StateIO - Open Source Economic Input-Output Models for the 50 States of the United States of America

Abstract

Subnational input–output (IO) tables capture industry- and region-specific production, consumption, and trade of commodities and serve as a common basis for regional and multi-regional economic impact analysis. However, subnational IO tables are not made available by national statistical offices, especially in the United States (US), nor have they been estimated with transparent methods for reproducibility or updated regularly for public availability. In this article, we describe a robust StateIO modeling framework to develop state and two-region IO models for all 50 states in the US using national IO tables and state industry and trade data from reliable public sources such as the US Bureau of Economic Analysis. We develop 2012-2017 state IO models and two-region IO models at the BEA summary level. The two regions are state of interest and rest of the US. All models are validated by a series of rigorous checks to ensure the results are balanced at state and national levels. We then use these models to calculate a 2012–2017 time series of macro economic indicators and highlight results for 11 states that have distinct economies with respect to size, geography, and industry structure. We also compare selected indicators to state IO models created by popular licensed and open-source software. Our StateIO modeling framework is consolidated in an open-source R package, stateior, to ensure transparency and reproducibility. Our StateIO models are US-focused, which may not be transferrable to international accounts, and form the economic base of state versions of the US environmentally-extended IO models.

Keywords

multi-regional input-output model environmentally-extended input-output model open source software

Introduction

The United States (US) Environmental Protection Agency (EPA) developed and maintains the United States Environmentally-Extended Input-Output (USEEIO) family of models, which are nationally-representative, single-region, environmentally-extended input-output (EEIO) models based on national input-output (IO) tables along with a diverse set of environmental data compiled from public data sources (US Environmental Protection Agency 2020; Yang et al. 2017). These models were initially developed to serve the EPA’s Sustainable Materials Management (SMM) program. This program first used a national EEIO model to form the basis of an analytical framework to prioritize goods and services for policy and non-policy actions to reduce their life cycle impacts in the “Road Ahead” report (US Environmental Protection Agency, Office of Land, and Emergency Management 2009). The USEEIO model now serves as the basis for this SMM analytical framework at a national level, and it is intentionally composed of only publicly available data. Furthermore, all source code for preparing inputs and creating USEEIO models are publicly available under open source licenses by design, to promote transparency and reusability (Li, Ingwersen, Young, et al. 2022b). The “Road Ahead” report included a recommendation that the US state agencies harmonize the quantitative frameworks they are using for materials management, and that policies and programs be developed to target materials and related industries and supply chains with the greatest opportunities to reduce life cycle impacts. National EEIO models like USEEIO are inadequate to use for prioritization of goods and services at the state level due to the stark differences in the composition of economies and industry-level environmental performance among the 50 US states and District of Columbia (DC). Unlike the national IO tables underlying the USEEIO framework that are provided and regularly updated by the US Bureau of Economic Analysis (BEA) without license constraints, state IO tables are not currently provided or maintained by any public entities. Those provided by private entities are closed source and come with license requirements that limit their use. To provide a framework that can be adopted by any state or potential stakeholder interested in evaluating production and consumption in a state context that is fully transparent and freely available, state IO tables constructed with publicly provided and maintained data are needed. In addition, because the US states are highly integrated and economically interdependent, multi-regional EEIO models built with multi-regional IO (MRIO) tables are more suitable for characterizing impact reduction opportunities at the state level.

The MRIO modeling framework overcomes an important limitation of single-region IO models, which is the failure to include an explicit representation of interregional trade. In contrast to single-region IO models, MRIO models highlight the relevance of bidirectional trade linkages between regions. For example, a region that imports from another region that itself in turn imports from ultimately results in deep interregional economic dependencies, in which changes in demand for imports can have feedback effects on supply in the same region. Capturing these regional feedback effects is one justification for a MRIO model in place of single-region models (Miller and Blair 2009). As MRIO models can describe interactions between industries and households located in distinct regional contexts, they are usually used for two main purposes: (1) describing the sectoral structure of a given region and how it interacts with other regions through interregional trade; and (2), estimating Leontief multipliers that quantify how changes in final demand affect the regional industry output beyond the region experiencing the direct shock.

Previous research demonstrated that a two-region (state of interest [SoI] and rest of the US [RoUS]) MRIO-EEIO model was adequate for estimating environmental profiles of goods and services produced and consumed in a US state when compared to a 51-region model for most cases (Yang, Ingwersen, and Meyer 2018). In support of creating regionalized versions of the USEEIO models for states, the authors are proposing an original method for constructing two-region (SoI-RoUS) economic IO tables at the level of summary resolution provided by the BEA (US Bureau of Economic Analysis (BEA) 2019b). The method and model building approach detailed here are consolidated in a publicly available and open source R software package, stateior. Similar to the manner in which USEEIO is openly constructed and maintained, complete transparency in methods and procedures and use of exclusively public data have been paramount in the design of these models.

Literature Review

Both regional and multi-regional IO accounts are constructed by adjusting pertinent national IO tables and their derivatives (e.g., national direct requirements matrix) for regional trade and productivity differences (Lahr, Ferreira, and Többen 2020; Sargento 2009). Given the extent to which interindustry interactions and trade have increased over time and to which economies have become more open, multi-regional models are becoming increasingly popular among regional economists and other researchers for performing economic impact or economic contribution analyses. Although numerous contributions have been made to this formulation (Hewings 2020), the model developed by Leontief and Strout (Leontief and Strout 1963) and then modified by Polenske (Polenske 1980) is the most widely known. Nonsurvey approaches enhanced with superior data through the so-called “hybrid approach” (Lahr 1993, 2001) remain the primary means of developing MRIO models. Regionalization at the subnational level using nested MRIO databases (Faturay, Lenzen, and Nugraha 2017) has been applied to multiple countries, including China (Wang et al. 2015), Portugal (Ramos et al. 2015) and Canada (Bachmann, Roorda, and Kennedy 2015). MRIO models have been applied to high-level policy making, such as carbon footprinting in the United Kingdom (Barrett et al. 2013) and global material resource efficiency and decoupling (Wiedmann et al. 2015).

Besides the MRIO models in the US context, there are other experiences developed for other countries or economic regions. In the context of South America, there are several examples of the development of regional and multi-regional economic models for Brazilian regions (Guilhoto 1998; Guilhoto et al. 2010; Haddad, Gonçalves Júnior, and Nascimento 2017) and more recently for Colombia (Haddad et al. 2016). Other models have also been developed for countries like China (Zhang et al. 2013; Mi et al. 2018), Portugal (Ramos et al. 2015), Spain (Llano Verduras 2004; Pérez-Balsalobre, Verduras, and Díaz-Lanchas 2019), and the United Kingdom (Wiedmann et al. 2010).

Benefits of Multi-Regional Modeling

Multi-regional modeling better accounts for the effects of endogenous consumption expenditures. When modeling the 1970 Dutch economy for indirect impacts only, Oosterhaven et al. (Oosterhaven and Hewings 2021; Oosterhaven 1981), determined underestimations of the effects of regional income on regional final demand to be 1.1% for the relatively isolated rural Northern Netherlands region and 3.4% for the heavily urbanized greater Rotterdam region. However, if the effects of endogenous consumption expenditures are included in the multipliers (i.e., a Type II or SAM multiplier is used instead of a Type I multiplier), the neglect of interregional feedbacks increases the underestimations of regional income impacts to 3.1% and 6.6%, respectively. The larger feedbacks are driven by interregional commuting and shopping effects captured by endogenous household consumption. The same conclusion was observed by Ferreira et al. (Ferreira et al. 2018). More recently, Bouwmeester, Oosterhaven, and Rueda-Cantuche (Bouwmeester, Oosterhaven, and Rueda-Cantuche 2014) studied the European Union (EU) and compared the differences in income effects of EU27 exports to third countries using either a single-region or multi-regional model. The first-round intra-EU income spillover in a basic multi-region model was 7.7% greater than a single-region model. Furthermore, when running the same analysis in the full interregional EU27 model, the spillover effects appeared to be as large as 10.7% of the weighted-average domestic effect. Thus, transitioning to fully interconnected MRIO models might enable analysts to identify effects that might otherwise be underestimated or completely missed.

Data Limitations

A challenge of MRIO models is the larger amount of data relative to single-region models that are required to incorporate the magnitude of intra and interregional interdependency by industries and households, including regional gross outflows and inflows. The lack of interregional trade data and other economic data has also been well documented in the literature. In some circumstances, particularly in the case of goods, commodity transportation data might be available. Nevertheless, Sargento, Ramos, and Hewings (Sargento, Ramos, and Hewings 2012) enumerate four reasons why even these data are subject to inconsistencies: (1) services are not covered; (2) transport flows are sometimes expressed in physical terms; (3) regions with transport platforms might observe an over-estimation of trade flows; and, (4) detailed origin-destination freight transportation matrices often have non-numerical entries such as “unreliable value”.

Data on Gross Trade

The estimation of gross imports and gross exports is the basis of the so-called cross-hauling problem where countries/regions can simultaneously export and import products, such as cars. The right balance of local and imported consumption is not easy to assess. Polenske and Hewings (Polenske and Hewings 2004) draw attention to the need to address the issue of cross-hauling since it will continue to increase (worldwide) for two major reasons: first, regional trade leakages are growing in sophistication as firms look across the whole country (or even internationally) to find the most cost competitive locations to produce; and second, the search for product differentiation by firms is increasing to keep pace with growing consumer interest in products that can be produced locally or in other regions. Even before this, Treyz and Stevens (Treyz and Stevens 1979) suggested cross-hauling be the rule rather than the exception in most subnational regional economies. Even with this attention, the best method for addressing cross-hauling in MRIO models is still the subject of debate (Court and Jackson 2015).

The data required to adequately estimate international trade intensities at a subnational level are not always available despite these intensities being one of the critical steps in subnational MRIO model development. When constructing an MRIO model, the interregional trade must be balanced while obeying the restriction that everything that is imported must be exported interregionally and the sum of net exports must equal zero.

Interregional Trade Data

Direct collection of interregional trade data via survey is still considered to be too expensive (Lahr, Ferreira, and Többen 2020) and time consuming (Miller and Blair 2009), and, therefore, interregional trade is generally estimated. Lahr et al. (Lahr, Ferreira, and Többen 2020) identify four families of estimation techniques: Location Quotients (LQ), Flegg-Location-Quotients (FLQ), Cross-Hauling Adjusted Regionalization Method (CHARM), and econometric approaches. Of these, the LQ has the lowest data requirement, requiring data on employment, labor compensation, or production by industry or commodity and by region of the MRIO model. Stevens and Trainer (Stevens and Trainer 1980) knew that simple regionalization approaches—typically some form of LQ—in raw form were problematic since LQs do not permit cross-hauling. A rather long list of researchers note that ignoring cross-hauling leads to significantly biased estimates of subnational trade (Boero, Edwards, and Rivera 2018; Lehtonen and Tykkyläinen 2014; Mccann and Dewhurst 1998; Round 1983). Despite these criticisms, simple LQs are now used and will continue to be used as the need for a flexible procedure to infer an IO table and perform the economic analysis outweighs the need to accurately specify each sector in the set of interregional interdependencies.

FLQs are parametric LQs that are applied row-wise to a region’s technology matrix. FLQs and related parametric forms of the LQ and have been used in deriving interregional trade models (Flegg, Mastronardi, and Romero 2016; Flegg and Tohmo 2013, 2016; Jahn 2017; Kowalewksi 2015). FLQs require some a priori knowledge of interregional trade relationships for a nation. FLQ approaches also have been criticized (Lahr, Ferreira, and Többen 2020; Fujimoto 2019; Lamonica and Chelli 2018), because: (1) the value of the interregional parameter that is critical to establish the locally produced requirements is only possible to assess if interregional trade data are known; (2) some of the adjustments applied to LQs can be determined mathematically but there is a lack of meaningful economic interpretation of those values.

The third method for estimating interregional trade, CHARM, requires complete information on regional supply and demand. Depending on the data availability, CHARM can be easier or more complex to apply than FLQ (Lahr, Ferreira, and Többen 2020).

Among the econometric methods, those based on gravity theory have gained some popularity. According to Boero et al. (Boero, Edwards, and Rivera 2018) the method departs from a reference period for which observations on trade flows are available. Then, a linear regression is conducted on observed data to calibrate the parameters that allow for explaining trade flows of a certain commodity and a pair of regions while considering the distance between the regions. The estimation is done by industry using all the possible trade flows in the regression sample and thus leading to a single value for each parameter.

