Report #5
Methodology for Regionalization of 1997 Out-of-Scope Truck Commodity Flows

1. Introduction

The Freight Analysis Framework (FAF) program is developing a consistent base year 2002 set of inter-regional commodity flows. The FAF three-dimensional matrix (by commodity, region, and transportation mode) of freight flows will establish a primary base for the analysis of all freight shipments. The core component of the FAF is the Commodity Flow Survey (CFS) supplemented by other data sources. The CFS, however, did not include traffic flows originating for several "Out-of-Scope" business sectors. Specifically, truck traffic for Farm based, Fisheries, Logging, Construction, Services, Publishing and Retail activities, totaling 257 trillion ton miles (Table 1) in 2002 where not sampled. These national freight flows in Table 1 were estimated as part of the FAF program. For the 1997 National Out-of-Scope totals were developed following methods developed during the 2002 FAF base case data development process. This paper presents an approach to develop FAF region-to-FAF region commodity flows consistent with national estimates. Figure 1 shows a map of the 114 FAF regions.

Table 1. National Estimate of Truck Shipments of Out-of-Scope Economic Activities 2002

Sector	Value ($ millions)	Tons (thousands)	Ton Miles (millions)	Average shipping distance (miles)
Farm based	200,646	1,051,285	40,222	38.26
Fisheries	3,181	4,714	259	54.94
Logging	7,871	350,191	16,271	46.46
Construction	924,974	591,449	62,003	104.83
Services	284,601	277,413	30,500	109.94
Publishing	98,657	32,330	13,945	431.33
Retail	1,408,236	1,050,277	94,411	89.89
Total	2,928,166	3,357,659	257,611	76.72

Figure 1. 2002 Freight Analysis Framework Regions, Including Gateways.

Notes: Metropolitan areas shown in green, gateways (unnumbered) in purple. Not shown are Alaska (3), Honolulu (26), Hawaii (27), and the Anchorage Gateway.

Source: Southworth and Peterson, Generation of a U.S. Commodity Flows Matrix Using Log-Linear Modeling and Iterative Proportional Fitting, (Freight Analysis Framework Report 2), August 8, 2005.

In order to generate an expedient and reasonable regionalization of out-of scope commodity flows, one needs to reflect the relative regional differences in economic activity that generate the truck commodity flows using readily and openly available data on state and local economic activity. The regionalization of national truck freight flows followed in the development of the FAF2002 Out of Scope analysis used a three step process defining, (1) the allocation of the national freight estimates to county in which the freight generation occurs, (2) the estimation of county-to-county freight flows for each commodity shipped in the out-of scope business sectors, and (3) then the aggregation of the county-to-county flows to regional commodity flows used in the FAF matrix. This same procedure is used for the 1997 estimates, following the steps used in the 2002 estimates as closely as possible. However, because of variations in data availability several modifications were required and are described below.

This paper outlines the approach used for the regionalization to FAF regions of national estimates of out-of-scope commodity flows by first generating county-to-county information. Two features of the FAF region-to-region spatial context indicate a county-to-county analysis as a necessary intermediate step in developing the FAF region commodity flows. First the freight analysis flows to be regionalized appear to be relatively short local hauls with an average distance of under 100 miles. Key flows of interest are the FAF-to-FAF region flows, which include urban area flows to the rest-of-state (regions). However, there are many instances where several urban areas interact with the same "rest-of-state" area, in which the region-to-region distances are defined by the centroid of a relatively large rest of state region. To estimate the interregional flows of such local movements, it is useful to have more spatial discrimination, as in a county-to-county matrix to capture the movements within and around the urban metroplex represented by the short haul out-of-scope activities. The county-to-county flows can then be aggregated to the desired FAF regional scale.

