Applying the STIRPAT model in a post-Fordist landscape: Can a traditional econometric model work at the local level?
Introduction
The degree to which humans are responsible for changes in the environment has now become one of the most pressing questions in academic research. To what degree is anthropogenic activity forcing climate change? What human activities are primary genitors of environmental degradation, and to what magnitude are they liable? Out of this decades-long debate, the IPAT model has emerged as a primary method for conceptualizing the nature of relationships between people and their environment. Borne out of a debate over the relationships between consumption, population, and environmental degradation, the IPAT model has been applied in a multitude of empirical settings. Most recently, scholars have been applying the STIRPAT model -- a stochastic transformation of the IPAT model -- in order to econometrically test the theoretical model. Numerous studies have shown that the relationships posited by the model are statistically significant and positive determinants of environmental impact.
This research applies the STIRPAT model at the county level in the southeastern US. Using county-level carbon dioxide emissions as the dependent variable, the results of the regressions find that the traditional social and economic metrics used in the STIRPAT model present mixed results, complicating sign of the determinants as posited by the IPAT identity model. Population is statistically significant, positive, and a unit-elastic determinant of environmental impact, as expected from the IPAT identity model's assumptions. Affluence, on the other hand, is either a negative determinant or not significant in the analysis, depending on the model specification and dependent variable. Furthermore, other measures broadly labeled as 'technology' components -- variables intended to capture the landscape of development and manufacturing characteristic of the southeastern US -- illustrated significance in a variety of different models employed here. These results illustrate that when applying the model at a local level, the relationship between humans and environment becomes more complex, and that 'affluence' is an amalgamation of disparate economic components which exert force on the environment in differential ways.
Background
What role do demography and population play in an increasingly polluted planet? Are wealthy places responsible for increased stress on the global environment, or is it runaway growth of populated places? Initially put forth by Ehrlich and Holdren, and modified a multitude of times in the decades since, the IPAT model attempts to answer this question.
Coefficients in this model should be interpreted as elasticities, as they are akin to other elasticity research in econometrics. Coefficients that are at or significantly close to 1.0 are considered to be "unit elastic", the model is illustrating a proportional relationship between input and output. Coefficients greater than 1.0 illustrate an elastic relationship, where greater volumes of input result in proportionally greater volumes of output.
Typical STIRPAT research primarily focuses on modeling population and affluence, as there is little agreement about what constitutes technology and "no single operational measure of T that is free of controversy". In addition to being very difficult to define and transformed into a testable metric, technology -- unlike population and affluence -- is not so obviously a positive correlate of environmental impact. While the original intent of the model is to illustrate that there are environmental consequences to be paid for the process of industrialization, critics argue that this logic fails to take into account "green" technologies and the multitude of complex factors that are bound up in the overly-simplistic "technology". In short, there are almost an infinite number of ways to conceive of technology, making it difficult to treat it as an independent variable. Treating the residuals as a catch-all term for "technology" is certainly a potential avenue to work around this issue -- as some scholars have done -- but that approach leaves a variety of other unobserved factors embedded in the residual as attributed to technology. In order to work around this modeling quagmire, many analyses simply narrow the focus and test the signal strength population and affluence variables.
Similarly, the affluence term comes under significant scrutiny in its role as a positive driver of environmental degradation. Not only is it difficult to define affluence, but ecological modernization theory (EMT) also posits that the relationship between economy and environmental degradation is non-linear. As a take-off on Simon Kuznets' theory on wealth and inequality, EMT argues that this non-linearity manifests itself in the existence of an environmental Kuznets curve, an inverted U-shaped curve that describes the relationship between wealth and environmental degradation as having a "turning point", where very wealthy economies begin to see improvement in the environmental condition. Environmental degradation in this scenario is only initially a necessary byproduct of increasing industrialization. After a significant period of development and accumulation, sufficient marginal wealth is generated to initiate the process of improvement.
This idea has generated a great deal of debate. Critics of the EKC argue that empirical estimation methods are inadequate for forecasting human-environment relations, where the consequences for committing a Type 1 error are potentially disastrous. Theoretically speaking, critics argue that the logic relies upon dangerous utopian assumptions of unlimited growth and consumption; if indeed a Kuznets-type relationship does not exist, the level of degradation may be high enough that we have passed a threshold of no return. Additionally, even if an inflection point on the curve does exist, it could potentially be at a high enough level where environmental degradation might then be too costly to conceivably make any improvement. In this latter scenario, the investments expected in a transitioning economy which would go towards improving the environmental condition are inadequate and do little to ameliorate the problem. The estimation methods for detecting EKC relationships thus fail to address these potential non-linearities, thresholds, and feedback loops present in physical-environmental processes.
Empirically, the EKC is estimated by adding a quadratic (ln[A2]) term for affluence to the model; if an EKC-type relationship between wealth and environmental degradation is present, the affluence term will be positive and significant, while the quadratic term will be negative and significant. Naturally, the inclusion of this kind of term leads to rigorous debate about the proper econometric specification, and therefore a de facto political argument based on methodological choice.