Any of the four methods are capable of accurately estimating the interregional trade of distinct sectors and regions, and the debate of selection of these methods is ongoing. A valid rule of thumb Szabo (Szabó 2015) considers is that the smaller a territorial unit, the greater its dependence on external territories through trade as the self-supplying ability of regions shrinks. Lahr et al. (Lahr, Ferreira, and Többen 2020) suggest that the technique used should be grounded in location theory and that particular emphasis should be given to the analysis of the Regional Purchase Coefficient (RPC). Stevens and Trainer (Stevens and Trainer 1980) coined the term RPC when referring to the extent to which regional producers fulfill the same region’s demand for a commodity. The RPC will vary from sector to sector and from region to region, being close to 1 where the local supply is more important than exports and being close to zero when local production tends to satisfy demand generated abroad, independently of the level of local production. Treyz and Stevens (Treyz and Stevens 1985) and Lahr et al. (Lahr, Ferreira, and Többen 2020) converge on the idea that RPC tends to increase if the transportation costs of the manufactured products increase, if the (geographical or economic) size of region increases, or if the production of a certain commodity is spread throughout the regions and if the region is more exposed to international trade and tourism. In this sense, this indicator is instrumental to establishing the relationship between production and trade at the sectoral and regional level.

Methods/metrics for assessing accuracy of estimations

Due to the lack of data, it is practically impossible to compare the values obtained after applying a non-survey method to estimate interregional flows and the real numbers. In fact, researchers have been discussing what might be the best way to estimate interregional flows by comparing with the small number of regional accounts that correspond to official data. Kowalewksi (Kowalewksi 2015) discusses the adequacy of the FLQ method in determining the regional accounts for 1993 in Germany. Flegg and Tohmo (Flegg and Tohmo 2019) apply the FLQ method to study the errors generated in the regional accounts in South Korea. And recently, Lahr et al. (Lahr, Ferreira, and Többen 2020) compare distinct methods to establish interregional trade in the EU. All these studies agree that it is not enough to just estimate net imports. Countries import products that they also export and regional accounts need to reflect the difference between regions. Another thing that is not adequately incorporated by all the methods but is supported by Treyz and Stevens (Treyz and Stevens 1985), Sargento et al. (Sargento, Ramos, and Hewings 2012), and Lahr et al. (Lahr, Ferreira, and Többen 2020) is that there is heterogeneity among products. Manufacturing products will be more tradable than Services and products with high transportation costs will be less tradable than products that are easy to transport. This means that in some cases, like the case of stateior, derived regional accounts are both state-specific but also product-specific as they try to incorporate information for each product.

One of the common indicators in most studies and methods applied to derive interregional trade is the RPC, which measures how local demand is satisfied by local production. This measure excludes the dependence of international trade. The RPC is relevant since it is also a way to measure how local supply will react to a shock on the demand side at the regional level. The RPC formula is given by

R P C_{i} = \frac{q_{i} - x_{i}}{q_{i} - x_{i} + m_{i}}

(1)

where

0 \leq R P C_{i} \leq 1

q_{i}

is the local supply of a given product and

x_{i}

is the regional exports. Then,

m_{i}

corresponds to the part of the local demand that is imported from other regions. When the RPC is zero, it means that all the local supply is directed towards other regions and if the region has any consumption of the product, then it has to be regionally imported. Alternatively, an RPC of one (or close to one) means that the local demand is mostly satisfied by local supply.

IO models can take the form of open or closed models. Open models can be used to create Type I multipliers, which account for the direct and indirect effects on output of changes in final demand. Closed models also account for the effects of income in household consumption in the matrix “core” and can be used to generate Type II multipliers. Katz (Katz 1980) shows how Type II multipliers are always higher than Type I multipliers as they also account for the induced effects of a shock. By induced effects, Katz (Katz 1980) refers to the shock of the income distributed by industries in the household consumption. Emonts-Holley, Ross, and Swales (Emonts-Holley, Ross, and Swales 2015) underline the danger of underestimation in the case of using Type I multipliers and the problem of overestimation in the case of Type II multipliers since not all the household consumption depends on the direct distribution of income by industries.

Existing MRIO Applications

After being derived, MRIO models can be extended with social-accounting matrices or environmental satellite accounts, to extend the analysis of the impacts to other dimensions besides the strictly economic impacts. Towa et al. (Towa, Zeller, and Achten 2020a) underline the role of environmental IO applied at the sub-national level as a form of performing a more accurate consumption-based analysis of carbon footprint impacts. As nations rely on the production of other countries, they in turn also rely on natural resources and products from other regions. Trade also extends the responsibility for the impacts of emissions, wastes, and natural resource extraction beyond political or administrative boundaries (Minx et al. 2009; Towa et al. 2020a, 2020b).

In the context of the US, several subnational MRIO models have been developed, varying both in geographical and temporal scope. IMPLAN is a widely used tool for regional economic impact assessment based on a US subnational MRIO model. IMPLAN was originally developed by the United States Department of Agriculture (USDA) Forest Service and is now maintained by the Minnesota IMPLAN Group, Inc. (Dahal, Henderson, and Munn 2015; Minnesota IMPLAN Group, Inc. (MIG) 2004). IMPLAN now includes MRIO-based datasets at the zip code, county, state, and national level. In IMPLAN, multipliers are used to estimate direct, indirect, and induced impacts resulting from changes in industry activity, employment, income, or other economic activity. According to Lindall et al. (Lindall, Olson, and Alward 2006), the IMPLAN trade model is a doubly-constrained gravity model that uses county level estimates of commodity supply and demand and guarantees that trade is balanced, such that the gross domestic imports are equal to gross domestic exports for all states.

With more specific applications, other multi-regional models have also been quite important in performing specific economic impact or contribution analyses. Besides describing the IMPLAN model, Rickman and Schwer (Rickman and Schwer 1995) also devote some attention to REMI and RIMS II. The REMI model was produced by Regional Economic Models, Inc. of Amherst, Massachusetts (Treyz, Rickman, and Shao 1991) and has the non-survey-based IO component linked to an econometric component. The econometric specifications of REMI are derived from the neoclassical idea of interregional equilibrium. The econometric component in REMI is applied for manufacturing sectors, using the Census County Business Pattern data and the Census of Transportation. For non-manufacturing sectors, supply-demand ratios and subjective intraregional trade coefficients are used to calculate the RPCs (Treyz and Stevens 1985). The RIMS II model, last updated in 2016, was developed by the US Department of Commerce/Bureau of Economic Analysis (Cartwright 1981). This model has been severely criticized (Drake 1976; Rickman and Schwer 1995) since it assumes that local demand is satisfied first, with the remainder of an industry’s output assumed to be exported and not allowing for cross-hauling between regions.

Other examples of MRIO models are the WiNDC model from the University of Wisconsin (Wisconsin National Data Consortium 2021), the Chicago model from the Regional Economic Applications Laboratory (Hewings, Okuyama, and Sonis 2001); the RUBMRIO model for Texas (Kockelman et al. 2005); and the NIEMO model for the 50 states plus Washington D.C. (Park et al. 2007). Unfortunately, these MRIO models either lack trasparency for verifying model outputs are following the reported methods or suffer from their proprietary nature that obstructs the public from automatically updating or reproducing model results. Our economic modeling framework, built with stateior, is developed to overcome these barriers and provide open source, transparent MRIO models in the US context to the public.

Methodology

An original method for constructing economic IO tables in the form of two-region, SoI and RoUS, at the level of summary resolution as defined by the BEA (US Bureau of Economic Analysis (BEA) 2019b) is described. This method is implemented in a publicly available and open-source R software package, stateior. Three main steps are taken to construct StateIO tables: (1) the creation of a state supply model, (2) the creation of a state demand model, and (3) estimation of interregional trade to create a two-region IO model.

Standard matrix algebra notation is used to delineate the steps of creating the state Supply and Use models, using conventions for variable names commonly used in the IO literature and the existing USEEIO model documentation when possible. Capital letters indicate matrices and lowercase letters denote vectors. A “^” over a variable represents the diagonalization of a vector as a matrix. An exponent of “-1” represents an inverse. A “ $'$ ” represents the transposed (rows and columns switched) form of a matrix or vector.

The model was designed to have the structure and components described in Figure 1. The structure reflects an IO model that is open with respect to households and other institutions.

Figure 1.

Bi-regional IO model for StateIO.

Inside the bold border in Figure 1, representing the “core” of the StateIO framework and comprising the supply and use matrices, are the basic elements for estimating the Leontief Inverse (

L

) and corresponding multipliers. The supply (or production) matrices,

V

, describe each commodity’s production structure, i.e. the commodities produced by each industry, in each region. The use matrices,

U

, describe the consumption of commodities and value added (VA) inputs by each industry, in each region, according to the origin of the production. The regional matrices are constrained by “domestic flows”, meaning they only include the products produced within regional economies and national borders (the international imports –

M

and

T M

- are treated separately). All transactions are expressed at “basic prices”, i.e., excluding the value-added tax and the other taxes less subsidies on products (

T

). According to the BEA, trade and transportation margins are treated as inputs provided by the Retail and Wholesale Trade Services or the Transport Services (integrated in

U

and

F

). Notation details of the matrices are presented in Table 1.

Table 1.

Notation Within the StateIO Framework.

Matrix/Vector	Elements	Description	Level of Detail
$U^{i, j}$	$u_{r, s}^{i, j}$	i = SoI, RoUS; $j$ = SoI, RoUS - intermediate consumption of region $i$ commodity $r$ , used by industry $s$ located in region $j$ .	Commodity by industry
$F^{i, j}$	$f_{r, c}^{i, j}$	i = SoI, RoUS; $j$ = SoI, RoUS - final demand of region $i$ commodity $y$ , used by component $c$ located in region $j$ .	Commodity by final demand component
$T C O^{i}$	$t c o_{r}^{i}$	i = SoI, RoUS; $j$ = SoI, RoUS - total output of commodity $r$ , produced in region $i$ , at basic prices.	Commodity
$V^{i}$	$v_{r, s}^{i}$	i = SoI, RoUS; $j$ = SoI, RoUS - commodity production of $r$ by industry $s$ in region $i$ .	Commodity by industry
$T I O^{i}$	$t i o_{s}^{i}$	i = SoI, RoUS; $j$ = SoI, RoUS - total output of industry $s$ in region $i$ .	Industry
$T {(U)}^{i}$	$t_{s}^{i}$	i = SoI, RoUS - taxes (less subsidies) on production falling upon industry $s$ , in region $i$ .	Industry
$T^{i}$		i = US - total taxes (less subsidies) on production.	Scalar
$M {(U)}^{i}$	$m_{s}^{i}$	i = SoI, RoUS - international imports used by industry $s$ , in region $i$ .	Industry
$M {(F)}^{i}$	$m_{c}^{i}$	i = SoI, RoUS - international imports used by final component $c$ , in region i	Final demand component
$M^{i}$		i = US - international imports.	Scalar
$m u_{r}^{i}$		i = US, SoI, RoUS - international trade adjustment by commodity $r$ , in region i	Commodity
$T U^{i}$	$t u_{s}^{i}$	i = US - total intermediate consumption by industry $s$ in region $i$ , at purchaser’s prices.	Industry
$T F^{i}$	$t f_{c}^{i}$	i = SoI, RoUS - total final demand by final demand component $c$ in region $i$ , at purchaser’s prices.	Final demand component
$T U^{i}$ + $T F^{i}$		i = SoI, RoUS; $j$ = SoI, RoUS - intermediate consumption of region $i$ commodities, used by industries located in region $j$ .	Scalar
$V A^{i}$	$v a_{s}^{i}$	i = SoI, RoUS - value added in industry $s$ in region $i$ .	Industry
$T V A^{i}$		i = US - national value added.	Scalar

Final demand in the StateIO framework is comprised of the same components as in the BEA tables (Table 2). This table also shows the relative weight of these components in total final demand for the year 2017.

Table 2.

Final Demand Components and Relative Weights in 2017.

Final Demand Component	Description	Weight, %
F010	Personal consumption expenditure	59.7
F02E	Nonresidential private fixed investment in equipment	3.8
F02N	Nonresidential private fixed investment in intellectual property products	3.9
F02R	Residential private fixed investment	3.6
F02S	Nonresidential private fixed investment in structures	2.8
F030	Change in private inventories	0.1
F040	Exports of goods and services	9.9
F06C	National defense: Consumption expenditures	2.9
F06E	Federal national defense: Gross investment in equipment	0.2
F06N	Federal national defense: Gross investment in intellectual property products	0.3
F06S	Federal national defense: Gross investment in structures	0.0
F07C	Nondefense: Consumption expenditures	1.9
F07E	Federal nondefense: Gross investment in equipment	0.1
F07N	Federal nondefense: Gross investment in intellectual property products	0.5
F07S	Federal nondefense: Gross investment in structures	0.1
F10C	State and local government consumption expenditures	8.4
F10E	State and local: Gross investment in equipment	0.1
F10N	State and local: Gross investment in intellectual property products	0.2
F10S	State and local: Gross investment in structures	1.4
F051	International trade adjustment	0.3

Since our goal is to establish 51 state models (the 50 states and DC) that are coherent with the national IO tables, the sum of flows between an SoI and its corresponding RoUS must equal the national totals. Therefore, all transactions in the national territory must be allocated to the regions. However, for some economic indicators, the sum of the 51 regional values is close but not exactly equal to the national total. The cause of this divergence is the overseas adjustment.¹ To keep the consistency between state and national flows, a 52^nd region is created, defined as Overseas, that is only used for closing the balance with the transactions not allocated to one of the 51 geographic regions.

