Using Commodity Flow Survey Microdata and Other Establishment Data to Estimate the Generation of Freight, Freight Trips, and Service Trips: Guidebook (2016)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

19 The modeling approach adopted in this guidebook is a major improvement over traditional approaches. It has solid conceptual foundations, and more importantly, produces estimates that are more accurate than alternative methodologies. (For a comparison, see Holguín-Veras et al. 2013a.) However, this does not mean that these models are perfect; far from it. FSA is the result of complex interactions between the economic factors that determine the production and consumption of freight and services, and a host of logistical decisions. These interactions cannot be captured by simple models, such as the ones in this guidebook. The research conducted by NCFRP Project 25(01) team members confirms this assertion. Sánchez-Díaz et al. (2014) used advanced econometric techniques to estimate two sets of models. The first set considered traditional independent variables such as employment and other establishment attributes. The second set considered spatial variables such as proximity to similar businesses and large population centers. The difference in the explanatory power of the models was remarkable. The coefficients of determination (R2) in the traditional models—in the range of 0.10–0.20—jump to 0.76–0.94 with the inclusion of spatial variables and the use of spatial econometrics. The use of spatial proxies of the determinants of economic activity leads to models that provide a better explanation of FSA; however, using spatial econometric techniques and spatially defined independent variables is beyond the reach of most practitioners. Data availability also is an issue. In applications such as site impact analysis and freight demand fore- casts, it is not possible to even guess the values that spatially defined variables are likely to take. Simpler (though admittedly less accurate) models are needed. These considerations led the team to estimate the simplest conceptually valid models using employment as the sole independent variable because (1) employment is an expression of the economy (represented by number of employees) and (2) data about employment at different levels are publicly available, making it easier for practitioners to apply the models. Among their numerous advantages, these models can use the employment estimates produced by federal agencies such as the Census Bureau to estimate FSA at a fine level of detail. More- over, they establish a direct connection with the Commodity Flow Survey (CFS), an underused resource that could be further exploited for transportation modeling purposes. The analyses conducted by the research team led to the conclusion that the guidebook models should also be: • Establishment-level (disaggregate) models. Estimating FSA at the establishment level leads to more accurate models because there is a more direct connection between FSA and employment. Importantly, these estimates can then be aggregated to larger levels of geography (e.g., city block, ZIP code, and transportation analysis zones [TAZs]) using suitable aggregation procedures. C h a p t e r 5 Modeling Principles

20 Using Commodity Flow Survey Microdata and Other establishment Data to estimate the Generation of Freight, Freight trips, and Service trips • Economic-based. Instead of using variables like square footage that do not measure the intensity of the activity performed at the establishment, the models use employment, which correctly measures the intensity of the use of space, which leads to better forecasts of FSA. • Applicable to any land use classification system. The models’ disaggregate and economic- based nature allows their use regardless of how the land use classes are defined. Model trans- ferability is thus enhanced, because land use classes change from city to city, whereas the FSA for an industry sector is significantly more stable across the country. Establishment-Level (Disaggregate) Models The guidebook models are based on the fundamental principle that, for a model to adequately predict transportation demand, it must correctly capture the underlying processes that generate the demand. In the case of FSA, these economic and logistical processes take place at the establishment level. This emphasis reflects transportation modeling experience, which unambiguously shows that disaggregate models are better able to correctly capture the determinants of transportation demand. These models are very efficient because they (1) require smaller samples than their aggregate counterparts, (2) establish a direct connection between the attributes of the establishment and the measure of FSA that is being estimated, and (3) could be aggregated to any level of geography. Economic-Based Nature of the Models The guidebook models use economic variables to estimate FG, FTG, and STG. For the follow- ing reasons, the use of economic variables enables the models to produce better estimates of FSA, and increases their ability to work well in different land use patterns: • The amounts of cargo consumed and produced at an establishment constitute the inputs and outputs of an economic process. For that reason, employment and the industry sector are better predictors of FSA than are variables like square footage, which denotes little about the activity taking place at the establishment (see Holguín-Veras et al. 2013a). • The economic nature of the models allows for a differentiation between FG and FTG. This is significant because, whereas FG is the output of an economic process, FTG is the output of logistical decisions concerning frequency of deliveries and shipment sizes. As a result, no one- to-one correlation exists between FG and FTG, because increasing shipment sizes could allow carriers to transport larger amounts of FG without necessarily increasing FTG. This distinction provides a more nuanced view of how operational or policy changes might alter freight activity. Given the economic-based nature of the models, it is important to use a formal economic classification system of industrial activities. Using these systems improves model quality because the establishments within a group share common characteristics, which reduces the internal variability of the data in that group. The NAICS is used as the system of classification because: • it enables direct use of official statistics—particularly employment—that are regularly released by agencies such as the U.S. Census Bureau and the Bureau of Transportation Statistics, and • it provides a comprehensive classification of all the economic activities that can take place in the formal economy, including in freight intensive and service sectors. Applicability to Various Land Use Configurations A key principle that guided model development was the desire for the resulting models to be applicable irrespective of the prevailing land use patterns. This was accomplished by estimating models that do not use land use variables. To this effect, the team exploited the disaggregated

Modeling principles 21 and economic nature of the models. Instead of estimating FSA models based on land use patterns, the team estimated the models by industry sectors. This decision is advantageous in multiple ways, such as the following. • The FSA generation patterns by industry sector are more stable than those associated with land use. Land use patterns can change radically from city to city, but FSA patterns are determined largely by industry-wide practices and regulations, which reduces variability. • Industry-based FSA models can be folded into any land use pattern. Given that land use ordinances regulate the economic activities that are possible at any given zone, FSA models by industry sector can easily be mixed to replicate any land use pattern. Undoubtedly, a city block in a highly dense commercial area in a large city is likely to generate a different level of FSA compared to a city block in a mid-size or small city. Close examination of the FSA generation patterns at the establishment-level reveals, however, that these patterns are relatively similar. This is because establishments in the same lines of business tend to use similar technologies and operational practices. True, differences in FSA occur on account of land use values, proximity to arterials, and other local factors (Sánchez-Díaz et al. 2014). For practical purposes, however, the assumption that FSA patterns are the same for all establishments in the same industry sector is reasonable. Differences in FSA patterns that are observed beyond the establishment level, such as at the level of city block and buildings, are likely to result from factors such as the density of commercial establishments, the mix of industry sectors present in the area, and the logistical adjustments made by vendors to deliver supplies to environments of various densities.