a contested analysis.

Dataset Sensitivity and Methodological Issues

Professor Shekhar further argues that the synthetic control method lacks a unique ‘synthetic India’ and that its results are highly sensitive to various factors. Slight changes in data or parameters drastically alter the study’s conclusions.

Data Vintage: The outcome significantly shifts when using World Bank data or a previous Penn World Table release instead of PWT 11.0.

GDP Concept: Altering the GDP measure, such as using the recommended growth rate metric instead of the expenditure-side measure for living standards, dramatically changes the findings.

Donor Pool: Expanding the group of comparison countries used in the model can also substantially modify the reported results.

In many of these alternative scenarios, India’s actual economic performance is shown to be at parity or even ahead of its synthetic counterparts. The article concludes that such statistical exercises construct an ‘imaginary India’ as a benchmark, which then unfairly criticizes the real economic experiences of 1.4 billion people.

These analyses often report a false sense of certainty that the statistical design cannot genuinely support, according to Shekhar. They risk misrepresenting complex economic realities by relying on inherently unstable models.