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.
45 5. FIRM-LEVEL SPATIAL ANALYSIS We conducted analysis of firm-level data for two regions with rail, Dallas-Fort Worth and Portland, Oregon. We chose these two regions based on initial diagnostic regression analysis of block-level worker-by-industry data from the LEHD database. We aimed at understanding how relatively new transit systems have affected the pattern of spatial development in a region over time. We had initially intended to obtain firm-level data from the State of New Jersey, which we had used in the past. However, due to budget cuts in the Department of Labor, we were unable to obtain these data. Instead, we purchased a large database for two regions: Portland, OR, and Dallas, TX, consisting of a twenty-year time-series of geo-coded firm-level data on workers and retail sales. These data do not include information on average wages or firm revenues, which are required for an explicit productivity analysis. Instead, our analysis focused on how the location of stations and the introduction of light rail transit influenced the location of firms, as well as both the number of employees in each firm and their total sales. Although we were unable to make an explicit linkage to any changes in productivity associated with the transit systems, the spatial effects are of interest to that question because they address the nature of the transit-to- agglomeration link, which is the precursor to most of the expected productivity effects of transit capacity investments. The firm-level analysis gives support to some of our case study results and provides additional evidence as to how different regions have leveraged or failed to leverage the potential of their transit systems. The dataset used here is a complete set of Dun & Bradstreet (D&B) reports for every firm that is or was located within the metropolitan areas between 1990 and 2009. These areas are the three counties in each region served by light or commuter rail. Specifically, these are Collin, Dallas, and Tarrant Counties in Texas and Clackamas, Multnomah, and Washington Counties in Oregon. D&B time-series used herein are the estimates of each firmâs annual number of employees and the dollar value of its sales. These data may be directly reported by the firm or estimated based on economic census data, proprietary modeling by Dun & Bradstreet, or imputation by the data vendor, Walls & Associates. They are available for every year that the company is in business in the metropolitan area. Along with these data are included a geocode (latitude and longitude) and a six-digit industrial classification (NAICS) code, again updated annually. After data cleaning, the dataset included 1,025,441 firms in the Dallas-Fort Worth area and 336,158 in the Portland area. From there, geocodes were mapped using ArcGIS and spatially joined to the appropriate block, as defined by Census 2000. These data were then aggregated at the block level while being disaggregated by two-digit NAICS sector. The result was industry-specific employment and sales counts for each census block by year. From there, the distances from the centroid of each block to the central business district of the respective metropolitan area and the nearest rail station that was open in the corresponding year were calculated for all blocks, of which there were 60,923 in the Dallas-Fort Worth area (meaning that, as the data is a 20-year panel in long form, there are 1,218,460 observations) and 28,270 (565,400 observations) in the Portland area. Using block-level data, four ratios were computed: employees per acre, employees per firm, sales per firm, and sales per employee. The first of these was computed for each NAICS sector as well as across all sectors, while the other three were only computed for all firms. These ratios became the dependent variables in panel regressions. Fixed-effects and random-effects models were specified with independent variables consisting of dummy variables representing each year of the data (except 1990, the reference year) and rail station distance.
46 With regard to the latter, six variables were specified identifying [1] blocks whose centroid is located within ¼ mile of a station situated in the central business district, [2] within ½ mile of a CBD station, [3] within one mile of a CBD station, [4] within ¼ mile of a non-CBD station, [5] within ½ mile of a non-CBD station, and [6] within one mile of a non-CBD station. (Naturally, the reference category is all blocks not within one mile of a rail station.) These terms are mutually exclusive, in that blocks are assigned to the âclosestâ range that applies to it and that only the station closest to the centroid is included in the analysis. Central business districts are defined according to the transit agencies themselves; Dallas Area Rapid Transit (DART) identifies a âDowntown Dallasâ area on its route map, while Tri-Met designates a city center âFree Ride Zoneâ in Portlandâs urban core. Straight-line distance to the central business district in miles was also included in random-effects models only. A summary of findings is provided on the next page, followed by statistical outputs in the appendices. Results In the first set of analyses using the variables above, we specified a panel model that uses time-series econometric techniques to better compute correlations over the course of the study period. This was done first using what is known as a fixed-effects model, which imposes statistical constraints on the model by assuming the independent variables are non-random, and a generalized least squares or random-effects model that does not have this constraint. Using a Hausman test, it was determined that the data are in fact non-random in all cases and, therefore, that the fixed-effects model is more appropriate. Results for Dallas and Portland contrast substantially. In Dallas, the presence of transit stations was largely negatively correlated with three of the four dependent variables: employees per acre, employees per firm, and sales per firm. For instance, being located within a quarter- mile of a CBD transit station in Dallas was associated with a reduction of 20 employees per acre and the $900,000 in retail sales per firm. However, there was a $11,768 increase in sales per employee for businesses located within a quarter-mile of a non-CBD rail station. In Portland, results were far more ambiguous. Most coefficients were not statistically significant, though as in Dallas there was again a positive correlation with rail proximity and sales per worker. Detailed output is available in Appendix J. From here, it was decided that sector-specific analysis would be appropriate, specifically as it pertained to employees per acre. Hence, the ratio was computed for workers employed in the 20 two-digit NAICS categories, again in both Dallas and Portland. There was indeed substantial variation across industries, as detailed in Table 11 below and in the regression outputs in Appendix J. Clearly, there is no single decisive trend; overall, a majority of coefficients reported an absence of a statistically significant relationship. This was especially true in Portland, where none of the sectors indicated a strong correlation in either direction. (A correlation is considered to be âstrongâ if a majority of dummy variables indicate the same directional relationship.) The only finding consistent for the two metropolitan areas is that rail access has no correlation with manufacturing employees per acre.
47 TABLE 11 Impacts of rail stations on employees per acre by two-digit NAICS sector, 1990- 2009 NAICS Sector Dallas Portland NAICS Sector Dallas Portland Agriculture Ambiguous Slightly Neg. Real Estate Positive Slightly Neg. Mining Positive None Prof. Services Positive Slightly Neg. Utilities Negative None Management Ambiguous None Construction Positive Slightly Neg. Administration Ambiguous None Manufacturing None None Education Ambiguous None Wh. Trade Slightly Neg. Slightly Pos. Health Care Positive Ambiguous Retail Trade Slightly Neg. Ambiguous Arts & Ent. Positive None Transportation Slightly Neg. None Hotels/Dining Positive None Information Slightly Neg. None Other Services Positive None Fin. & Ins. Ambiguous None Public Admin. Slightly Neg. None Finally, we carried out some regressions cross-sectionally, rather than as a panel, to measure change over time as a single phenomenon rather than a year-to-year one. The dependent variables of the ordinary least squares regressions were the change in each of the original measurements between 1996, the year Dallas inaugurated its rail system, and 2009, the most recent year in the dataset. Independent variables were the six rail dummies plus the straight-line distance to the central business district (in miles). Findings indicated that, in Dallas, presence of rail stations depressed (in increasing order of magnitude) employees per acre, employees per firm, and sales per firm. Across all three, however, being located within ¼ mile of a CBD rail station had strong negative impacts. There appeared to be no impact on sales per employee in Dallas. Meanwhile, in Portland, the effect on employees per acre was positive within ½ mile of CBD stations, while the effect on employees per firm was consistently negative (implying decreasing firm size near rail stations, but more firms), and no impact was found with regard to sales.
