Rensselaer Polytechnic Institute

Civil and Environmental Engineering Department

 

CIVL6230–Transportation Economics–

 

Jose Holguin-Veras, Ph.D., P.E.                   Office room: JEC 4030, Telephone: 276-6221

Email: jhv@rpi.edu                                        Office hours: W 4:00-6:00pm/ by appointment

Lecture hours: Monday & Thursdays 4:00-6:00 PM

Prerequisite: CIVL 2030+DSES4140 or their equivalents 

 

Laboratory #2: Business Location Modeling

Objective:

1.      To introduce the students to advanced econometric techniques involving the estimation of discrete choice models.

2.      To introduce the students to estimation of discrete choice models using sampling of alternatives techniques.

 

The case study

The main objective of the modeling process is to gain insights into the determinants of business location. Among other things, this knowledge provides insights into the most effective mechanisms to attract business and therefore foster economic development.

 

The analyses are based upon employment data obtained from the New Jersey Department of Commerce and Economic Growth and the New Jersey Office of Business Research. This database contained the information that companies moving into New Jersey provide when registering with the State. The data represented 1,017 firms that relocated into New Jersey from outside the state or the country from 1990 through 1999. Intrastate moves or firms leaving the state were not included. For that reason, the analyses conducted here are representative only of the revealed behavior of firms relocating to New Jersey from elsewhere. This limitation is to be kept in mind when reading this report. Additionally, neither the reason that a firm decided to relocate nor whether the move represents an opening of a branch or a complete move are not included.

 

The 1,017 firms employed an estimated total of 108,000 employees and represented 74 industries as defined by different NAICS (North American Industrial Classification System) codes. The data set included the name of the firm, the year of registration, the destination county and address in New Jersey, the state or country of origin, the number of employees and the SIC number (to two digits), and description of the “line of business.” A substantial number of firms were missing addresses, number of employees, and/or SIC codes. Missing addresses were obtained from the Internet or telephone books. The project team converted SIC two digit codes to NAICS three digits codes, and the missing ones were determined from the “line of business” description. For firms missing information on number of employees, the average number of employees for firms with the same NAICS code was used.

 

The firms in the database were geolocated using Geographic Information Systems (GIS) for spatial analyses. The resulting GIS was used to generate a set of zoning systems to facilitate the modeling process. A key objective of the zoning systems was to discretize the choices so that­—instead of having a continuous of choices in space­—the choices are part of a discrete set. The first trials used zoning systems based on ZIP codes and counties. Since no acceptable models resulted from these attempts, the project team created zoning systems based on internally “homogenous” zones, in terms of key socio-economic characteristics such as income. The zoning system used in the lab is one of these zoning systems.

 

The original database was complemented with estimates of transportation accessibility, and socio-economic attributes of the different zones (i.e., population, population density, area, median income) extracted from the population census. Two different sets of accessibility measures were considered. The fist one captures the travel impedances from each zone to the major population centers of New York City and Philadelphia, using the travel times as a proxy. The second set of accessibility measures captured the travel impedance to destinations in New Jersey by means of the accessibility index suggested by Allen et al. (1993). Two accessibility indexes were generated for (a) highway and (b) transit modes. All of the accessibility variables were estimated using the travel time data from the NJDOT travel demand model. The resulting database was then used for model estimation, after converting it to the appropriate format.

 

Figure 1: Zoning system used in the analyses

 

 

Data set used for estimation

The data set used for estimation included variables that characterize the firm, as well as variables that measure transportation accessibility. The original set of variables is shown in Table 1.

Table 1: Variables used in discrete choice modeling

 

The type of economic activity was represented using the NAICS (North American Industrial Classification System) codes (for more information see http://www.bls.gov/bls/naics.htm). The original values of the NAICS were further aggregated into six supergroups, as outlined in Table 2. An even larger grouping system that classified the economic activities in either: goods producing (NAICS 11 to 33) and service providing (42 and above) was created.

 

Table 2: Original NAICS and supergroups (number of observations in parentheses)

Supergroup #1:

21 Mining (0)

22 Utilities (3)

23 Construction (9)

Supergroup #2:

31 Food manufacturing (70)

32 Wood product manufacturing (68)

33 Primary metal manufacturing (123)

Supergroup #3:

42 Wholesale Trade (64)

44 Retail trade (55)

45 Sporting goods, hobby, book, and music stores (47)

Supergroup #4:

48 Transportation (30)

49 Postal service and warehousing (109)

Supergroup #5:

51 Information (47)

52 Finance and insurance (100)

53 Real estate and rental and leasing (5)

54 Professional, scientific, and technical services (53)

Supergroup #6:

55 Management of companies and enterprises (3)

56 Administrative and support and Waste management / remediation services (23)

62 Health care and Social assistance (12)

71 Arts, entertainment, and recreation (5)

72 Accommodation and Food services (13)

81 Other services (except Public administration) (9)

92 Government (5)

 

Objectives of the analyses:

a)      To determine the key variables explaining the business location process

b)      To define the set of transportation policies that would translate into an increase in business relocations to the most impoverished areas of Northern New Jersey

c)      To quantify the impacts of these policies to determine their effectiveness

 

1^st part of the lab: Estimation of discrete choice models using the complete choice set

Data

The data could be downloaded from the course web site. The zip file contains:

a)      the input data in TXT format (zs2LIMDEP.TXT)

b)      a LIMDEP command to read the data (zs2Input.lim)

c)      a LIMDEP command to do a basic set of Transformations (Transformations.lim)

d)      a file with the zonal characteristics (ZS2zonaldata.xls)

 

To read the data in LIMDEP:

a)      unzip all the files and save it folder XXXX

b)      change Project/Settings/Number of Cells to 19000000

c)      open the file zs2Input.lim using LIMDEP and change the folder address to XXXX, save the file when done

d)      Run zs2Input.lim from LIMDEP using Run/Run File

e)      After it finishes reading the data, usie Run/Run File to run Transformations.lim

f)        Save the project so that all transformations and data are saved

 

To run the example models:

a)      run Example.lim from LIMDEP using Run/Run File

b)      once you run it, you could try different models

 

2^nd part of lab

a)      use sampling of alternatives techniques to estimate the best models obtained in the first part of the lab

b)      compare results