State Supply Model

Supply (or Make) tables describe commodity production by industries in a given region. The US Make table published by BEA is not available at any sub-national geographical level. The state Make tables constructed here have industries as rows and commodities as columns just like the BEA US Make table allowing for byproduct production, so each cell contains the dollar value of each commodity produced by a given industry. In each state Make table, the sum of each row represents state industry output, while the sum of each column represents state commodity output.

The BEA regional accounts do not publish state industry ( $T I O^{S o I}$ ) or state commodity output ( $T C O^{S o I}$ ). Therefore, state information on VA by industry ( $V A^{S o I}$ ) is used, as a proxy for industry output, to regionalize national industry output, ( $T I O^{U S}$ ), and estimate state industry output, ( $T I O^{S o I}$ ), in equation (2).

t i o_{s}^{S o I} = t i o_{s}^{U S} \times \frac{v a_{s}^{S o I}}{v a_{s}^{U S}}

(2)

In equation (2), $v a_{s}^{S o I}$ is the value-added of industry $s$ in state ( $S o I$ ), and $v a_{s}^{U S}$ is national value-added for industry $s$ . In this process, as BEA summary industries in the national Make and Use table do not correspond perfectly to the industries in the state VA data, a crosswalk is used to map VA industries to BEA summary level industries. Within the Appendix, the Federal Spending By State section and the Detailed Approach of Constructing State Supply Model section specify the steps applied for mapping retail, real estate and federal, state and local government sectors where differences in industry definitions among these two data sources are present.

Now that the total state industry output, $T I O^{S o I}$ , is known, a determination must be made regarding the value of commodities produced by each industry in the state, i.e. corresponding industry row in the state Make table $V^{S o I}$ , is still unknown. BEA clarifies that establishments are classified to industries according to the primary product that they produce (or the primary activity that they perform) (Horowitz and Planting 2009). Commodities are not produced by a single industry as required in a simplified industry-by-industry square model. According to the national accounts, each industry can produce more than one type of commodity. For example, a farm specialized in fluid milk production can also sell some meat or butter as a byproduct. This additional production is called secondary production or production of by-products. Secondary products must be allowed for the state Make table as well, and therefore for a given industry, the commodity output will not necessarily equal industry output.

The primary source of commodity output is the national Make table. For commodities where data are available that can serve as proxy values for commodity output, initial state commodity output is estimated independently. State agricultural commodity output are obtained from USDA Census of Agriculture; state fisheries commodity output is obtained as values of state-level fishery landings from National Oceanic and Atmospheric Administration (NOAA) Fisheries; state forestry commodity output is obtained as cut values of forestry from the United States Forest Service (USFS) Forestry Inventory. State output of manufactured and transported commodities are obtained from the Freight Analysis Framework (FAF) developed by the Oak Ridge National Laboratory (ORNL) (Oak Ridge National Laboratory (ORNL) 2020). The alternative data sources are used to create SoI-RoUS commodity output ratios as shown in equation (3)

t c o_{s}^{S o I} = t c o_{s}^{U S} \times \frac{c p_{s}^{S o I}}{c p_{s}^{U S}}

(3)

where

c p_{s}

is the commodity output proxy variable for a given industry and is defined both for SOI and US.

To estimate the final value of each cell in the state Make table, $V^{S o I}$ , the RAS method² is applied. The RAS method transforms an initial table to establish a final table that obeys two sets of restrictions (sum of the components of each row and column). In this case, the initial Make table for each state is generated by multiplying the US Make table by the diagonal matrix of state-US industry output ratios. The RAS method is applied to each industry to assure the sum of state industry production are equal to the national industry output.

The approach of constructing the state supply model is illustrated in Figure 2 with a more detailed version in the Appendix in the Detailed Approach of Constructing State Supply Model section (Figures A1–A2). At this point, a state supply model contains one Make table (industry by commodity) and the sum of its rows and columns corresponds to the sum of state industry output and the sum of state commodity output, respectively.

Figure 2.

Approach of constructing state supply model.

State Demand Model

With industry and commodity production at the state level determined, consumption of commodities by industry is estimated, also called intermediate consumption $U^{S o I}$ , and the final consumption of commodities, i.e. the final demand $F^{S o I}$ . BEA (Horowitz and Planting 2009) establishes that the consumption of intermediate inputs are purchases of goods and services used for the production of other goods and services and do not include the labor and taxes that are components of VA. The $U^{i, j}$ in Figure 1 is an essential component of IO modeling as it describes the industry production technology, which is the value of inputs that are consumed to produce a certain value of output. The additional columns in Figure 1 represent final demand, $F^{i, j}$ , which is the “sales by each sector to final markets […] such as personal consumption purchases and sales to the federal government. For example, electricity is sold to businesses in other sectors as an input to production (an interindustry transaction) and also to residential consumers (a final-demand sale)” (Miller and Blair 2009). This component describes the personal consumption expenditures (PCE), private fixed investment, local change in inventories, international exports, and federal, state and local government expenditures and investment. A general rule for IO models is that production, or the industry and commodity output estimated in the Make table, $V$ , must be consumed domestically or abroad for the system to be balanced. The balance between production and consumption must be verified independently of the geographical scope of the IO framework.

Before deriving the state Use table, a vital adjustment is applied to solve the discrepancy between the US import matrix and the imports of commodity in the BEA national Use table before redefinition (import column, F050), referred to as international trade adjustment (ITA) hereafter. The ITA values are added as a column with a component code of F051³ in the domestic Use table in order to balance national and state production and consumption. Details about ITA are described in the Appendix in the Adjustments in the National Matrix section.

State Intermediate Consumption

For StateIO models, it is recommended by BEA to use the hypothesis of a homogeneous production function to derive state intermediate consumption, $U^{S o I}$ in Figure 1, i.e. the demand of commodities by industries in their production process. This means that, for StateIO models, the production technology at the state level is assumed to be the same for every state in the model. Models built via this method have a structure that assumes the “industry-technology assumption, meaning that “all products made in the same industry have the same production function” (US Bureau of Economic Analysis (BEA) 2019a). The StateIO models follow a clear and generic rule that applies to every commodity, industry, state and year. As stated by the BEA (US Bureau of Economic Analysis (BEA) 2019a), in the absence of more detailed data, assuming the same input consumption structure is an adequate procedure.

As Miller and Blair (Miller and Blair 2009) recognize: “Soft drinks of a particular brand that are bottled in Boston probably incorporate basically the same ingredients in the same proportions as are present in that brand of soft drink produced in Kansas City or Atlanta or in any other bottling plant in the US. But, electricity produced in eastern Washington by water power represents quite a different mix of inputs from electricity that is produced from coal in Pennsylvania or by means of nuclear power or wind farms elsewhere.” The problem of industry specificities led Lahr and Stevens (Lahr and Stevens 2002) to prove that one can obtain a more accurate representation of industries’ input consumption structure if a more disaggregated model is available for further distinguishing between close, yet different, sectors.

The application of the rationale explained above means that, in mathematical terms (equation (4)), elements of state industry intermediate consumption, $T U^{S o I}$ , are estimated by allocating national intermediate consumption by industry based on state-to-US industry output ratios.⁴

t_{s}^{S o I} = t_{s}^{U S} \times \frac{t i o_{s}^{S o I}}{t i o_{s}^{U S}}

(4)

Thus, theoretically every industry in every state can have a unique input consumption structure. However, data limitations make it difficult to describe even one sector in one state. Survey methods for subregions are too costly and expensive (Miller and Blair 2009; Boero, Edwards, and Rivera 2018). At this point, it is important that the users less familiar with IO can understand the implications of such limitation. According to Lahr (Lahr 1998), if the regional modeler’s goal is to describe one sector in one given state in detail, the identification of critical sectors and uses and the collection and application of superior data will significantly improve model accuracy.

State Final Demand

Final demand is composed of ten distinct components. Household consumption, designated as PCE, makes up the largest share of final consumption. According to Kim, Kratena and Hewings (Kim, Kratena, and Hewings 2015), this component accounted for 70% of gross domestic product in the US as of 2011, compared to 56% on average for OECD (Organisation for Economic Co-operation and Development) countries. The other components are private fixed investment, change in private inventories, international exports, federal, state and local government expenditures and investments. In this section, the distinct methods and data used to regionalize each US final demand component to the corresponding state final demand component is described.

Personal Consumption Expenditures

Personal consumption is not assumed to be proportional to the population in every state. Different states might exhibit distinct consumption patterns as the population can be very heterogeneous across states. It is well documented that consumption patterns can change according to socio-economic characteristics of the population: age, educational background, urban, suburban or rural contexts, among others (Druckman and Jackson 2009; Kim and Hewings 2015; Ferreira et al. 2018). More importantly, the level of consumption is highly correlated with the level of income that households attain in a certain geographical area (Varlamova and Larionova 2015). Personal consumption expenditures data by state are used to distribute the consumption among the 50 states plus DC.

International Exports

States exports are not assumed to be directly proportional to output, as states can have a smaller or larger propensity to export their goods to foreign markets. This can depend, for instance, on a state industry’s position in the supply chain, the distance to the border, or the existence of transportation infrastructure such as a cargo airport or a port. Farole and Winkler (Farole and Winkler 2014) explore the differences in state exports and show how the distance to national agglomerations can affect the openness to the global market. From the StateIO perspective, this means that to deal with such an important level of heterogeneity, the estimation of exports can largely benefit from the inclusion of region specific data. Foreign trade data from the US Census Bureau (US Census Bureau 2020b) is used to estimate the state commodity export ratios. The state trade data are created in part by models of the exports from the region where the commodities were produced, and the models do not always result in reliable estimates.⁵ Using these data in StateIO, there are cases where state commodity exports are greater than state commodity output. This is not possible in economic terms as exports is a component of the commodity production. So, in the rare cases where these data generated this incompatibility, the model automatically corrects this imbalance by using the state commodity output ratio, $\frac{t c o_{r}^{S o I}}{t c o_{r}^{U S}}$ , as an alternative way to allocate the international exports of that commodity in that state and then distributes the excess throughout the other regions. Nevertheless, when this step is completed, the model has assigned international exports for all the commodities in the 51 US regions and a new component of the final demand is estimated.

Federal, State, and Local Government Expenditures and Investment

Federal government spending data available at USAspending (USAspending.gov 2020) are the main source used to distribute the national flows to the state level. The spending data are first categorized into (1) consumption expenditures or (2) investments in structure, equipment, and intellectual property according to their product or service code (US General Services Administration and Federal Acquisition Services 2020). The investments were further separated into defense and non-defense based on the awarding agency. The data for defense and non-defense investment were obtained separately using data on each particular investment equation (5), while the data on defense and non-defense consumption expenditures are derived based on a combination of intermediate federal government spending, state employment compensation ratios, and national government expenditure weighting factors equation (5).

For the state and local government consumption expenditures and investments, US Census Bureau data (US Census Bureau 2020a) on state and local government expenditures were used to calculate a ratio of state expenditures, $φ$ , that is then transformed in a regional distribution of each component.

f_{C}^{S o I} = ϕ \times f_{C}^{U S}

(5)

Final Demand Remaining Components

The remaining components of final demand not including imports (residential private fixed investment, non-residential private fixed investment, and change in private inventories) composed about 15% of US final demand in 2017. For lack of state-specific data, state-US commodity output ratios ( $\frac{t c o_{r}^{S o I}}{t c o_{r}^{U S}}$ ) are applied to allocate national values to states.

International Imports

International imports are reported in BEA tables as final demand because final-use transactions consist of the transactions that make up the final-expenditure components of the GDP and international imports are the most important negative component of GDP (Horowitz and Planting 2009). However, as the IO matrices are commonly used to reflect the impact of changes in the national or regional economy, international imports must be included in a different way. The reason is that international imports represent leakages in the economy; in other words, they have no multiplier effects within the regions. For example, if an industry only used inputs that were produced abroad, changes in their economic activity would not produce changes in the national economy but will instead impact the industries located abroad throughout its intermediate consumption. In this case, the local impacts would be limited to wages and capital transfers. The need to treat international imports for economic impact purposes is well described by Horowitz and Planting (Horowitz and Planting 2009, chaps. 12–5): “IO tables are frequently used to calculate the impact of changes in final uses on domestic output, income, or employment of industries using the total requirements matrix… but before calculating the domestic portion of the requirements, the industry inputs from foreign sources need to be removed. This removal is accomplished using an import matrix—that is, a matrix that shows the use of the imports by industries and final uses.”