2. Procedure

2.1 Regional Economic Freight Generation

The national estimates are allocated to the county based on local traffic generating activity, but taking into account State variations in truck usage by the business sector as identified in the Vehicle Inventory and Use Survey (VIUS). The VIUS provide annual freight truck miles generated by major business sectors in each State as well as providing commodity (two digit SCTG code) totals for the state for the year 2002. The approach is to allocate the national total of out-of-scope freight by each commodity and business to the states based on the states’ share of national activity as depicted in VIUS or other source when VIUS data are not applicable. The estimate for each state is then allocated to the counties within the state based on an appropriate measure of the economic activity that generates the freight activity. The county weight will typically was based on the 2002 County Business Patterns for the 2002 estimates. However, for the 1997 estimates the 1998 County Business Patterns is used. The 1997 CBP was based on the earlier Standard Industrial Classification (SIC) industry definitions. As an expedient for the estimates of the 1997 FAF flows, the 1998 CBP was used as it is the first year of the North American Industry Classification System (NAICS) definitions were used in the CBP and also consistent with the base used in generating the 2002 FAF flow estimates. It was felt that the difference in the year was less than the error that would be introduced by converting 1997 SIC to 1997 NAICS at the detailed level we were interested in.

This two step process can be implemented as a single allocation using a set of allocation coefficients (i.e., a nation to state and a state to county coefficients). For a specific commodity carried by an out-of-scope business sector the allocation can be represented as:

Equation 1:
Truck Tonnage Originating in County I for the business sector = 
     (County’s share of state’s earnings for the business sector) X
          (State’s share of national truck miles for the business sector) X
              (National tonnage estimate for the business sector)

Depending on the state and county data available for allocation four different strategies are considered. The first strategy (focusing on the business sector) used for the Construction, Services and Retail trade sectors, uses the business sector information on truck freight mileage from VIUS. Although in each business sector a variety of commodities are shipped and need to be regionalized, the freight generation activity and market area in which the freight movements take place is similar and thus the same allocation strategy would be reasonable to regionalize each commodity carried by the business sector. Thus the same share coefficient will be used for all commodities carried by a specific business for this first strategy. The coefficient is the truck mileage carried by the business sector in the state as compared to the national total. In this strategy the state estimate is then allocated to the counties in the state using an allocation weight reflecting the business sector’s local activity as reflected by employment or earnings in the County Business Patterns employment or earnings. This approach was used for both the 2002 and the 1997 estimates. (The candidate allocation variables for this case, "VIUS Sector," are presented in the first set of rows of Table 2.)

The second strategy focuses on commodity characterization for the three farm-based commodities and Logging. In these situations the commodity mix of the appropriate aggregate business sector may vary considerably from state to state and a more refined focus is required. For both the farm-based activities and logging, additional information in VIUS on commodities carried within each state can be used to assist in the regionalization rather than the more aggregate business sector information. In addition county commodity production data (Census of Agriculture and county round wood production from the National Forest Service) is available to better characterize the sub-state regionalization as well. The candidate variables for the regionalization of this set of sectors, "VIUS commodities" is presented in the section set of rows in Table 2.

This approached was used for the 2002 estimates, but was not available for the 1997 estimates. The 1997 VIUS data on commodity shipments did not include truck miles but only number of trucks, which would provide a quite different state allocation result. However, we did not wish to hold the state shares constant by using the 2002 ratios, so we decided to use a different weighting variable. In each of the commodities that we used the VIUS commodity, we had a county series to do the final allocation. The state summary of the county series was used to allocate the national totals to the state.

The third strategy involves Fisheries and Printing both of which are special cases. Fisheries are part of the farm based business sector and Live Fish as a commodity is part of the SCTG 01 (Live animals and live fish). Both the sector and the commodity are a small part of the total, and the importance in the sector may very greatly from one state to another. Consequently, it seems more reasonable to develop a set of allocation variable that are more tightly related to the regional level of activity in the sector. The case for printing is similar in that the appropriate sector (information) is too broad to capture the nature of the printed material considered in the truck shipments. Moreover, the data coverage on State level truck shipments of the commodity printed materials in VIUS is very limited (less than 20%). In this case the County Business Patterns information on employment or earnings appears as a reasonable proxy for both the state and county level regionalization. The candidate variable for regionalization of this set of sectors (Other) is presented in the third set of Table 2.

