Interactions Between Transportation Capacity, Economic Systems, and Land Use (2012)

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.

26 C h a p t e r 5 Structure of regressions Focus on Direct Job Impact Regression analysis is a statistical technique used to calculate the magnitude of incremental impact that various explanatory factors (variables) can have on outcomes, holding all else equal. For this study, regression analysis was used to identify factors that are statistically significant in explaining project impacts in terms of (a) long-term job growth at nearby businesses or (b) the ratio of job impact per $1 million of project cost. Whereas the Chapter 4 tabulations represented job impacts in terms of total job growth attributable to a project, the sta- tistical analyses results reported in this chapter examine fac- tors affecting direct impact, which counts only added jobs occurring at nearby locations benefitting from improved access or enhanced travel conditions. It excludes other aspects of total impact, such as growth of suppliers to the directly affected businesses that may be located elsewhere in a broader surrounding area. This was done to enhance statistical accu- racy for prediction because total impacts can be calculated by multiplying each project’s direct impact by an input-output economic multiplier that is specific to the project area and its economic profile. The direct impact area is usually defined as a neighborhood or corridor, although the corridor may be many miles in length. The total impact area typically is defined as a metro- politan area, county, or aggregation of multiple counties. On average, the direct impact accounts for approximately 70% of the total impact, although that ratio can vary between 50% and 100%, depending on the specific project and its setting. Classification of Project Types For statistical analysis, the 100 case study projects were pooled into three classes, designed to distinguish fundamental differ- ences in project length (size) and traffic volumes: • Roadway corridors: Beltway, bypass, major highway, widen- ing (44 projects). • Point-to-point: Interchange, access road, bridge, connec- tor (37 projects). • Intermodal: Terminals for passenger and freight road/rail transfers (19 projects). The research team developed and tested a series of separate regression equations for U.S. “roadway” and “point-to-point” classes of projects to determine the most important factors affecting the magnitude of their job impacts. Independent (Explanatory) Variables Independent variables are the explanatory factors that are hypothesized to affect observed job impacts; in this case, they describe the nature of the project or its setting. Location- specific data were obtained from the Bureau of Economic Analysis, Bureau of Labor Statistics, Census Bureau, and Esri GIS database. The independent variables that were tested as explanatory factors fell into seven categories, each with a hypothesized behavioral impact, as noted here. • Level of traffic activity. Projects with higher levels of AADT (traffic count) or VMT (total vehicle miles traveled) are most likely to be facing congestion delays, which can have particularly important consequences for access and travel time reliability. • Scale of project. Projects involving the highest number of lane miles are most likely to be connecting urban areas or linking urban activity centers to their surrounding markets. • Urbanization. Projects set in areas of higher average popu- lation density are most likely to be in urbanized areas, where congestion is a particularly important consideration. • Market scale. Projects with the largest size market (measured in terms of population within a 40-minute drive) are most likely to be within large metropolitan regions, where access Statistical Analysis of Job Impacts

27 is a particularly important consideration. They may also be more likely to have rail and air facilities located nearby, which can also gain from highway access improvements. • Terrain. Projects in mountain terrain are most likely to face limited route options and higher sensitivity to slow vehicle or accident delays. • Economic health. Projects in areas that are already economi- cally healthy (measured in terms of higher income and lower unemployment rates) are more likely to enable economic development without facing other barriers (occurring in economically distressed areas) that need to be addressed before additional business investment can occur. • Underlying growth trend. Projects in regions that are already strong and growing (in terms of jobs and income) can be particularly dependent on additional transportation capac- ity enhancement to successfully attract new business. Statistical analysis of Job Impact Regression Results for Explanatory Use The first set of regressions had total job impact as the depen- dent variable. Findings are summarized in Table 5.1 for various alternative combinations of project class (roadway and point- to-point) and setting (expressed in terms of metro, rural, or mixed classification). Only explanatory variables that were found to be statistically significant at the 90% confidence level are shown. The results were tested for “multi collinearity” to ensure that the power of each explanatory variable is estimated in an efficient manner that is not biased by correlations among the explanatory variables. The overall explanatory power of each regression is shown in terms of the R2 value, which repre- sents the portion of variance in the dependent variable (job impact) that is explained by the explanatory variables. The results showed that some explanatory variables had a statistically significant effect for all combinations of project class and setting, while others were statistically significant only for a subset of project-setting combinations. All seven catego- ries of independent variables had some explanatory power for at least some project-setting combinations. The location set- ting variables that most consistently emerged as important were the level of traffic activity, market scale, urbanization, and underlying growth trend. Regression Results for Predictive Use The underlying economic growth trend is an important fac- tor in understanding why the economic impact of highway projects varies from place to place. However, at the time of project planning, one may not be able to assume that local Table 5.1. Regression Results: Factors Affecting Job Impact Project-Setting Combinations Statistically Significant Explanatory Variables (those with more than 90% statistical significance) R2adj Rural setting, point-to-point, and roadway projects •  Level of traffic activity (VMT) •  Market scale (population size) •  Underlying economy: per capita income growth trend •  Economic health (per capita income level) .70 Metro and mixed setting, roadway projects •  Level of traffic activity (AADT) •  Project scale (lane miles) •  Urbanization (population density) •  Market scale (population size) •  Underlying economy: local population and job growth trend .81 Metro setting, roadway projects •  Level of traffic activity (AADT) •  Project scale (lane miles) •  Urbanization (population density) •  Underlying economy: local population and job growth trend .91 Mixed setting, roadway projects •  Level of traffic activity (AADT) •  Project scale (lane miles) •  Urbanization (population density) •  Market scale (population size) •  Terrain (mountain terrain) .91 Urban setting, point-to-point projects •  Economic distress (dummy variable) •  Underlying economy: regional job and income growth trend .58 Rural and mixed setting, point-to-point projects •  Level of traffic activity (VMT) •  Urbanization (population density) •  Underlying economy: regional and local income growth trend •  Economic health (per capita income level) .88