Therefore, the first step of estimating state international imports is to expunge all the flows detailed in the IO framework of the part of the demand that are generated abroad. Our model excludes the possibility of reexports⁶ of a particular commodity, so there are no imports that are directed towards exports of the same product. At the state level, the value of imports is obtained by multiplying the corresponding value on the import matrix, $m_{r, s}^{U S}$ and $m_{c, s}^{U S}$ , with the share of each state’s industry consumption and final demand in the US economy (equations (6) and (7)).

m_{s}^{S o I} = m_{s}^{U S} \times \frac{t i o_{s}^{S o I}}{t i o_{s}^{U S}}

(6)

The same is applied for final demand components.

m_{c}^{S o I} = m_{c}^{U S} \times \frac{t f_{c}^{S o I}}{t f_{c}^{U S}}

(7)

This procedure estimates international imports at the state level and allows allocation of national international trade adjustment, $μ$ , detailed in Adjustments in the National Matrix section, to states (equations (8) and (9)).

m u_{s}^{S o I} = m u_{s}^{U S} \times \frac{m_{s}^{S o I}}{m_{s}^{U S}}

(8)

The same is applied for final demand components.

m u_{c}^{S o I} = m u_{c}^{U S} \times \frac{m_{c}^{S o I}}{m_{c}^{U S}}

(9)

When $\frac{m_{s}^{S o I}}{m_{s}^{U S}}$ and $\frac{m_{c}^{S o I}}{m_{c}^{U S}}$ are not viable, total consumption ratios, $\frac{u_{s}^{S o I}}{u_{s}^{U S}}$ and $\frac{f_{c}^{S o I}}{f_{c}^{U S}}$ , are used as alternatives, respectively.

Since international imports and trade adjustment are part of the final demand, they are then included in two separate rows outside the core of the matrix and will be considered exogenous.

State Value Added

The state VA data set released by BEA has been used in previous sections to derive the state-US VA ratio, which is further used to regionalize US Make or Use table to states. To estimate state VA, $V A^{i}$ , the same state-US VA ratio is used again to regionalize US VA to states.

The overall approach of constructing the state demand model is illustrated in Figure 3 with a more detailed version in the Appendix in the Detailed Approach of Constructing State Demand Model section (Figures A3–A4). At this point, the consumption and production in each state are fully derived. However, to have a multi-regional model it is also necessary to estimate the inter-dependencies that occur between commodities, industries, and final demand components at the regional level since not all consumption that happens in one region is supplied by industries in the same region. This will be addressed in the next section.

Figure 3.

Approach of constructing state demand model.

Interregional Trade Estimation

In the previous section, the processes for estimating the total supply and total demand for each industry and final demand component at the state level was described. However, one important aspect is still missing. In a bi-regional IO model, the multiplier effect of a given shock will be significantly different if one industry is directed towards the internal market (in this case, the state) or if instead is more directed towards external markets (other US states or abroad). With Make and Use tables prepared for each state, the model can produce accurate numbers in terms of net exports. If a state has a large volume of supply and almost no-demand, then in net terms, it is expected that the state will export. However, that does not imply that the volume of imports is zero. Indeed, given the fact that the world is more and more interdependent, states usually import goods that they also export. Additionally, to estimate the economic multipliers data on net exports is not enough. It is essential that the gross imports and gross exports are known and that they are balanced in the sense that everything that is being exported from a region must be imported by another region.

In a two-region framework like the one representing a SoI and the RoUS, the Use table will detail the local consumption (hereafter referred as SoI2SoI and RoUS2RoUS) and the interregional trade between the two areas (SoI2RoUS and RoUS2SoI). For a given pair of SoI and RoUS, the two-region Use table (Figure 4) has a four-quadrant structure overall, where

• Red quadrant shows use of commodities by industries in SoI that are produced in SoI;

• Pink quadrant shows use of commodities by industries in RoUS that are produced in SoI;

• Blue quadrant shows use of commodities by industries in SoI that are produced in RoUS;

• Purple quadrant shows use of commodities by industries in RoUS that are produced in RoUS;

• Bars on the right represent final demand of commodities in SoI and RoUS with the same color meaning in the quadrants, e.g. pink bars show final demand of commodities in RoUS that are produced in SoI;

• Orange boxes at the bottom are VA amounts in the two regions.

Figure 4.

Two-region Use table.

The goal of the StateIO models is to allow users to estimate a bi-regional model that describes the economic interdependencies between an SoI and its RoUS. This model comprises all the economic activity divided in a group of four distinct economic flows:

1) a SoI2SoI table demonstrating commodities produced and consumed in SoI,

2) a SoI2RoUS table demonstrating commodities produced in SoI but exported to and consumed in RoUS,

3) an RoUS2RoUS table demonstrating commodities produced and consumed in RoUS

4) an RoUS2SoI table demonstrating commodities produced in RoUS but exported to and consumed in SoI.

The models also include the Make tables that describe the production of commodities by industries in the SoI and in the RoUS. Until this point, the model provides the information on what is produced and what is consumed but does not include an estimation of the trade flows between the regions. It is known how much Regions A and B produce, but at this stage how much of Region A consumption comes from its own region or Region B, and vice-versa, must be estimated. This information is essential to perform an accurate economic impact assessment since state economies are interconnected. For example, industries in many states rely on Texas production of oil or Idaho’s production of potatoes.

It is important to understand that this step of the model will work in two stages. First, the products are classified according to their tradability. Some products are considered tradable and others are more locally consumed. To estimate the different degrees of tradability, the model first estimates something designated as the domestic Interregional Commodity Flow (ICF) Ratio. Then, in a second part, this ICF is applied to the model as it is described in the State Demand Model section and the trade is adjusted to balance all consumption, production, and trade flows that are incorporated in a bi-regional model by incorporating a set of economic restrictions that must be present in multi-regional models.

According to the recent tradition in modern IO modeling and international and interregional studies, the StateIO model admits the existence of cross-hauling since regions import commodities that they also export. Polenske and Hewings (Polenske and Hewings 2004) discuss why this happens and will happen even more in the future as production and societies become more and more interdependent and the level of product differentiation increases.

Interregional Trade Estimation

To estimate the trade flows between regions, the StateIO model draws on several publicly available data sources to prepare a set of ICF ratios. For each commodity and state a set of 4 ICF ratios, corresponding to the four distinct economic flows described in the previous section, are calculated.

Interregional Trade in Goods

Interregional trade of goods is estimated based on the Freight Analysis Framework (FAF) domestic commodity flow model (Oak Ridge National Laboratory (ORNL) 2020), where FAF goods correspond to StateIO commodities. The FAF model estimates the weight and dollar value of commodities that are transported from an origin state to a destination state. These data are the main source for a first estimation of the proportions of a commodity produced in one region that stays in the same region or is domestically exported to another US state. Dollar values of total shipments by FAF good are used to develop ICFs as shown in equations (10) and (11).

I C F^{S o I, S o I} = \frac{C F^{S o I, S o I}}{T D^{S o I}}

(10)

I C F^{R o U S, S o I} = \frac{C F^{R o U S, S o I}}{T D^{S o I}}

(11)

where

C F^{S o I, S o I}

is the flow of a commodity within the state, while

T D^{S o I}

is the total demand of that commodity in the state. An

I C F^{S o I, S o I}

close to one implies that most of the demand is satisfied through local production while a value close to zero suggests that demand is largely satisfied by imports.

Movements of electricity, water, and natural gas provided by the utility sector are not tracked in the FAF model, and therefore require sector specific data for modeling interregional trade. ICF ratios for sector “22 - Utilities” are calculated at a more detailed subsector level, including ICF ratios for three subsectors: (1) “221100 - Electric power generation, transmission, and distribution”; (2) “221200 - Natural gas distribution”; and (3) “221300 - Water, sewage, and other systems”.

For “221100 - Electric power generation, transmission, and distribution”, state electricity consumption and interstate trade data from the US Energy Information Administration’s State Energy Data System (US Energy Information Administration 2021) are used. Although the electricity is transmitted across state borders, the amount of electricity transmitted from one state to another is not accurately detailed. Therefore, net interstate trade of electricity is used to represent interstate export or import. If a state’s “net interstate flow of electricity and associated losses” is less than zero, it is considered a net exporter of electricity as in equation (12).

\hat{I C F^{S o I, R o U S}} = \frac{| N I T_{221100}^{S o I} |}{C_{221100}^{R o U S}}

(12)

where

N I T

is the net interstate trade, and

C

is total consumption. Values for electricity are in million kilowatt hours.

If SoI is a net exporter, consequently $\hat{I C F^{S o I, R o U S}} = 1 - \hat{I C F^{S o I, R o U S}}$ . By contrast, if a state’s net interstate trade is larger than zero, then the inverse situation applies.

For “221200 - Natural gas distribution” and “221300 - Water, sewage and other systems”, all economic activities are assumed to occur locally, in other words, there is no interstate trade, hence $I C F^{S o I, S o I}$ and $I C F^{R o U S, R o U S}$ are one while $I C F^{S o I, R o U S}$ and $I C F^{R o U S, S o I}$ are zero. To determine the ICF ratios of “22 - Utilities”, a weighted average of the ratios of “221100”, “221200”, and “221300” are calculated using state total employment in the respective sector (US Bureau of Labor Statistics (BLS) 2020) as the weighting factor. Where data for interstate transport of goods are unavailable, ratios of state commodity output are used to calculate ICFs.

Interregional Trade in Services

Some services are by nature local because provision of these services must occur within the region of interest. For example, construction or housing activities (imputed or rented) are an extreme case of locally consumed services. According to the rules of regional accounts for state GDP (Adams et al. 2021), the place where such activities occur is where the associated income is generated. For the same reason, international trade in these sectors is non-existent or considered residual. In the StateIO models, locally produced services, including construction, housing, transit, and ground passenger transportation, other transportation, and all government sectors, have an $I C F^{S o I, S o I}$ and $I C F^{R o U S, R o U S}$ of 1, while $I C F^{S o I, R o U S}$ and $I C F^{R o U S, S o I}$ are 0.

Services that are not local by definition can be classified as local or tradable according to how the services are produced and consumed in the US. These designations were published by the US Cluster Mapping Project (Harvard Business School and US Department of Commerce, Economic Development Administration 2020). Industries were classified as “traded” or “local” by NAICS (North American Industry Classification System) code. For the StateIO model, the “traded” or “local” NAICS classifications are mapped to the StateIO industries and combined with SoI and RoUS commodity output to approximate commodity ICF ratios.

Sector-specific methods are used for transportation, warehousing, and waste management services. Since the FAF model also provides transportation modes of commodity flows, it is used to estimate the ICF ratios for transportation services. The assumption is that if a trade flow happens within a state all the service is directly provided by the corresponding state transportation industry but if the transportation happens between states, then half of the economic activity of the transportation activity will stay in the origin and half is imputed to the state of destination. For these cases, the $I C F^{S o I, S o I}$ is adjusted using equation (13).

\hat{I C F^{S o I, S o I}} = c + I C F^{S o I, S o I} \times (1 - c)

(13)

where

c

is 0.5.

Ratios for RoUS are handled in the same way. The same logic is applied to the warehousing and storage sector, assuming all commodities would be stored in warehouses before transported to destination.

A specific approach is used for the ICF ratio estimation for sector “562 - Waste management and remediation services”. This sector includes the following activities:

• 562111 - Solid waste collection

• 562212 - Solid waste landfilling

• 562213 - Solid waste combustors and incinerators

• 562910 - Remediation services

• 562920 - Material separation/recovery facilities

• 562HAZ - Hazardous waste collection treatment and disposal

• 562OTH - Other waste collection and treatment services

ICFs associated with “562 - Waste management and remediation services” are calculated with data on net physical waste treatment by state in a method that is the inverse to how the net physical data for electricity trade were used in equation (12). Imports of waste from RoUS to SoI for treatment and disposal in SoI are not considered ICF from RoUS to SoI, instead they are exports of management service from SoI to RoUS. In this case, data from the EPA’s RCRAInfo Biennial Report (US Environmental Protection Agency 2018) are used to calculate ICF ratios for “562HAZ - Hazardous waste collection treatment and disposal” using equation (14)

I C F_{562 O T H}^{S o I, R o U S} = \frac{R_{562 O T H}^{S o I}}{M_{562 O T H}^{S o I}}

(14)

where

R_{562 O T H}^{S o I}

is the total amount (in tons) of interstate hazardous waste received by SoI, and

M_{562 O T H}^{S o I}

is total amount of hazardous waste that are managed by facilities in SoI.

As for the other waste management services, the commodity output ratio is used as a proxy for their ICF ratios. Then, to determine the ICF ratios of “562 - Waste management and remediation services”, the ratios of hazardous waste management services and non-hazardous waste management services are summed on an equal basis.