Applying the model at the local level
A question that plagues IPAT research is whether or not the model scales. Does it describe human-economy-environment relations at all levels, or merely at the large scales of analysis, where most IPAT studies have taken place? Taken to its logical extreme, can IPAT be applied to the individual or household level? Although individuals certainly possess the ability to effect behavioral changes that positively impact the environment, there are many economic machinations that make these micro-scales of analysis less palatable for empirical research. At larger scales of analysis, greater numbers of economic processes become endogenous, as more and more economic activities are "included" in each unit i. For these reasons -- and along with more readily available data at bigger scales -- most empirical work has focused on variance between nation-states, where processes can be "containerized".
This final question about scale is germane to this research, and represents the principal gap in the literature that is hoped this research will address. Economic geographers have paid considerable attention to the ways that economic processes operate at multiple scales. Chief among these concerns are the ways in which economies undergo change and restructure internally. While the US has undergone several 'regional shifts' -- for example, the increasing importance of the Midwest to supply eastern production regions, aerospace and defense industries in California, and the sunbelt migration of northeastern manufacturing firms -- it is this latter that has had the most profound impact on the overall geography of the US economy. Research using the nation-state as the unit of analysis treats this great shift as unobserved heterogeneity, thus potentially ignoring critical changes in the environment of a particular region. Analysis at local level theoretically offers relief to this issue.
Methods and data
The study area is a nine-state area defined as the Southeastern United States, including the states of Arkansas, Louisiana, Mississippi, Alabama, Georgia, Florida, South Carolina, North Carolina, and Tennessee. Counties are the unit of analysis in this research, and there are 755 counties in the study area. While there are varying definitions over what constitutes "the South", this seems to approximate the sub-region according to very general definitions, particularly as they relate to zones of recent economic restructuring.
As a result of recent economic restructuring, the US Southeast offers a compelling case study in estimating the magnitude of human-environment relations. The shift in manufacturing and skilled, blue-collar labor from the urban northeast to the sunbelt -- both the southeastern and the southwestern US -- has drastically changed the landscape of the US industrial economy. Along with this great restructuring and a new spatial division of labor comes a breakdown in traditional distinctions in industrial location and economic geography. The continual search for lower costs and operating expenses by manufacturing firms has complicated the notion of an urban node as the center of the manufacturing hierarchy. Not only are heavy-industry firms moving from the rustbelt areas of the north to the sunbelt, but they are also seeking out what are non-traditional locations in these new areas. In short, many types of sites which contribute to degradation are more often found in areas that are atypical. Applying the IPAT model in a region that has been subjected to a tremendous influx of employment and industry will elucidate relationships between economy and environment that are not found elsewhere.
Results and discussion
The results of the regression are presented in [Table 1], [Table 2] and [Table 3]. Because both spatial dependency and heteroscedasticity are apparent in the error terms, results for an OLS regression with White's robust standard errors are provided in addition to the spatial-lag model specified in the methodology. The White procedure produces an adjusted standard error that is consistent and unbiased in a regression with heteroscedasticity in the disturbance term; failing to make this adjustment may produce biased standard errors, and by extension t-scores that indicate an erroneously significant result. These were run as separate procedures for each model because no procedure exists to account for producing both a spatial-lag regression and White's consistent standard errors under conditions of heteroscedasticity. Note that in general there are no material or substantive differences between the OLS models and the spatially-lagged models in terms of the significance and sign of each variable.
Conclusions
This research uses the STIRPAT framework to model social and economic relationships with the environment at the local level. Population and technology proxy variables such as urbanization and the density of manufacturing employment show a positive and statistically significant relationship with carbon output. Substantively speaking, the population variables are unit elastic -- an increase in one unit of population indicates a proportional increase in environmental impact. This is in agreement with a host of other research on ecological elasticities. Only Cramer found a scale factor significantly less than unity. At the same time, measures of affluence and income in the American southeast provide a much more muddled picture. Although income is a significant negative determinant for both industrial and residential CO2, indicating that the sources of carbon are stronger in less affluent and poorer areas, these results should ultimately be read as inconclusive. Further work is necessary to separate wealth-generated relationships from other processes unobserved in this analysis.
The goal of this research was to examine whether the traditional assumptions about sign and significance for IPAT variables hold at a local level. Using the Southeastern US as a case study allowed for analysis in a region that has undergone a tremendous economic change, and has received a number of people and firms formerly located in the industrial Northeast and Rustbelt. While the results of these estimations illustrate a complicated relationship between people, economy and environment in the region, they should also be interpreted as an illustration of the ways these relationships change when traditional economic-base formulations are complicated by fundamental shifts in the economic landscape. Further investigations of the STIRPAT model should aim to take into account these complex and often contradictory processes of development and economy.