28 or regional economies will continue to trend over time in the same way as they have in the past. For that reason, it is also useful to consider regression equations in which the underlying growth trend is not available as an explanatory variable. Accordingly, Table 5.2 summarizes revised regres- sion results in which only known or planned project char- acteristics and existing preproject socioeconomic factors are used as explanatory variables. Although the resulting explanatory power of the regression equation drops, the results still confirm the importance of differences in project class and setting, including factors such as project scale, level of traffic activity, urbanization, market scale, and eco- nomic health. Those results are also used as a basis for the predictive impact calculator called “My Projects” in the T-PICS web tool. Statistical analysis of Job Impact per Dollar Objective It is not surprising that there is a statistical relationship between project cost and resulting economic impacts. That certainly does not mean that spending more money on a project automatically leads to a larger economic impact. Rather, it indicates that, all else equal, larger scale projects tend to lead to larger scale economic impacts. Furthermore, decisions to fund most major highway projects involve some form of (explicit or implicit) consideration of the benefit relative to cost, so projects that have a high expected cost and low expected benefit are unlikely to ever be built. Although there is a general relationship between project cost and economic impact, it can be useful to identify the nature of that relationship and the extent to which it is affected by other factors associated with either the project type or set- ting. Accordingly, the research team conducted a statistical analysis of ways to relate cost and impact. Analysis Design To assess the statistical relationship of job impacts to project cost, several alternative regression specifications were tested. Explanatory variables included in various regression estima- tions were combinations of project cost (adjusted to con- stant dollars), magnitude of cost scaled by highway size (measured in terms of length), and multiplicative terms combining the cost metrics with a measure of traffic: either AADT (average annual daily traffic) or VMT (vehicle miles traveled). Those variables were examined for the entire set of projects, for the pooled classes of highway and point-to- point projects, and for classes of rural and metropolitan settings. Previous statistical analysis, shown in Table 5.2, showed that we can explain a large share of the variation in job impacts among the case studies by considering project cost and addi- tional factors such as project type, traffic level, and urbaniza- tion of the study area. For example, projects in metropolitan areas are more likely to be implemented to reduce congestion than to have a primary objective of creating jobs. In addition, certain types of projects are initiated specifically to facilitate job development, such as roads that connect highways with office or industrial parks. In this situation, it would be expected Table 5.2. Regression Results: Limited to Factors Known Before Construction Project-Setting Combinations Significant Explanatory Variables (preproject knowledge only) R2adj Rural setting, point-to-point and roadway projects •  Project scale (lane miles) .42 All settings, roadway projects •  Level of traffic activity (AADT) •  Project scale (lane miles) •  Urbanization (population density) •  Market scale (population size) .41 Metro and mixed setting, road- way projects •  Level of traffic activity (AADT) •  Project scale (lane miles) •  Urbanization (population density) •  Market scale (population size) .35 Mixed setting, roadway projects •  Level of traffic activity (AADT) •  Project scale (lane miles) •  Urbanization (population density) •  Market scale (population size) •  Terrain (mountain terrain) .91 Rural and mixed setting, point- to-point projects •  Level of traffic activity (AADT) •  Project scale (lane miles) .61