Table 2. Candidate Variables for National to State and State to County Allocations

Type of Allocation	Business Sector	Commodity	State Allocation	State to County Allocation
1. VIUS Sector	Construction	All	VIUS Sector Activity	CBP Sector Employment
1. VIUS Sector	Services	All	VIUS Sector Activity	CBP Sector Employment
1. VIUS Sector	Retail	All	VIUS Sector Activity	CBP Sector Employment
2. VIUS Commodity	Farm Based	Animals	VIUS Commodity	Value in Farm Sales (USDA)
2. VIUS Commodity	Farm Based	Cereal	VIUS Commodity (2000) Value in Farm Sales (1997)	Value in Farm Sales (USDA)
2. VIUS Commodity	Farm Based	Other Agriculture	VIUS Commodity (2000) Value in Farm Sales (2000)	Value in Farm Sales (USDA)
2. VIUS Commodity	Logging	Logs and Other Wood	VIUS Commodity (2000) Round Wood Production	Round Wood Production (NFS)
3. Other	Printing	Printed Materials	CBP Industry Employment	CBP Industry Employment
3. Other	Fisheries	Live Fish	CBP Industry Employment	CBP Industry Employment
3. Other	Household moves	All	none	County-to-county migration

Household and Business sector uses a fourth strategy, which uses the Census information on county-to-county migration during the period 1990 to 2000 to identify relative regional growth and decline upon which to allocate the national flows. The same migration data was used for both the 2002 and 1997 estimates.

2.2 Estimation of Local Market Commodity Flows

The second step (used for both the 1997 and the 2002 estimates) is to expand the freight generation at county origin to the destination flow. As discussed in the introduction, the out-of-scope truck traffic examined here, appear to reflect short haul movements that are likely to remain within the local market area. With this view, we defined a reasonable market area for each origin, business sector, and commodity and then estimate the Market Potential. We then allocated the total freight to each of the flows in the market in proportion to that flow’s contribution to the total market potential. As an expedient, we selected a proxy variable that we expect is proportionate to the market metric. Table 3 augments Table 2, by adding a candidate economic activity variable for each of the Business Sectors in the Out-of-Scope activities to be allocated to local commodity flows.

Table 3. Candidate Variables for National to State and State to County Allocations for Market Potential

Type of Allocation	Business Sector	Commodity	State Allocation	State to County Allocation	Market Potential
1. VIUS Sector	Construction	All	VIUS Sector Activity	CBP Sector Employment	CBP Sector Employment
1. VIUS Sector	Services	All	VIUS Sector Activity	CBP Sector Employment	CBP Sector Employment
1. VIUS Sector	Retail	All	VIUS Sector Activity	CBP Sector Employment	Population
2. VIUS Commodity	Farm Based	Animals	VIUS Commodity	Value in Farm Sales (USDA)	CBP Animal Slaughtering and processing Employment
2. VIUS Commodity	Farm Based	Cereal	VIUS Commodity	Value in Farm Sales (USDA)	CBP Grain and Oil Seed milling Employment
2. VIUS Commodity	Farm Based	Other Agriculture	VIUS Commodity	Value in Farm Sales (USDA)	CBP Food Manufacturing Employment
2. VIUS Commodity	Logging	Logs and Other Wood	VIUS Commodity	Round Wood Production (NFS)	CBP Wood Products Employment
3. Other	Printing	Printed Materials	CBP Industry Employment	CBP Industry Employment	Population
3. Other	Fisheries	Live Fish	CBP Industry Employment	CBP Industry Employment	CBP Seafood Products Employment

For each origin activity to be allocated, all destinations with 350 hundred miles (over twice the typical average distance)[1] will be considered. The value of each destinations market potential (candidate variable) discounted by a distance (using a specified lambda value) was summed to determine the Total Market Potential for that decay value. Each flow was then allocated a proportionate share of the generated freight based on its contribution to the market potential.

Flow(i,j) = ( [P(j)/d(i,j)^λ] ÷ ∑[P(j)/d(i,j)^λ] ) X F(i)

where:

F(i) is freight at origin i to be allocated.
P(j) is the market potential at destination j
[P(j)/d(i,j)^λ] is the contribution to the total market potential for i derived from the interaction with j.
lamda (λ) is the distance decay coefficient for the potential model.
∑ [P(j)/d(i,j)^λ] is the total market potential at i from all potential destinations.

Flow(i,j) is the (allocated) proportionate share of the freight activity assigned between i and j.

