Employment Location Choice Model

In this model, we predict the probability that a job that is either new (from the Economic Transition Model), or has moved within the region (from the Employment Relocation Model), will be located at a particular site. The grid cells used as the basic geographic unit of analysis in the current model implementation contain variable quantities of space to be occupied by jobs. The number of locations available for a job to locate within a grid cell will depend mainly on the total square footage of nonresidential floorspace in the cell, and on the density of the use of space (square feet per employee). Given the possibility that some jobs will be located in residential units, however, housing as well as nonresidential floorspace must be considered in job location. We have defined a maximum rate of home-based employment, determined using local data for a particular metropolitan region, to identify the potential set of spaces available for home-based employment. The set of job locations available for placing a job, then, are the union of the spaces in nonresidential floorspace and a subset of the residential units in the cell. The model is specified as a multinomial logit model, with separate equations estimated for each employment sector.

$S_{blt}$	is the number of job space for building type at location ,
$s_{bl}$	is a scalar representing the total nonresidential square footage of floorspace for building type in location ,
$r_{bl}$	is a space utilization rate for building type in location (e.g. non-residential sqft per non-home-based employee, or housing units per home-based employee).

Since we process each building type in parallel, we can eliminate subscript

in the following equations.

For both the employment location and household location models, we take the stock of available space as fixed in the short run of the intra-year period of the simulation, and assume that locators are price takers. That is, a single locating job or household does not have enough market power to influence the transaction price, and must accept the current market price as given.

The variables included in the employment location choice model are drawn from the literature in urban economics. We expect that accessibility to population, particularly high-income population, increases bids for retail and service businesses. We also expect that two forms of agglomeration economies influence location choices: localization economies and inter-industry linkages.

Localization economies represent positive externalities associated with locations that have other firms in the same industry nearby. The basis for the attraction may be some combination of a shared skilled labor pool, comparison shopping in the case of retail, co-location at a site with highly desirable characteristics, or other factors that cause the costs of production to decline as greater concentration of businesses in the industry occurs. The classic example of localization economies is Silicon Valley. Inter-industry linkages refer to agglomeration economies associated with location at a site that has greater access to businesses in strategically related, but different, industries. Examples include manufacturers locating near concentrations of suppliers in different industries, or distribution companies locating where they can readily service retail outlets.

One complication in measuring localization economies and inter-industry linkages is determining the relevant distance for agglomeration economies to influence location choices. At one level, agglomeration economies are likely to affect business location choices between states, or between metropolitan areas within a state. Within a single metropolitan area, we are concerned more with agglomeration economies at a scale relevant to the formation of employment centers. The influence of proximity to related employment may be measured using two scales: a regional scale effect using zone-to-zone accessibilities from the travel model, or highly localized accessibilities using queries of the area immediately around the given grid cell. Most of the spatial queries used in the model are of the latter type, because the regional accessibility variables tend to be very highly correlated, and because agglomerations are expected to be very localized. Note that the use of radial queries surrounding grid cells also avoids the problems of arbitrary zonal aggregations.

Age of buildings is included in the model to estimate the influence of age depreciation of commercial buildings, with the expectation that businesses prefer newer buildings and discount their bids for older ones. This reflects the deterioration of older buildings, changing architecture, and preferences, as is the case in residential housing. There is the possibility that significant renovation will make the actual year built less relevant, and we would expect that this would dampen the coefficient for age depreciation. We do not at this point attempt to model maintenance and renovation investments and the quality of buildings.

Density, the inverse of lot size, is included in the location choice model. We expect businesses, like households, to reveal different preferences for land based on their production functions and the role of amenities such as green space and parking area. As manufacturing production continues to shift to more horizontal, land-intensive technology, we expect the discounting for density to be relatively high. Retail, with its concentration in shopping strips and malls, still requires substantial surface land for parking, and is likely to discount bids less for density. We expect service firms to discount for density the least, since in the traditional urban economics models of bid-rent, service firms generally outbid other firms for sites with higher accessibility, land cost, and density.