29 that the cost of a project would be related to the scale of devel- opment to be served. It should also be noted that jobs are only one way of mea- suring the economic impact of highway development. Expan- sive (and expensive) projects generally are conceived to generate significant user benefits, including personal time savings for drivers and passengers and household cost sav- ings, although such user benefits are not part of an economic development impact analysis in this report. Similarly, envi- ronmental, social, and safety impacts may be important con- siderations for some or many of the projects studied here. It is reasonable to assume that major highway investments would not be undertaken without assuming that the benefits are equal or greater than the costs involved. However, this aspect of the analysis focuses only on job creation impacts. Findings Table 5.3 shows the outcome of four alternative regression specifications. It shows that, when considering the full pool of all case study projects, total cost emerges with a stronger relationship to job impact than cost per lane mile. Similarly, total VMT emerges with a stronger relationship to job impact than AADT and highway length. By considering both the cost of a project and its VMT level, we can account for as much as 55% of the variation in jobs generated by all projects. To gain a second perspective, the data set was split into “road- way” projects (which do not have a specific destination point) and “point-to-point” projects (which generally have defined start and end points). The 19 intermodal freight projects and intermodal passenger projects were excluded for this analysis. The analysis again considered combinations of project cost, VMT, AADT, and length. Results are shown in Table 5.4. Results again showed that the strongest statistical relationship was between jobs and total project cost. The regressions explained approximately 83% of the variance in job impacts for point-to- point projects but less than 50% of the variance for continuous roadway projects. There are several explanations for this difference. After all, “point-to-point” projects generally create access to industrial parks, office parks, and other economic development nodes. Moreover, it is likely that state and local area officials are will- ing to invest in high-cost, point-to-point highway develop- ment for strong and foreseeable jobs and benefit returns on investments. In contrast, continuous roadway projects may be created to relieve congestion, in which case there is a less pronounced job impact, or job creation may be generated hundreds of miles from the project investment or may have a robust local job impact. Therefore, the variation of jobs gen- erated by continuous roadway projects does not reflect investment as smoothly as for point-to-point projects. Projects were further divided into metro and rural as a third test to account for the relationship of project cost and jobs. Those results show that the regression explained between 44% and 53% of the job impact variance for urban projects and between 47% and 70% of the variance for rural projects (Table 5.5). Table 5.3. Relationship of Project Cost and Job Impact (Dependent Variable Is Job Impact) Equation Specification Explanatory Variables Coefficient T-Score R2adj A •  Constant term •  Project cost 3.42* 9.14* .46 B •  Constant term •  Project cost •  AADT 1.80 8.83* 2.06* .47 C •  Constant term •  Project cost •  AADT •  Length 1.07 8.26* 2.24* 1.88* .49 D •  Constant term •  Project cost •  VMT 2.24* 8.98* 4.62* .55 * Statistically significant with better than 90% confidence level. Table 5.4. Relationship of Project Cost and Job Impact by Project Class (Dependent Variable Is Job Impact) Project Class Explanatory Variables Coefficient T-Score R2adj Point-to-point •  Constant term •  Project cost 2.15* 11.83* .83 Roadway •  Constant term •  Project cost 2.95* 5.66* .38 Point-to-point •  Constant term •  VMT •  Project cost 1.54 3.28* 5.67* .83 Roadway •  Constant term •  VMT •  Project cost 1.59 0.97 11.69* .48 Point-to-point •  Constant term •  AADT •  Project cost 1.99* –0.29 11.63* .83 Roadway •  Constant term •  AADT •  Project cost 0.43 3.21* 4.93* .48 Point-to-point •  Constant term •  AADT •  Length •  Project cost –1.39 0.21 0.40 10.95* .82 Roadway •  Constant term •  AADT •  Length •  Project cost –0.42 3.49* 1.56 4.56* .49 * Statistically significant with better than 90% confidence level.