The importance of distance (lambda) in the above equation is directly related to the average mile shipped estimation that characterizes the market area. The larger the value of lambda, the more resistance from distance and the smaller the market radius and average distance shipped. It is typical to use a Spatial Interaction Model for determining the lambda and spatial flows consistent with an average distance shipped. The flows are organized as a large matrix and then through iterative computations using various Lambdas, the value that brings about the target average flow is determined. This approach is difficult in the FoxPro environment that was used to generate the other regional allocations. As an expedient, a two step procedure was adopted. First a set of lambdas covering the typical range was selected (0.5, 1.0, 1.5, 2.0, 2.5 and 3.0) and market calculations were done for each of the lambda values. Then in the second stage, the results were visually examined and the county flows set whose market potential data generated the average distance that was closest to the target national values for that business sector was chosen as the appropriate lambda value for that business sector. The sector lambda was then used to select the associated set of flows for all commodities considered for the business sector.

As the general market of interest is for the shipments of the business sector, we tuned to a single lambda for each business sector. The exception was farm based activity, because of the potentially very different regional patterns of market area for the different commodities in the sector. We also used different variables to determine the market areas, where as in the other sectors the same variable was used to determine the market area for all commodities. The target "national" average shipping distance reflected the FAF National Estimates of the Out-of-Scope Activities.

In most cases, at least one of the computed distance values (i.e., lambda) for each business sector was close to the target value, and the estimates associated with it were used. Because the average shipping distance of publishing and household & business moves were considerably larger than the other sectors, it was assumed that these two activities had a national scope as opposed to the local (under 350 miles) use for the rest of the sectors. In the case of publications, we used the same methodology as the other sectors, except that we included all counties thus permitting long distance shipments based on the potential model. Including all counties did stress the software/hardware set we were using, and we could not run the full set of lambda’s simultaneously as we did for the other sectors. As an alternative we elected to start with the less restrictive decay and then progress till we matched the target.

For the household and business moves, it was thought that the simple potential model might not capture some of the significant changes in recent regional growth that might stimulate inter-regional moves. The recent availability of the county-to-county migration flows (CENSUS 2000) between 1990 and 2000 provide a snapshot of such recent changes. The relative county-to-county flows provide an alternative measure to a standard market potential that does reflect changing regional population growth that induces household and business movement of household goods and business equipment. The average distance of this method is approximately 485 miles somewhat higher than the national estimate but given the approximate nature of both the national values and regional allocation not too extreme to reject the approach. As time and effort warrant some alternative analysis of this sector would be of interest.

The step above provides commodity flow tables that have values for each of the key county-to-county freight flow characteristics: business sector, commodity, origin-county (FIPS code), destination-county (FIPS code), distance from origin to destination, tons trucked. Then for each commodity, the national value-to-tons ratio is applied to the allocated freight tons to derive the associated freight value for each of the estimated flows.

As part of the project, ORNL developed a cross walk from the FIPS county codes to the FAF region codes. Using this cross walk the county-to-county flow table is augmented with the appropriate origin FAF Region and the destination FAF Region codes. The resultant file is then aggregated to a "Matrix" table with the FAF commodity flow matrix dimensions: FAF origin, FAF destination, Business Sector, SCTG 2 digit Commodity, Truck Mode. The measures for each FAF to FAF truck flow include the Freight tons, Freight Value, and the average ton miles for that flow.

This "out-of-scope" truck freight matrix can then easily be integrated into the full FAF Commodity Flow Matrix.

3. Data Sources used in the Out-of-Scope Regionalization:

1. U.S. Census Bureau, 1997 and 2000 Economic Census, Vehicle Inventory and Use Survey (VIUS), Geographic Area Series.

2. U.S. Census Bureau, 1998 and 2002 County Business Patterns, Census Bureau.

3. U.S. Census Bureau, 2000 CENSUS of Population, County-to-County Migration Flow Files.

4. U.S. Census Bureau, County population and estimated components of population change, all counties: April 1, 2000 to July 1, 2004.

5. U.S. Department of Agriculture, 1997 and 2002 Census of Agriculture.

6. Bureau of Economic Analysis, 2002 Regional Economic Information System (REIS), (county population and income).

7. National Forest Service, 1997 and 2002 Round Wood Production.

8. 1997 and 2002 National Estimates of Out-of-Scope Freight, Freight Analysis Framework working papers.

9. County-to-county highway distances, ORNL.

10. County-to-FAF region cross walk, ORNL.