We might expect that certain sectors, particularly retail, show some preference for locations near a major highway, and are willing to bid higher for those locations. Distance to a highway is measured in meters, using grid spatial queries. We also test for the residual influence of the classic monocentric model, measured by travel time to the CBD, after controlling for population access and agglomeration economies. We expect that, for most regions, the CBD accessibility influence will be insignificant or the reverse of that in the traditional monocentric model, after accounting for these other effects.

Calibration of the model is based on a geocoded establishment file (matched to the parcel file to link employment by type to land use by type). A sample of geocoded jobs in each sector is used to estimate the coefficients of the location choice model. As with the Household Location Choice Model, the application of the model produces demand by each employment type for cell locations.

The employment location model processes each job in the mover queue individually, and queries grid cells for alternative locations to consider. These alternatives are sampled in proportion to the capacity of the built space in the cell for accommodating jobs, and the number of alternatives to consider may be determined by the user. Note that jobs may be located in housing units, as is increasingly the case with home-based employment through telecommuting and small independent home-based businesses. A logit model is applied to estimate the probability that the current job will move to each of the alternative job spaces under consideration. Monte carlo simulation is used to generate a decision to locate in a particular alternative, and once this choice is made, the job is assigned to the cell, and the respective quantities of vacant and used space in the cell are updated. If a preferred alternative for a job becomes unavailable during a simulation run, having been chosen and occupied by a previously locating job, the currently locating job is assigned its next best available alternative.

The independent variables used in the employment location choice model can be grouped into the categories of real estate characteristics, regional accessibility, and urban-design scale effects as shown below:

The job location pairs set is defined to contain all pairs of jobs and locations that correspond to jobs occupying a particular location.

$\displaystyle M^l_{jt}$	$\displaystyle = \{\, (j, l) \mid j \in J_{t}, l \in L^J_{t},$ job is placed at location $\displaystyle \,\},$	(12)
$\displaystyle J^U_{t}$	$\displaystyle = \{\, j \mid j \in J_{t}, \forall l \in L_{t} \; (j, l) \notin M^l_{jt} \,\},$	(13)
$\displaystyle ?\vert J\vert _{lt}\vert$	$\displaystyle = \{\, \vert j\vert \mid \forall l \in L_{t} \; (j, l) \in M^l_{jt}, \}\,$	(14)
$\displaystyle L^U_{t}$	$\displaystyle = \{\, l \mid l \in L_{t}, s_l - c_l > 0, s \in S_{t}, c \in L^j_{t} \,\},$	(15)
$\displaystyle P^l_{jt}$	$\displaystyle = \{\, P(j, l) \mid j \in J^U_{t}, l \in L^U_{t},$ is the MNL probability function of job locating in $\displaystyle \,\}$	(16)

Monte Carlo sampling of the location choices for each job occurs over the distribution given by $P^l_{jt}$ .

$\displaystyle \hat{M}^l_{jt} = \{\, (j, l) \mid j \in J^U_{t}, l \in L^U_{t},$

monte carlo choice of

from $P^l_{jt}$ $\displaystyle \,\},$

(17)

$M^l_{jt}$	is the set of new job/location pairs,
$\vert J^L_t\vert$	is the number of jobs for location set L,
	is the set of locations with available space to place jobs,
	is the set of jobs that don't match to any locations,
$P^l_{jt}$	is the probability for job location in ,
$\hat{M}^l_{jt}$	is the set of new job/location pairs created using a monte carlo sampling from $P^l_{jt}$ .

For all the jobs in $J^U_{t}$ to be placed, The cardinality of available space $S_{t}$ has to be larger than or equal to that of $J^U_{t}$ . If not, the cardinality of the new job/location pairs is equal to the cardinality of unplaced jobs or available space, whichever is smaller.