30 Calculations in t-pICS Web tool The T-PICS web tool enables users to either (a) search for case studies meeting specified criteria (the “Case Search” feature) or (b) allow the system to calculate a range of potential impacts that is consistent with previous cases, given a specified type of project and setting (the “My Project Tools” feature). This latter feature is enabled by using fac- tors and statistical relationships drawn from the tabulations and regression analyses. The economic impact estimation function of “My Proj- ect Tools” is divided into five modules: (1) initial user entry, Table 5.5. Relationship of Project Cost and Job Impact by Urban Setting (Dependent Variable Is Job Impact) Urban Setting Dependent Variables T-Score R2adj Metro or mixed •  Constant term 3.56* .44 •  Project cost 7.82* Rural •  Constant term 1.41 .50 •  Project cost 4.76* Metro or mixed •  Constant term 2.41* .53 •  VMT 3.85* •  Project cost 7.73* Rural •  Constant term 1.05 .71 •  VMT 4.10* •  Project cost 5.86* Metro or mixed •  Constant term 2.09* .45 •  AADT 1.35 •  Project cost 7.63* Rural •  Constant term 1.0 .47 •  AADT 0.04 •  Project cost 4.64* Metro or mixed •  Constant term 1.37 .46 •  AADT 1.55 •  Length 1.53 •  Project cost 7.11* Rural •  Constant term 0.82 .69 •  AADT -0.26 •  Length 3.94* •  Project cost 5.72* Metro or mixed •  Constant term 2.81* .45 •  Length 1.33 •  Project cost 7.37* Rural •  Constant term 0.87 .71 •  Length 4.02* •  Project cost 5.86* Note: Sample size: 77 metro or mixed projects, 23 rural projects. * Statistically significant with better than 90% confidence level. (2) initial system feedback, (3) preliminary economic impact estimation, (4) user adjustments, and (5) final economic impact estimation. Each module is discussed here. Initial User Entry User inputs are Project Type, Region, Setting (Metro/Rural/ Mixed), Local Economy (Distress) Rating, and Length of the Project (miles). Initial System Feedback Given the user inputs, the T-PICS system estimates typical baseline traffic level (in terms of AADT) and typical project cost, as well as a range for typical economic impacts (in terms of jobs, income, and output, based on the impacts of appli- cable case studies). The traffic level is estimated based on the median for each project type, adjusted by the setting classification. The project cost is calculated using the median cost per mile for each project type, multiplied by the project length (in miles). Users may adjust the traffic and cost values if more accurate num- bers are available, although changing the cost alone will not affect economic impact outcomes. Preliminary Economic Impact Estimation The range of estimated job, income, and business output impacts is presented in terms of direct impact and total impact. The direct impact is calculated based on statistical relationships between the average project impact per mile (or per project) and each of the five classes of user entry variables. The calculation draws on regression results described in this chapter. The total impact is calculated by applying applicable input-output economic multipliers for each study area. User Adjustments T-PICS users may adjust five factors, which will lead to recal- culation of the estimates of impacts on jobs, wages, and output. They are the following: • Project length (miles); • Project baseline traffic level (AADT); • Infrastructure conditions (including water, sewer, telecom, broadband); • Land use and development policies; and • Business climate policies, including availability of financial incentives.

31 Final Economic Impact Estimation Based on analysis of the case study database, the estimated economic impacts are scaled by project size (as reflected by a combination of highway length and traffic level) and adjusted upward or downward based on the policy adjustment factors shown in Table 5.6. In this table, it may be noted that the potential for upward adjustment in economic impacts is larger than the potential for downward adjustment. The rea- son is that the range of actual project impacts is far broader above the median than it is below the median (because few projects have net impacts below zero). In actual use, the eco- nomic impact calculations shown in “My Project Tools” can reflect the compounded effects of any or all of these factors. Finally, the estimated magnitude of estimated project impacts is capped at 1.2 times the largest value observed to date from any case studies of the applicable project type. This serves to prevent anomalous occurrences generating unreasonably large impact estimates. Table 5.6. Impact of T-PICS Adjustment Factors on Estimated Economic Impact Factor Max Negative Impact (%) Max Positive Impact (%) Local economy (distress) rating -11 38 Local economy rating × per mile  scale factor -14 46 Local economy rating × traffic  volume factor -7 31 Urban/Rural settings -58 121 Infrastructure conditions -40 32 Land use policies -34 24 Business climate -12 20 Table 5.7. Percentage of Cases with the Predicted Job Impact Accurate within 1 Standard Deviation of Observed Impact Project Type Percentage of Cases Accurate Within 1 Standard Deviation (%) Total Cases Access road 100 7 Beltway 63 8 Bridge 78 9 Bypass 100 11 Connector 88 8 Interchange 100 12 Major highway 100 13 Widening 100 9 % within range 92 77a a The analysis excluded intermodal terminals, international  projects, and mega-projects. Validation of Predictive Accuracy To test whether the predicted values of direct jobs fall within a reasonable range of accuracy, the project team calculated the mean and standard deviation of actual job impacts asso- ciated with each project type. This enabled statistical confi- dence intervals to be constructed for the observed impacts. It was done for all U.S. highway projects (omitting intermodal terminals, international projects and mega-projects). Pre- dicted values were compared against those confidence inter- vals, and it was found that 92% of the U.S. highway projects had a predicted value within 1 standard deviation of the actual (observed) impact (Table 5.7).