Two-Region Domestic Use Tables

With the ICF ratios determined, interregional flows between the two regions in our model, SoI and RoUS, are estimated. Each commodity in each region is multiplied by the respective ICF ratio to create a new table (equations (15) and (16)) that already includes a first estimation for interregional imports and exports in both SoI and RoUS regions.

u_{r, s}^{S o I, S o I} = u_{r, s}^{S o I} \times I C F^{S o I, S o I}

(15)

f_{r, c}^{S o I, S o I} = f_{r, c}^{S o I} \times I C F^{S o I, S o I}

(16)

Interregional imports, or the demand for products in SoI supplied by the RoUS region, $u_{r, s}^{S o I, R o U S}$ and $f_{r, c}^{S o I, R o U S}$ , are then estimated in equation (17) and (18), respectively.

u_{r, s}^{S o I, R o U S} = u_{r, s}^{S o I} - u_{r, s}^{S o I, S o I}

(17)

f_{r, c}^{S o I, R o U S} = f_{r, c}^{S o I} - f_{r, c}^{S o I, S o I}

(18)

The sum of the row for each commodity provides the total imports of commodity $r$ in SoI satisfied by supply from the RoUS region. Alternatively interregional exports from SoI to RoUS of commodity $r$ , $e_{r}^{S o I, R o U S}$ , is given by

e_{r}^{S o I, R o U S} = t c o_{r}^{S o I} - (u_{r}^{S o I, S o I} + f_{r}^{S o I, S o I})

(19)

In other words, all the production from SoI region that is not consumed locally or exported internationally must be exported to the RoUS region. This method is also applied in the RoUS side and produces values for interregional imports by SoI with origin in the RoUS and by the RoUS with origin in SoI and interregional exports from the SoI to the RoUS and vice-versa. Summarizing, four interregional trade indicators are produced.

While all industries are considered to be affected by interregional trade, not all components of final demand were assumed to be interregionally tradable, including nonresidential private fixed investment in structures, residential private fixed investment, change in private inventories, international exports and imports, and government gross investment in structures.

This first estimation produces inconsistencies because the model uses data from distinct sources. The first inconsistency that has be solved is that in mathematical terms the interregional exports estimated can be negative. This happens if the supply for a given commodity is smaller than the local consumption. However, this is impossible according to economic theory. In the case of deficits where local consumption is greater than local supply, the negative interregional export values are proportionally allocated across industries and final demand sectors in that region, according to the weight of the commodity consumption in total state consumption equations (20) and (21).

u_{r, s}^{S o I, S o I} = u_{r, s}^{S o I, S o I} + e_{r}^{S o I, R o U S} \times \frac{u_{r, s}^{S o I, S o I}}{\sum_{s = 1}^{71} u_{r, s}^{S o I, S o I} + \sum_{c = 1}^{13} f_{r, c}^{S o I, S o I}}

(20)

f_{r, c}^{S o I, S o I} = f_{r, c}^{S o I, S o I} + e_{r}^{S o I, R o U S} \times \frac{f_{r, c}^{S o I, S o I}}{\sum_{s = 1}^{71} u_{r, s}^{S o I, S o I} + \sum_{c = 1}^{13} f_{r, c}^{S o I, S o I}}

(21)

After this step, all interregional imports and interregional exports are greater than zero. But there is still an important imbalance that needs to be resolved. Imports from the SoI might not be equal to the exports made from RoUS, and vice-versa. This discrepancy, $ϵ$ , is solved by distributing the values across both regions using the state commodity output ratio equations (22) and (23).

u_{r, s}^{S o I, S o I} = u_{r, s}^{S o I, S o I} + ε^{S o I, c} \times \frac{u_{r, s}^{S o I, S o I}}{\sum_{s = 1}^{71} u_{r, s}^{S o I, S o I} + \sum_{c = 1}^{13} f_{r, c}^{S o I, S o I}}

(22)

f_{r, c}^{S o I, S o I} = f_{r, c}^{S o I, S o I} + ε^{S o I, c} \times \frac{f_{r, c}^{S o I, S o I}}{\sum_{s = 1}^{71} u_{r, s}^{S o I, S o I} + \sum_{c = 1}^{13} f_{r, c}^{S o I, S o I}}

(23)

where

ϵ^{S o I, r} = (\sum_{c = 1}^{13} f_{r, c}^{S o I, R o U S} - e_{r}^{R o U S, S o I}) \times \frac{c p^{S o I}}{c p^{U S}}

The method described above to estimate and balance interregional trade is applied to most of the industries. The exceptions are the industries in which all consumption was assumed to be local. In this case, since no interregional trade is assumed, the distribution of a discrepancy between supply and demand affecting other regions would necessarily result in a model with negative interregional imports, which is untenable from the economic point of view. In this case, the difference between supply and demand is included as a new component of final demand that is called the “export residual”. For commodities that have SoI2SoI ICF ratio equal to 1, their “export residual” equals their “interregional exports”.

At this stage, the model is balanced. Interregional imports from one region equal the interregional exports from the other region. Also, all the domestic supply, $V$ , is consumed domestically in SoI or RoUS, $U^{S o I}$ , $F^{S o I}$ , $U^{R o U S}$ , and $F^{R o U S}$ , or exported out of the country and all the domestic consumption is associated with commodities produced in SoI, RoUS or imported from international origins.⁷ In other words, the model is balanced because everything that is produced is consumed according to the rules of national accounts.

By having established the SoI2SoI table and the RoUS2RoUS table, the other two matrices, one representing the commodities produced in SoI and used in RoUS, SoI2RoUS, and the other commodities produced in RoUS and used in SoI, RoUS2SoI, are estimated by subtracting the value of total consumption from the intra-regional accounts. The final assembly of these four tables results in the two-region domestic Use tables Figure 4.

Two-Region Total Use Tables

The methodology described in the previous section offers StateIO model users a domestic Use table that represents domestic flows and that is suitable for analyzing and interpreting the supply and consumption of products that are produced in the US economy. In economic terms, the domestic Use table can yield multipliers that identify the effects of economic shocks to the national or regional economy. For example, if the demand for a good imported from another country suffers a positive shock, the multiplier in the US should be zero as the demand relies on another economy.

In other cases, the goal of the user is to understand the dependency on a commodity independently of where it is produced. Such cases require what are commonly referred to as Total Use tables, where the uses of commodities include international imports. The Use tables for the US provided by BEA are in this form (US Bureau of Economic Analysis (BEA) 2019b, 2019c). Cells in the Total Use table convey the total consumption of a commodity by a given industry, regardless of national origin. The Total Use table was prepared using equation (24).

\overset{ˇ}{u_{r, s}^{S o I}} = u_{r, s}^{S o I} + m_{r, s}^{S o I}

(24)

The value of the total use of commodity $r$ by industry $s$ in SoI region, $\overset{ˇ}{u_{r, s}^{S o I}}$ , is the value of the domestic use, $u_{r, s}^{S o I}$ , plus the value that is internationally imported, $m_{r, s}^{S o I}$ . This value is then distributed by SoI and RoUS according to the destination of the commodity, SoI or RoUS.

Model Validation

To assure the two-region Use tables are successfully built, a set of validation steps were performed. Validation is applied to assure that all the tables produced follow the economic rationale that a balanced model must observe. Equations for model validation presented in equations (25)–(28) are used to guarantee that errors produced are limited to minor rounding errors.

\sum_{S o I = 1}^{51} V^{S o I} \approx V^{U S}

(25)

\sum_{S o I = 1}^{51} U^{S o I} + \sum_{S o I = 1}^{51} F^{S o I} \approx U^{U S} + F^{U S}

(26)

\sum_{S o I = 1}^{51} T I O^{S o I} \approx T I O^{U S}

(27)

\sum_{S o I = 1}^{51} T C O^{S o I} \approx T C O^{U S}

(28)

Finally, the most important validation is when the Use matrix produced is used to estimate the technical coefficient (direct requirements) matrix and then transformed to estimate the Leontief matrix, $L$ . This matrix, when multiplied by total final demand (directed towards SoI and RoUS), must reproduce the initial values of state commodity output and state industry output that are both represented in the initial version of the Make table and in the Use table, with a tolerance of ±1 million dollars.

This step (equation (29)) was applied successfully in the cases of all the states and all years covered by the StateIO models.

L \times [\begin{array}{l} F^{S o I} \\ F^{R o U S} \\ 0 \\ 0 \end{array}] = [\begin{array}{l} T C O^{S o I} \\ T C O^{R o U S} \\ T I O^{S o I} \\ T I O^{R o U S} \end{array}]

(29)

Additional validation is presented in the Appendix in the Model Validation section.

Results and Discussion

Results from the StateIO models are evaluated through examination of macroeconomic indicators for selected US states that produce commodities that satisfy a substantial fraction of US domestic demand. The RPC metric obtained using the StateIO models are compared state by state and year by year with the RPCs obtained from other available US state IO models. The software implementation of the StateIO models is briefly described.

Sectoral Structure and Interregional Trade

In this sub-section, macroeconomic indicators are presented from the two-region models for 11 distinct states for 2017, including total gross value added (GVA) and gross imports and exports. The 11 states were pre-selected to represent different sizes (by gross product), key economic industries, and a representative set of geographic locations (e.g. the East, West, and Midwest regions). Results from the year 2017 are shown in Figure 5. In this figure, the 11 states are ranked by their total GVA in 2017. Their total exports to the rest of US and imports from the rest of US are shown next to the total GVA.

Figure 5.

State Total GVA and Interregional Trade in 2017.

As Figure 5 highlights, the structure and the size of the economies are very different from state to state, and the related percentage contribution of interregional trade (using the largest of imports or exports) to total GVA ranges from 44% (California) to 76% (in Iowa). This outcome aligns with economic theory and the idea that smaller regions are more open as they tend to depend more on trade with the outside, while larger regions are often less dependent on exports or imports.

StateIO models also reveal signature commodities produced by states. A 2012–2017 time series of interregional exports from the 11 states is shown in Figure 6. For each of the selected 11 states and years, the contribution of the top five exported commodities/services is highlighted.

Figure 6.

Interregional exports by commodity.

Overall, exports from the selected states to their corresponding RoUS gradually increased from 2012 to 2017, which is consistent with other studies that have shown how trade is growing between countries, regions, and sectors (Timmer et al. 2015; Los, Timmer, and Vries 2015). The 11 states shared some of the top 5 exports despite their varying economic structure. “42 - Wholesale trade” and “5412OP - Miscellaneous professional, scientific, and technical services” were the common top exports, even the largest and second largest, in all the states except Iowa where “5412OP” was not in the top 5. These sectors are highly aggregated and include some tradable activities and firms that operate nationwide. “111CA - Farms,” “311FT - Food and beverage and tobacco products,” and “524 - Insurance carriers and related activities” made up the top 3 exports from Iowa. The first two exports come as little surprise from a state known for its agriculture-oriented economy. In Delaware, “524 - Insurance carriers and related activities,” “521CI - Federal Reserve banks, credit intermediation, and related activities,” and “532RL - Rental and leasing services and lessors of intangible assets” were the top 3 exports from 2012 to 2017. These financial exports likely reflect the high percentage of firms that incorporate in Delaware (Delaware Division of Corporations 2022) and it is the home of a large percentage of US consumer credit card companies (Tsosie 2017). New York had top-ranking exports that are similar to Delaware, but “523 - Securities, commodity contracts, and investments” ranks top 2 or 3, which is explained by New York being a hub for investment firms and securities exchanges (Bajpai 2021).

The 3 states that were the largest interregional exporters for raw and manufactured commodities from 2012 to 2017 are highlighted in Figure 7.

Figure 7.

Top 3 export states for raw and manufactured commodities.

California was the largest exporter of farm (111CA) and forestry and fishery products (113FF) to the rest of the US. Two states in the Midwest, Iowa and Nebraska, emerged in second and third place in terms of farm product exports, while Florida and Oregon were the second and third largest exporters of forestry and fishery products. It is worth noting that Oregon was also the largest exporter of wood products (321), meaning that besides raw timber, Oregon also exported significant quantities of manufactured wood products. As for other raw products like crude oil and natural gas (211), Texas was the most important supplier and exporter, followed by Oklahoma and Alaska and more recently, Colorado. However, looking at exports of petroleum and coal products (324), which are crude oil-based products that have been further processed in the refinery, Texas was still the most important exporter, but Louisiana and California, and even Ohio, played an important role in the market in 2012–2017. Petroleum was further processed to produce chemical products (325), which were mainly exported from California, Indiana, and Texas, and plastics and rubber products (326) that were mainly exported from Ohio, Illinois, and North Carolina during these years. California was also the largest exporter of computer and electronic products (334), followed by Massachusetts and North Carolina.

Mined products (212) were primarily exported from West Virginia, Wyoming, and Arizona. Nonmetallic mineral products (327) were mainly exported from Ohio, Pennsylvania, Wisconsin, and Texas. The Southeastern states, including Georgia, North Carolina, South Carolina, and Virginia were the top exporters of food (311FT) and textile products (313 TT). Wisconsin, Pennsylvania, and three Southeastern states, including Alabama, South Carolina, and Georgia, were top exporters of paper products (322).

Our results also show that “Rust Belt” states are the main providers of heavy equipment and metal-based products. Indiana, Ohio, and Pennsylvania were the top exporters of primary metals (331); Ohio and Illinois were the top exporters of fabricated metal products (332); and Illinois, Ohio, Iowa, and Wisconsin were the top exporters of machinery (333). Michigan was the main exporter of motor vehicles (3361 MV), followed by Indiana and Ohio and more recently, Tennessee.

