World Bank Document

2.2 Estimation Techniques
The majority of empirical studies use statistical methods to estimate the various effects of 
interest. A few studies also use structural modeling, particularly computable general equilibrium 
(CGE) modeling, to assess the impacts of infrastructure on economic growth. We start with 
discussion of econometric/statistical approaches followed by general equilibrium modeling. Since 
the focus of this work is on the empirical studies investigating the effect of infrastructure on 
economic performance, inequality, and poverty, this study did not survey studies conceptualizing 
the effect of infrastructure on the various parameters (Glomm & Ravikumar 1994; Holtz-Eakin & 
Schwartz 1995, among others).
While the basic principles of the econometric approach should not be affected by how 
infrastructure is measured, in practice there are some specific issues in the mathematical 

---

9 
techniques that result in the estimates of the relationship depending on how infrastructure is 
measured. Therefore, we have divided the discussion below between methodologies used in 
studies that use public expenditure and studies that use physical infrastructures for the same 
purpose. 
2.2.1 Econometric/Statistical techniques
In this section, we discuss the statistical/econometric approach, including key limitations 
faced by these approaches and how some of these limitations have been addressed over time. 
Studies representing infrastructure through public expenditure: To establish the 
relationship between the infrastructure investment and economic output, the early studies mostly 
expressed the real economic output (e.g., per capita real GDP) as a function of labor input, capital 
input and public expenditures (or government expenditures on infrastructure) expressed in 
monetary units. To this end, public expenditures representing infrastructure, in many studies, refer 
to the flow of capital. The reason is that these values are available in countries’ national accounting 
systems. 
The first studies estimating the relationship between economic growth and infrastructure 
employed a production function approach, where the estimation techniques vary across studies. 
Many studies use ordinary least squares (OLS) in levels (Aschauer 1989a, 1989b, 1989c; Munnell 
1990, Munnell & Cook 1990; Garcia-Mila & McGuire 1992; Lynde & Richmond 1992; Holtz-
Eakin & Schwartz 1995; Rives & Heaney 1995; Sturm & De Haan 1995; Wylie 1996; Lewis 1998, 
among others). However, some studies use cross-sectional and panel data techniques to control for 
(unobservable) characteristics (Evans & Karras 1994; Holz-Eakin 1994, among others), first 
differences (Hulten & Schwab 1991; Evans & Karras 1994; Sturm & De Haan 1995, among 
others), and instrumental variable techniques (Duffy-Deno & Eberts 1991; Esfahani & Ramı́rez 
2003, among others).
The methodologies used in the early studies had several limitations, thereby questioning 
the credibility of their findings: key issues raised included spurious correlations, endogeneity