Regional Purchase Coefficients

RPCs from the StateIO models for the selected 11 states are compared with RPCs from three other common models: IMPLAN (Minnesota IMPLAN Group, Inc. (MIG) 2004), IO-Snap (IO-Snap 2022), and WiNDC (Wisconsin National Data Consortium 2021). Figure 8 shows the overall RPC values per state between 2012 and 2017 for these models.

Figure 8.

StateIO and other model RPC comparison.

RPCs from all four models are generally stable over the period of 2012–2017, only fluctuating by 1–3%. IO-Snap RPCs for Delaware fluctuate more, having a difference of 8% between the lowest and highest values over the period. StateIO RPCs are consistently smaller than IMPLAN and IO-Snap RPCs by an average of 10% and 16%, respectively. The RPCs for Iowa are an exception where the RPCs from IMPLAN are nearly identical to those of StateIO. IO-Snap RPCs are the highest among all except for in Delaware, where they are noticeably fluctuating and lower than IMPLAN RPCs, and in New York, where they are only slightly lower than IMPLAN RPCs in 2015–2017. Alternatively, the WiNDC model produces much smaller RPCs that are on average 30% smaller than StateIO. One possible explanation is that WiNDC relies solely on commodity flow data and commodity output⁸ to model interregional trade flows and therefore it is more suited to the study of material flows (Rutherford and Schreiber 2019). By treating all commodities as tradable, local preference for some commodities (more specifically services) might be underestimated, which likely overestimates the “openness” of the state economies. Variation in RPCs are likely caused by differences in methods related to inter-regional trade, aggregation bias, etc.

Figure A1.

State supply model - Part 1.

Figure A2.

State supply model - Part 2.

Figure A3.

State demand model - Part 1.

Figure A4.

State demand model - Part 2.

To better understand the basis for the RPC estimated through the StateIO model, an ordinary least squares (OLS) econometric model is used to identify the determinants of the StateIO RPCs. Table 3 summarizes the results. The OLS results represent a way to summarize our estimations and demonstrate that, independently of the unobserved deviance to real numbers, the RPC ends up being strongly dependent on four aspects that are more commonly referred in the literature (LeSage and Pace 2008; Helpman, Melitz, and Rubinstein 2008).

Table 3.

OLS Econometric Model of StateIO RPCs.

Variable	Coeff.	Std. Error	p-value
Constant	0.3219	0.07485	9.04 × 10⁻⁵ ***
GVA	3.640 × 10⁻¹⁴	4.47 × 10⁻¹⁵	2.13e-10 ***
Supply/Demand	0.1435	0.05097	0.00721 **
Share_Local	0.4466	0.06007	2.33 × 10⁻⁹ ***
Share_Intl_Export	0.5373	0.1764	0.00388 **
Signif. codes: 0 ‘*’ 0.001 ‘’ 0.01 ‘*’ 0.05 ‘.’ 0.1’ ’ 1
R-squared: 0.7176, adjusted R-squared: 0.6925
F-statistic: 28.59 on 4 and 45 DF, p-value: 7.541 × 10⁻¹²

First, the positive signal in the GVA variable shows that the RPC will increase as the relative size of a state economy increases. Second, the positive value for the supply and demand ratio shows more supply contributes to increase the RPC indicating that the local provision of local demand is higher. Third, as the weight of more locally provided commodities like housing, renting, and construction increases, the overall RPC of the state also increases. And, finally, as a state is relatively more directed towards international exports, then it will also better satisfy its own local demand. This can be a sign that more competitive states, with more relative capacity to operate in external markets, are also the ones more capable of satisfying their own demand.

Table A1.

State GDP, Interregional Trade, and RPC.

State	Year	Total GVA (Billion $)	Interregional Exports (Billion $)	Interregional Imports (Billion $)	Net (Interregional) Exports (Billion $)	Overall RPC
Arizona	2012	274.628	119.952	151.310	−31.358	0.689
Arizona	2013	282.111	118.555	154.822	−36.266	0.693
Arizona	2014	291.730	123.110	160.708	−37.597	0.694
Arizona	2015	304.318	129.616	164.770	−35.154	0.696
Arizona	2016	318.751	132.651	165.917	−33.265	0.703
Arizona	2017	336.081	142.069	177.938	−35.868	0.700
California	2012	2139.805	930.152	920.412	9.740	0.746
California	2013	2248.220	973.547	959.846	13.702	0.748
California	2014	2367.524	1054.964	1018.601	36.363	0.747
California	2015	2513.002	1086.462	1040.804	45.658	0.750
California	2016	2614.776	1122.192	1081.597	40.595	0.750
California	2017	2776.050	1227.127	1151.349	75.778	0.749
Delaware	2012	63.031	44.452	33.361	11.091	0.653
Delaware	2013	62.429	43.800	34.678	9.122	0.652
Delaware	2014	69.172	52.181	37.581	14.601	0.642
Delaware	2015	72.861	55.006	38.128	16.878	0.639
Delaware	2016	70.427	49.417	36.054	13.363	0.653
Delaware	2017	69.816	48.480	37.679	10.801	0.651
Florida	2012	789.398	277.482	425.222	−147.740	0.708
Florida	2013	823.375	294.964	438.047	−143.083	0.710
Florida	2014	865.539	310.342	464.551	−154.208	0.710
Florida	2015	926.311	323.555	475.189	−151.634	0.717
Florida	2016	974.366	339.823	495.124	−155.301	0.719
Florida	2017	1023.891	366.727	528.265	−161.538	0.716
Georgia	2012	452.156	271.756	263.828	7.927	0.669
Georgia	2013	469.334	283.525	273.341	10.184	0.669
Georgia	2014	495.403	302.346	291.660	10.686	0.667
Georgia	2015	527.798	308.904	291.000	17.904	0.678
Georgia	2016	555.374	330.172	301.739	28.433	0.679
Georgia	2017	582.373	352.150	321.933	30.218	0.675
Iowa	2012	159.932	125.914	106.937	18.976	0.634
Iowa	2013	163.701	126.727	106.632	20.095	0.641
Iowa	2014	174.818	138.961	114.583	24.379	0.638
Iowa	2015	182.388	140.557	111.038	29.519	0.649
Iowa	2016	183.370	136.433	108.591	27.842	0.654
Iowa	2017	185.866	140.377	113.773	26.604	0.648
Maine	2012	54.255	24.340	32.988	−8.648	0.676
Maine	2013	55.060	24.508	33.455	−8.947	0.676
Maine	2014	57.123	25.641	34.199	−8.558	0.680
Maine	2015	58.999	25.112	33.344	−8.232	0.691
Maine	2016	61.257	26.241	34.143	−7.902	0.693
Maine	2017	63.348	29.063	36.800	−7.737	0.682
Minnesota	2012	301.253	191.311	166.181	25.130	0.681
Minnesota	2013	314.215	193.691	172.470	21.221	0.683
Minnesota	2014	328.350	204.724	180.353	24.371	0.683
Minnesota	2015	340.068	193.670	174.562	19.108	0.697
Minnesota	2016	348.924	200.738	179.245	21.493	0.695
Minnesota	2017	359.552	205.652	188.605	17.046	0.692
New York	2012	1342.934	790.112	622.033	168.079	0.699
New York	2013	1380.810	809.418	645.327	164.092	0.700
New York	2014	1448.922	866.109	686.424	179.686	0.698
New York	2015	1509.541	893.542	686.844	206.698	0.703
New York	2016	1576.481	927.866	697.833	230.033	0.706
New York	2017	1628.595	960.169	732.177	227.992	0.704
Oregon	2012	176.309	86.475	97.202	−10.727	0.687
Oregon	2013	181.346	89.642	100.083	−10.441	0.688
Oregon	2014	190.261	94.055	104.493	−10.438	0.690
Oregon	2015	203.465	98.720	105.828	−7.109	0.698
Oregon	2016	214.612	103.509	110.388	−6.879	0.699
Oregon	2017	225.975	108.938	117.696	−8.758	0.698
Washington	2012	405.090	169.094	208.382	−39.287	0.720
Washington	2013	423.910	175.490	218.595	−43.105	0.720
Washington	2014	446.134	186.912	236.331	−49.419	0.715
Washington	2015	473.388	194.729	233.466	−38.737	0.726
Washington	2016	495.166	205.973	241.195	−35.221	0.724
Washington	2017	527.842	226.746	257.181	−30.435	0.723

A complete table of 2012–2017 state GDP, trade, and RPCs is available in the Appendix in the Summary of State GDP, Interregional Trade and Overall RPC section (Table A1).

Software Implementation

stateior v0.1.0 was used to generate the StateIO models that are described in this manuscript (Li, Ingwersen, Ferreira, et al. 2022a). The stateior package is maintained and updated in a public git repository. The main data products of stateior are the two-region (SoI-RoUS) Make, Use, and Domestic Use tables along with industry and commodity output vectors. stateior v0.1.0 generates these data products for all states for years 2012–2017 at the BEA summary level of sector resolution. The resulting models are written to R data format (*.RDS) and pushed to the EPA Data Commons (Zhuang and Balassiano 2021). As well as incorporating all the code to build the models from the raw sources, stateior provides convenient functions for users of the two-region tables to load these tables or vectors for year and state of interest. stateior data products will be used by EPA’s useeior R package (Li, Ingwersen, Young, et al. 2022b) in the near future to construct state-level EEIO models in the USEEIO framework. Plans for future releases of stateior can be found on the stateior Wiki pages.

Conclusion

This article describes a robust StateIO modeling framework for building complete state and two-region IO models and calculating inter-regional trade in the US. The resulting StateIO models produce results that align with known data and understandings of state economies and are comparable to other state IO models. The main contribution of this article is to clearly present to potential users and researchers the robust methodology with its underlying assumptions facilitated by open source programming code available in the stateior package. The transparency and reproducibility of the model allows for deeper review of the models’ core data, algorithms, and assumption, such that it can be more easily compared to other state IO models and methods.

The StateIO models can be used as general purpose IO models for US states from which Type I multipliers can be produced for regional economic impact assessment or the full two-region tables can be used for broader regional science applications or research purposes alongside or in place of existing off-the-shelf models where the available years, level of sectoral resolution, and two-region format of the StateIO models are fit for purpose.

Future work will include development of environmental and employment accounts to align with these models and expansion of the useeior package to build state versions of USEEIO models using the StateIO models. The resulting state EEIO models can be used for state-specific applications by EPA (US Environmental Protection Agency, Office of Land, and Emergency Management 2009) or state government agencies, but also for other regional applications where EEIO models are useful. Planned applications include a state-specific prioritization tool to reveal in which sectors opportunities exist to reduce impacts of state production and a consumption-based greenhouse gas inventory for selected US states (Bridge and Ingwersen 2018), and preparing consumption-based greenhouse gas inventories for US states (Ingwersen, Young, and Cashman 2021).

Footnotes

Acknowledgments

The authors would like to thank Catherine Birney and Andrew Schreiber for helpful comments on earlier versions of this article.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was funded by the U.S. EPA’s Sustainable and Healthy Communities Research Program. This research is supported through U.S. EPA contract HHSN316201200013 W, Task Order EP-G16H-01256 with General Dynamics IT (GDIT). Bill Michaud (GDIT) and Bhagya Subramanian (EPA) assist with contract and project management.

Disclaimer

The US Environmental Protection Agency, through its Office of Research and Development, funded and conducted the research described herein under an approved Quality Assurance Project Plan (K-LRTD-0030017-QP-1–3). It has been subjected to review by the Office of Research and Development and approved for publication. Approval does not signify that the contents reflect the views of the Agency, nor does mention of trade names or commercial products constitute endorsement or recommendation for use.

ORCID iD

Mo Li

Notes

Appendix

Acronyms

Acronym	Definition
BEA	Bureau of economic analysis
CHARM	Cross-hauling adjusted regionalization method
D	Defense
DC	District of Columbia
E	Employment
EEIO	Environmentally-extended input-output model
EPA	Environmental protection agency
EU	European union
FAF	Freight analysis framework
FGD	Federal government demand
FLQ	Flegg-location-quotients
GVA	Gross value added
IC	Intermediate consumption
ICF	Interregional commodity flow
IO	Input-output
ITA	International trade adjustment
LQ	Location quotients
MRIO	Multi-regional input-output model
NAICS	North american industry classification system
NIPA	National income and product accounts
ND	Non-defense
NOAA	National oceanic and atmospheric administration
NTU	Net total use
OECD	Organisation for economic Co-operation and development
ORNL	Oak ridge national laboratory
RoUS	Rest of the United States
RPC	Regional purchase coefficient
SMM	Sustainable material management
SoI	State of interest
US	United States
USEEIO	United States environmentally-extended input-output model
USDA	United States department of agriculture
USFS	United States forest service
VA	Value added

References

Adams

Alex

Aten

Bettina

Figueroa

Eric

Kublashvili

Solomon

Parajuli

Krishna

Yoon

Albert

York

Jack

. 2021. “Preview of the 2021 Annual Update of the Regional Economic Accounts.” https://apps.bea.gov/scb/2021/07-july/pdf/0721-regional-annual-revision-preview.pdf

Bachmann

Chris

Roorda

Matthew J.

Kennedy

Chris

. 2015. “Developing a Multi-Scale Multi-Region Input–Output Model.” Economic Systems Research. https://doi.org/10.1080/09535314.2014.987730

Barrett

John

Peters

Glen

Wiedmann

Thomas

Scott

Kate

Lenzen

Manfred

Roelich

Katy

Le Quéré

Corinne

. 2013. “Consumption-Based GHG Emission Accounting: A UK Case Study.” Climate Policy. https://doi.org/10.1080/14693062.2013.788858

Bajpai, Prableen.

2021. “The World’s Leading Financial Cities.” Investopedia. October 2021. https://www.investopedia.com/articles/investing/091114/worlds-top-financial-cities.asp

Boero

Riccardo

Edwards

Brian K.

Rivera

Michael K.

. 2018. “Regional Input–Output Tables and Trade Flows: An Integrated and Interregional Non-Survey Approach.” Regional Studies. https://doi.org/10.1080/00343404.2017.1286009

Bouwmeester

Maaike C.

Oosterhaven

Jan

Rueda-Cantuche

José M.

. 2014. “A New SUT Consolidation Method Tested by a Decomposition of Value Added and CO2 Embodied in EU27 Exports.” Economic Systems Research. https://doi.org/10.1080/09535314.2014.892473

Bridge

Jarrod

Ingwersen

Wesley

. 2018. “Sustainable Materials Management Prioritization Tools: National and State Models.” Presented at the EPA tools and resources webinar series. https://www.epa.gov/research-states/sustainable-materials-management-smm-webinar-archive

Cartwright

Joseph V

. 1981. RIMS II: Regional Input-Output Modeling System : Estimation, Evaluation, and Application of a Disaggretated Regional Impact Model. US Department of Commerce, Bureau of Economic Analysis, Regional Economic Analysis Division.

Court

Christa D.

Jackson

Randall W.

. 2015. “Toward Consistent Cross-Hauling Estimation for Input-Output Regionalization.” Regional Research Institute West Virginia University Working Paper Series. https://researchrepository.wvu.edu/cgi/viewcontent.cgi?article=1021&context=rri_pubs

10.

Dahal

Ram P.

Henderson

James E.

Munn

Ian A.

. 2015. “Forest Products Industry Size and Economic Multipliers in the US South.” Forest Products Journal. https://doi.org/10.13073/FPJ-D-14-00083

11.

Delaware Division of Corporations . 2022. “About the Delaware Division of Corporations.” March 2022. https://corp.delaware.gov/aboutagency/

12.

Drake

Ronald L

. 1976. “A Short-Cut to Estimates of Regional Input-Output Multipliers: Methodology and Evaluation.” International Regional Science Review. https://doi.org/10.1177/016001767600100201

13.

Druckman

Angela

Jackson

Tim

. 2009. “The Carbon Footprint of UK Households 1990–2004: A Socio-Economically Disaggregated, Quasi-Multi-Regional Input–Output Model.” Ecological Economics, Methodological Advancements in the Footprint Analysis. https://doi.org/10.1016/j.ecolecon.2009.01.013

14.

Emonts-Holley

Tobias

Ross

Andrew

Kim Swales

. 2015. “Type II Errors in IO Multipliers.” https://www.strath.ac.uk/media/1newwebsite/departmentsubject/economics/research/researchdiscussionpapers/15-04.pdf

15.

Farole

Thomas

Winkler

Deborah

. 2014. “Firm Location and the Determinants of Exporting in Low- and Middle-Income Countries.” Journal of Economic Geography. https://doi.org/10.1093/jeg/lbs060

16.

Faturay

Futu

Lenzen

Manfred

Nugraha

Kunta

. 2017. “A New Sub-National Multi-Region Input–Output Database for Indonesia.” Economic Systems Research. https://doi.org/10.1080/09535314.2017.1304361

17.

Ferreira

João Pedro

Ramos

Pedro

Cruz

Luís

Barata

Eduardo

. 2018. “The Opportunity Costs of Commuting: The Value of a Commuting Satellite Account Framework with an Example from Lisbon Metropolitan Area.” Economic Systems Research. https://doi.org/10.1080/09535314.2017.1357536

18.

Flegg

Anthony T.

Tohmo

Timo

. 2016. Estimating regional input coefficients and multipliers: the use of FLQ is not a gamble. Regional Studies, 50(2), 310–325. https://doi.org/10.1080/00343404.2014.901499

19.

Flegg

Anthony T.

Mastronardi

Leonardo J.

Romero

Carlos A.

. 2016. “Evaluating the FLQ and AFLQ Formulae for Estimating Regional Input Coefficients: Empirical Evidence for the Province of Córdoba, Argentina.” Economic Systems Research. https://doi.org/10.1080/09535314.2015.1103703

20.

Flegg

Anthony T.

Tohmo

Timo

. 2013. “Regional Input–Output Tables and the FLQ Formula: A Case Study of Finland.” Regional Studies. https://doi.org/10.1080/00343404.2011.592138

21.

Flegg

Anthony T.

Tohmo

Timo

. 2019. The regionalization of national input‐output tables: A study of South Korean regions. Papers in Regional Science, 98(2), 601–620. https://doi.org/10.1111/pirs.12364

22.

Fujimoto

Takashi

. 2019. “Appropriate Assumption on Cross-Hauling National Input–Output Table Regionalization.” Spatial Economic Analysis. https://doi.org/10.1080/17421772.2018.1506151

23.

Guilhoto

Joaquim J. M

. 1998. “Análise Inter e Intra-Regional das Estruturas Produtivas das Economias do Nordeste e do Resto do Brasil: 1985 e 1992 Comparados”. Departamento de Economia, Administração e Sociologia - ESALQ - USP, Mimeo.

24.

Guilhoto

Joaquim

Azzoni

Carlos Roberto

Massaru Ichihara

Silvio

Kadota

Decio Katsushigue

Haddad

Eduardo A.

. 2010. “Matriz de Insumo-Produto Do Nordeste e Estados: Metodologia e Resultados (Input-Output Matrix of the Brazilian Northeast Region: Methodology and Results),” December. https://doi.org/10.2139/ssrn.1853629

25.

Haddad

Eduardo Amaral

Júnior

Carlos Alberto Gonçalves

Oliveira Nascimento

Thiago

. 2017. “Matriz Interestadual de Insumo-Produto para o Brasil: Uma Aplicação do Método IIOAS.” Revista Brasileira de Estudos Regionais e Urbanos 11 (4): 424–46.

26.

Haddad

Eduardo

Rodrigues Faria

Weslem

Galvis-Aponte

Luis

Hahn-De-Castro

Lucas Wilfried

. 2016. “Interregional Input-Output Matrix for Colombia, 2012.” Borradores de Economia. Banco de la Republica de Colombia. https://doi.org/10.32468/be.923

27.

Harvard Business School, and U.S. Department of Commerce, Economic Development Administration . 2020. “U.S. Cluster Mapping.” 2020. https://clustermapping.us/

28.

Helpman

Elhanan

Melitz

Marc

Rubinstein

Yona

. 2008. “Estimating Trade Flows: Trading Partners and Trading Volumes*.” The Quarterly Journal of Economics. https://doi.org/10.1162/qjec.2008.123.2.441

29.

Hewings

Geoffrey J. D.

2020. “Regional Input-Output Analysis.” WVU Research Repository, 75.

30.

Hewings

Geoffrey J. D.

Okuyama

Yasuhide

Sonis

Michael

. 2001. “Economic Interdependence Within the Chicago Metropolitan Area: A Miyazawa Analysis.” Journal of Regional Science, 195–217. https://doi.org/10.1111/0022-4146.00214

31.

Horowitz

Karen J.

Planting

Mark A.

. 2009. Concepts and Methods of the U.S. Input-Output Accounts. US Bureau of Economic Analysis. https://www.bea.gov/sites/default/files/methodologies/IOmanual_092906.pdf

32.

Ingwersen

Wesley

Young

Ben

Cashman

Sarah

. 2021. “Consumption-Based GHG Inventories for States: An Approach with USEEIO.” Presented at the meeting of the climate and materials workgroup of the northeast waste management officials’ association. https://cfpub.epa.gov/si/si_public_record_report.cfm?dirEntryId=355183

33.

IO-Snap . 2022. “Input-Output State and National Analysis Program.” 2022. https://www.io-snap.com

34.

Jahn

Malte

. 2017. “Extending the FLQ Formula: A Location Quotient-Based Interregional Input–Output Framework.” Regional Studies. https://doi.org/10.1080/00343404.2016.1198471

35.

Katz

Joseph L

. 1980. “The Relationship Between Type I and Type II Income Multipliers in an Input-Output Model.” International Regional Science Review. https://doi.org/10.1177/016001768000500103

36.

Kim

Kijin

Kratena

Kurt

Hewings

Geoffrey J. D.

. 2015. “The Extended Econometric Input–Output Model with Heterogeneous Household Demand System.” Economic Systems Research. https://doi.org/10.1080/09535314.2014.991778

37.

Kim

Tae-Jeong

Hewings

Geoffrey J. D.

. 2015. “Aging Population in a Regional Economy: Addressing Household Heterogeneity with a Focus on Migration Status and Investment in Human Capital.” International Regional Science Review. https://doi.org/10.1177/0160017613484930

38.

Kockelman

Kara M.

Jin

Ling

Zhao

Yong

Ruíz-Juri

Natalia

. 2005. “Tracking Land Use, Transport, and Industrial Production Using Random-Utility-Based Multiregional Input–Output Models: Applications for Texas Trade.” Journal of Transport Geography. https://doi.org/10.1016/j.jtrangeo.2004.04.009

39.

Kowalewksi

Julia

. 2015. “Regionalization of National Input–Output Tables: Empirical Evidence on the Use of the FLQ Formula.” Regional Studies. https://doi.org/10.1080/00343404.2013.766318

40.

Lahr

Michael L.

1998. “A strategy for producing hybrid regional input-output tables.” Conference Proceedings of the 12th International Conference on Input-Output Techniques. New York City, USA. Available at: http://www.iioa.org/conferences/12th/pdf/strategy.pdf

41.

Lahr

Michael L.

2001. Reconciling domestication techniques, the notion of re-exports and some comments on regional accounting. Economic Systems Research, 13(2), 165–179.

42.

Lahr

Michael L.

1993. “A Review of the Literature Supporting the Hybrid Approach to Constructing Regional Input–Output Models.” Economic Systems Research. https://doi.org/10.1080/09535319300000023

43.

Lahr

Michael L.

Ferreira

João Pedro

Többen

Johannes R.

. 2020. “Intraregional Trade Shares for Goods-Producing Industries: RPC Estimates Using EU Data.” Papers in Regional Science. https://doi.org/10.1111/pirs.12541

44.

Lahr

Michael L.

Mesnard

Louis de

2005. “Biproportional Techniques in Input-Output Analysis: Table Updating and Structural Analysis.” Economic Systems Research. https://doi.org/10.1080/0953531042000219259

45.

Lahr

Michael L.

Stevens

Benjamin H.

. 2002. “A Study of the Role of Regionalization in the Generation of Aggregation Error in Regional Input –Output Models.” Journal of Regional Science. https://doi.org/10.1111/1467-9787.00268

46.

Lamonica

Giuseppe Ricciardo

Chelli

Francesco Maria

. 2018. “The Performance of Non-Survey Techniques for Constructing Sub-Territorial Input-Output Tables.” Papers in Regional Science. https://doi.org/10.1111/pirs.12297

47.

Lehtonen

Olli

Tykkyläinen

Markku

. 2014. “Estimating Regional Input Coefficients and Multipliers: Is the Choice of a Non-Survey Technique a Gamble?” Regional Studies. https://doi.org/10.1080/00343404.2012.657619

48.

Leontief

Wassily

Strout

Alan

. 1963. “Multiregional Input-Output Analysis.” Edited by Barna

Tibor

. Palgrave Macmillan UK. https://doi.org/10.1007/978-1-349-81634-7_8

49.

LeSage

James P.

Pace

R. Kelley

. 2008. “Spatial Econometric Modeling of Origin-Destination Flows.” Journal of Regional Science. https://doi.org/10.1111/j.1467-9787.2008.00573.x

50.

Ingwersen

Wesley W.

Ferreira

João Pedro

Court

Christa D.

Mengming

Meyer

David

. 2022a. stateior R package, v0.1.0. https://github.com/USEPA/useeior/releases/tag/v0.1.0

51.

Ingwersen

Wesley W.

Young

Ben

Vendries

Jorge

Birney

Catherine

. 2022b. “Useeior: An Open-Source R Package for Building and Using US Environmentally-Extended Input-Output Models.” Applied Sciences 12 (9): 4469. https://doi.org/10.3390/app12094469

52.

Lindall

Scott

Olson

Doug

Alward

Greg

. 2006. “Deriving Multi-Regional Models Using the IMPLAN National Trade Flows Model.” Journal of Regional Analysis & Policy. https://doi.org/10.22004/ag.econ.132316

53.

Llano Verduras

2004. “The Interregional Trade in the Context of a Multirregional Input-Output Model for Spain.” Estudios de Economia Aplicada 22: 1–34. https://ideas.repec.org/a/lrk/eeaart/22_3_8.html

54.

Los

Bart

Timmer

Marcel P.

de Vries

Gaaitzen J.

. 2015. “How Important Are Exports for Job Growth in China? A Demand Side Analysis.” Journal of Comparative Economics. https://doi.org/10.1016/j.jce.2014.11.007

55.

Mccann

Philip

Dewhurst

John H. L. L.

. 1998. “Regional Size, Industrial Location and Input‐Output Expenditure Coefficients.” Regional Studies. https://doi.org/10.1080/00343409850116835

56.

Zhifu

Meng

Jing

Zheng

Heran

Shan

Yuli

Wei

Yi-Ming

Guan

Dabo

. 2018. “A Multi-Regional Input-Output Table Mapping China’s Economic Outputs and Interdependencies in 2012.” Scientific Data 5 (1). https://doi.org/10.1038/sdata.2018.155

57.

Miller

Ronald E.

Blair

Peter D.

. 2009. Input-Output Analysis: Foundations and Extensions. Second. Cambridge University Press.

58.

Minnesota IMPLAN Group, Inc. (MIG) . 2004. “IMPLAN Professional: User’s Guide, Analysis Guide, Data Guide.” Stillwater. 2004.

59.

Minx

J. C.

Wiedmann

Wood

Peters

G. P.

Lenzen

Owen

Scott

, et al. 2009. “Input–Output Analysis and Carbon Footprinting: An Overview of Applications.” Economic Systems Research 21 (3): 187–216. https://doi.org/10.1080/09535310903541298

60.

Oak Ridge National Laboratory (ORNL) . 2020. “Freight Analysis Framework (FAF) V4.” Oak Ridge National Laboratory (ORNL). https://faf.ornl.gov/fafweb/

61.

Oosterhaven

Jan.

1981. Interregional Input-Output Analysis and Dutch Regional Policy Problems. Gower. https://pure.rug.nl/ws/portalfiles/portal/14479685/interregionale_input-output.pdf

62.

Oosterhaven

Jan

Hewings

Geoffrey J. D.

. 2021. “Interregional Input-Output Models.” Springer. https://doi.org/10.1007/978-3-662-60723-7_43

63.

Park

J. Y.

Gordon

Moore

J. E.

II Richardson

H. W.

Wang

. 2007. “Simulating The State-by-State Effects of Terrorist Attacks on Three Major US Ports: Applying NIEMO (National Interstate Economic Model).” In H. W. Richardson, P. Gordon, and J. E. Moore II (Eds.)., The Economic Costs and Consequences of Terrorism. Edward Elgar.

64.

Pérez-Balsalobre

Santiago

Verduras

Carlos Llano

Díaz-Lanchas

Jorge

. 2019. “Measuring Subnational Economic Complexity: An Application with Spanish Data.” JRC Working Papers on Territorial Modelling and Analysis 05/2019. Seville: European Commission, Joint Research Centre (JRC). http://hdl.handle.net/10419/202275

65.

Polenske

Karen R.

1980. “US Multiregional Input-Output Accounts and Model.” https://www.osti.gov/biblio/6506788

66.

Polenske

Karen R.

Hewings

Geoffrey J. D.

. 2004. “Trade and Spatial Economic Interdependence.” Papers in Regional Science. https://doi.org/10.1007/s10110-003-0186-7

67.

Ramos

Pedro

Cruz

Luís

Barata

Eduardo

Parreiral

André

Ferreira

João-Pedro

. 2015. “A Bi-Regional (Rectangular) Input-Output Model for Portugal: Centro and Rest of the Country.” In Assessment Methodologies: Energy, Mobility and Other Real World Application, edited by Godinho

Pedro

Dias

Joana

, 1st ed., 265–85. Imprensa da Universidade de Coimbra. https://doi.org/10.14195/978-989-26-1039-9_12

68.

Rickman

Dan S.

Schwer

R. Keith

. 1995. “A Comparison of the Multipliers of IMPLAN, REMI, and RIMS II: Benchmarking Ready-Made Models for Comparison.” The Annals of Regional Science. https://doi.org/10.1007/BF01581882

69.

Round

Jeffery I.

1983. “Nonsurvey Techniques: A Critical Review of the Theory and the Evidence.” International Regional Science Review. https://doi.org/10.1177/016001768300800302

70.

Rutherford

Thomas F.

Schreiber

Andrew

2019. “Tools for Open Source, Subnational CGE Modeling with an Illustrative Analysis of Carbon Leakage.” Journal of Global Economic Analysis. https://doi.org/10.21642/JGEA.040201AF

71.

Sargento

Ana L. M.

2009. “Regional Input-Output Tables and Models : Interregional Trade Estimation and Input-Output Modelling Based on Total Use Rectangular Tables.” https://estudogeral.sib.uc.pt/handle/10316/10120

72.

Sargento

Ana L. M.

Ramos

Pedro Nogueira

Hewings

Geoffrey J. D.

. 2012. “Inter-Regional Trade Flow Estimation Through Non-Survey Models: An Empirical Assessment.” Economic Systems Research. https://doi.org/10.1080/09535314.2011.574609

73.

Stevens

B. H.

Trainer

G. A.

. 1980. Error Generation in Regional Input-Output Analysis and Its Implications for Nonsurvey Models. In: Pleeter, S. (eds) Economic Impact Analysis: Methodology and Applications. Studies in Applied Regional Science, vol 19. Springer. https://doi.org/10.1007/978-94-011-7405-3_5

74.

Szabó

Norbert

. 2015. “Methods for Regionalizing Input-Output Tables.” Regional Statistics. https://doi.org/10.15196/RS05103

75.

Timmer

Marcel P.

Dietzenbacher

Erik

Los

Bart

Robert Stehrer de Vries

Gaaitzen J.

. 2015. “An Illustrated User Guide to the World Input–Output Database: The Case of Global Automotive Production.” Review of International Economics. https://doi.org/10.1111/roie.12178

76.

Towa

Edgar

Zeller

Vanessa

Achten

Wouter M. J.

. 2020a. “Input-Output Models and Waste Management Analysis: A Critical Review.” Journal of Cleaner Production. https://doi.org/10.1016/j.jclepro.2019.119359

77.

Towa

Edgar

Zeller

Vanessa

Merciai

Stefano

Schmidt

Jannick

Achten

Wouter M. J.

. 2020b. “Toward the Development of Subnational Hybrid Input–Output Tables in a Multiregional Framework.” Journal of Industrial Ecology. https://doi.org/10.1111/jiec.13085

78.

Treyz

George I.

Rickman

Dan S.

Shao

Gang

1991. “The REMI Economic-Demographic Forecasting and Simulation Model.” International Regional Science Review. https://doi.org/10.1177/016001769201400301

79.

Treyz

Stevens

B. H.

. 1979. Location analysis for multiregional modeling. Regional Science Research Institute. N° 113.

80.

Treyz

George I.

Stevens

Benjamin H.

. 1985. The TFS regional modelling methodology. Regional Studies, 19(6), 547–562.

81.

Tsosie

Clare

. 2017. “Why so Many Credit Cards Are from Delaware.” Forbes. https://www.forbes.com/sites/clairetsosie/2017/04/14/why-so-many-credit-cards-are-from-delaware/?sh=64b8fe411119

82.

US Bureau of Economic Analysis (BEA) . 2019a. “BEA Summary Import Matrices Before Redefinitions.” https://apps.bea.gov/industry/xls/io-annual/ImportMatrices_Before_Redefinitions_SUM_1997-2019.xlsx

83.

US Bureau of Economic Analysis (BEA) . 2019b. NIPA Handbook: Concepts and Methods of the U.S. National Income and Product Accounts. https://www.bea.gov/system/files/2019-12/All-Chapters.pdf

84.

US Bureau of Economic Analysis (BEA) . 2019c. “Input-Output Accounts Data.” https://www.bea.gov/industry/input-output-accounts-data

85.

US Bureau of Labor Statistics (BLS) . 2020. “Employment Projections Methods Overview.” 2020. https://www.bls.gov/emp/documentation/projections-methods.htm#industry_employment

86.

US Bureau of Economic Analysis (BEA) . 2021. “State Personal Income and Employment: Concepts, Data Sources, and Statistical Methods.” https://www.bea.gov/system/files/methodologies/SPI-Methodology.pdf

87.

US Census Bureau . 2020a. “Annual Survey of State Government Finances.” 2020. https://www.census.gov/programs-surveys/state.html

88.

US Census Bureau . 2020b. “Employment Projections Methods Overview.” 2020. https://www.census.gov/foreign-trade/data/index.html

89.

US Environmental Protection Agency . 2018. “National Biennial Hazardous Waste Report 2017.” 2018. https://www.epa.gov/hwgenerators/biennial-hazardous-waste-report

90.

US Environmental Protection Agency . 2020. “US Environmentally-Extended Input-Output (USEEIO) Models.” https://www.epa.gov/land-research/us-environmentally-extended-input-output-useeio-models

91.

US Energy Information Administration . 2021. “United States - SEDS - US Energy Information Administration (EIA).” 2021. https://www.eia.gov/state/seds/seds-data-complete.php?sid=US

92.

US General Services Administration, and Federal Acquisition Services . 2020. “Federal Procurement Data System Product and Service Codes (PSC) Manual.” https://www.acquisition.gov/sites/default/files/manual/PSC_Manual_October_2020.pdf

93.

US Environmental Protection Agency, Office of Land, and Emergency Management . 2009. “Sustainable Materials Management: The Road Ahead.” Collections and Lists. 2009. https://www.epa.gov/smm/sustainable-materials-management-road-ahead

94.

US Environmental Protection Agency, Office of Research, and Development . 2020. “EPA Researchers Working to Improve Life-Cycle Assessment Capabilities for Communities.” July 22, 2020. https://www.epa.gov/sciencematters/epa-researchers-working-improve-life-cycle-assessment-capabilities-communities

95.

USAspending.gov . 2020. “Government Spending Open Data USAspending.” https://usaspending.gov/

96.

Varlamova

Larionova

. 2015. “Macroeconomic and Demographic Determinants of Household Expenditures in OECD Countries.” Procedia Economics and Finance. https://doi.org/10.1016/S2212-5671(15)00686-3

97.

Wang

Haikun

Zhang

Yanxia

Nielsen

Chris P.

Jun

. 2015. “Understanding China׳s Carbon Dioxide Emissions from Both Production and Consumption Perspectives.” Renewable and Sustainable Energy Reviews. https://doi.org/10.1016/j.rser.2015.07.089

98.

Wiedmann

Thomas O.

Schandl

Heinz

Lenzen

Manfred

Moran

Daniel

Suh

Sangwon

West

James

Kanemoto

Keiichiro

. 2015. “The Material Footprint of Nations.” Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.1220362110

99.

Wiedmann

Thomas

Wood

Richard

Minx

Jan C.

Lenzen

Manfred

Guan

Dabo

Harris

Rocky

. 2010. “A Carbon Footprint Time Series of the Uk – Results From a Multi-Region Input–Output Model.” Economic Systems Research, https://doi.org/10.1080/09535311003612591

100.

Wisconsin National Data Consortium . 2021. “WINDC 3.0.” July 2021. https://windc.wisc.edu/

101.

Yang

Ingwersen

Wesley W.

Hawkins

Troy R.

Michael Srocka Meyer

David E.

. 2017. “USEEIO: A New and Transparent United States Environmentally-Extended Input-Output Model.” Journal of Cleaner Production 158: 308–18. https://doi.org/10.1016/j.jclepro.2017.04.150

102.

Yang

Ingwersen

Wesley W.

Meyer

David E.

. 2018. “Exploring the Relevance of Spatial Scale to Life Cycle Inventory Results Using Environmentally-Extended Input-Output Models of the United States.” Environmental Modelling and Software. https://doi.org/10.1016/j.envsoft.2017.09.017

103.

Zhang

Chen

Z. M.

Xia

X. H.

X. Y.

Chen

Y. B.

. 2013. “The Impact of Domestic Trade on China’s Regional Energy Uses: A Multi-Regional Input–Output Modeling.” Energy Policy 63: 1169–81. https://doi.org/10.1016/j.enpol.2013.08.062

104.

Zhuang

Xin

Balassiano

Kim

. 2021. “EPA Data Commons V0.1.” http://edap-data-commons.s3.amazonaws.com/data_commons_search.html