A Manual for Community and Fiscal Impact Modeling Systems

Practical Techniques for Building and Applying Community Level Models

Prepared by

David Swenson and Liesl Eathington
Department of Economics
Iowa State University

Download manual (PDF)

Introduction
Over the past few years several states have developed community and fiscal impact modeling systems. Beginning with the VIP model in Virginia, which was developed by Tom Johnson, and followed by Idaho, Iowa, and Missouri, several mid-western states established or enhanced their capacities to conduct community labor force and fiscal impact analysis. To date, models also have been developed or finalized in Kansas, Minnesota, Nevada, Nebraska, Pennsylvania, Wisconsin, and Kentucky. Models are under construction in Texas, Ohio, and Vermont. A model will soon be developed for Oregon.

The basic conceptual foundation for these models is straightforward. Exogenous shocks to local employment induce changes in local labor supply and changes in residential preferences regionally. The strength of these preferences and the probability of mobility (either as migration or commuting) are a function of the size of employment locally, the size of the labor force locally, the size of employment and the labor force regionally, and general community attributes. When communities experience labor force and population growth, demands for community services increase as do the marginal resources necessary to pay for those services, as illustrated in this simple schematic.

Changes in Labor	Changes in the	Changes in Local
Demand Lead to	Population, which Lead to	Service Demands

This compilation is designed to introduce the reader to community impact modeling systems. It draws upon our experiences at Iowa State University to develop and improve our modeling systems and upon our technical assistance experiences with several states interested in producing home-grown versions of the models. We also acknowledge our widespread dependence on the efforts of others in this area.

Each state that develops this modeling capacity has its own set of issues and particular intended uses – they invariably introduce innovations that are relevant and useful to the rest of us. In western states, for example, issues might center around federal fund and state fund transfers and just a few basic local government service needs. In other states, the fiscal elements may be discounted in favor of identifying expected changes in regional consumption and other local economic effects. Some states have good indigenous data to use for fiscal analysis: others, like Iowa, are highly dependent on data collected during the quinquennial census of local governments. Some have geographic barriers that create problems in the gravity components of the estimation model. Other states, like Iowa (and Nebraska), are the stereotypical "featureless planes" where calculating inter-regional labor relationships is much easier and statistically pleasing as compared to Appalachian or Rocky Mountain states. Finally, some western states have relatively few counties: estimations suffer from an absence of observations. No matter the limitations or the opportunities, however, there are several conceptual and procedural steps that must be taken to build the community impact model. We can break the these steps into at least six distinct parts.

Part 1. The Conceptual Model
Regional scientists are acutely aware of the hunger for good information among state and local decision makers. Analysts are also acutely aware that decisions at the federal level are often made without regard to the potential state or community impacts that might ensue. Changes in fiscal and structural federalism have revised the relationships of the federal government with the states and the states with local governments. As the 1990s end it is evident to any student of government finance and government policy that the overall structure, capacity, and missions of governments are much different than they were in the 1980s, which in turn were much different than much of the previous decade.

Not only are there differences in fiscal federalism, there are also profound differences in the structure and incentives of the nation's labor force. The nation currently is at full or near-full employment. Large fractions of the nation's labor supply are working. Where workers work and where workers live are important issues for community and regional land use, transportation, and other public services planning. Workers remain mobile. Suburbanization continues. The dominant transportation investment in the nation are highway systems. In short, where people live and where people work are dynamic systems that must be accounted before we can adequately anticipate labor force, economic, and fiscal changes in light of industrial growth or decline and changes in employment in an area.

Part 2. Collecting the Data
After finally swallowing the conceptual model, many prospective analysts choke on the apparent difficulty of acquiring the data needed for analysis. Labor force compilations need to be made for the counties using data from the 1990 census (now quite dated but still useful, nonetheless). Incommuters and outcommuters need to be estimated. Counts need to be compiled that estimate the number of workers in a county that both live and work in a county. Not only do we need to know the number of incommuters and outcommuters, we also need to know where they commute from. Most of the data are readily available from the U.S. Census Bureau web site or the BEA's Regional Economic Information Service CD-ROM.

Because these models are decision models for use primarily by local government officials, it is important to use reliable local government financial data. Many states do a good job compiling local government financial information; others, like Iowa, do not. The federal government compiles quinquennial censuses of all government finances. These data sets are massive. We provide information and assistance to states in accessing these data, including programs to read the raw data and write it for entry into other systems, already extracted local government revenue and expenditure data that are compiled as SAS system files, and spreadsheets of common revenue and expenditure items for all governments in a state.

The data acquisition becomes slightly more complicated if the analysts decide to build a municipality-based rather than or along with a county-based model. Government statistics are primarily collected at the county level – subcounty data compilations are rare and their quality is suspect. Still, we can build reasonably good city and county models. In Iowa we have built a county government model (which subsumes school district impacts) and a city government impact model. The models are run simultaneously to apportion impacts to the appropriate government and the appropriate territory; i.e., impacts accruing within a municipality and those outside of the municipality. This all assumes, of course, that spatial considerations are quite important in these models. Persons interested in compiling city-based models are encouraged to contact the authors for additional procedures.

Finally, labor and fiscal variables are not enough. Physical, economic, and social characteristics of the communities or counties that we are modeling are important for isolating variability among our communities. Most of those data are also available from the 1990 census, and some has been compiled in useful form in a variety of ftp sites (like www.ciesin.org ). Other data may by obtained from County Business Patterns or from state departments of revenue who maintain information on, for example, state sales tax collections and firms subject to those taxes.

Part 3. Compiling Variables and Selecting the Level of Analysis
A larger hurdle to clear involves the initial processing of the labor force and of the fiscal data. We have tried to make it a low hurdle instead of a high one. When we first compiled our model in Iowa we were required to read our data from then recently released census tapes. Now the labor force and the incommuter and outcommuter information are easily accessed via the internet or from commonly available CD-ROMs from the BEA. Although getting the data are now easy, there are still steps to processing the data that require a little effort.

The labor force component is gravity based: that means that the size of local employment and of the local labor force must be considered in light of external employment and the external labor force. It is necessary, then, to compile counts of external employment and the external labor force, and it is necessary to localize that employment spatially. To do so requires finding appropriate latitudes and longitudes relative to your counties of study and processing the data to arrive at appropriate distance calculations. For some of the more picky among us, care might also be taken at this point to arrive at a distance-decay function that best approximates a region's labor to employment characteristics. For the rest of us, however, distance squared seems to do the trick.

There is an additional important element of this section. It deals with the spatial and institutional relationships that are evident in your state of study. Simply put, local government service territories often overlap but do not align – school districts do not align with municipalities or with counties; drainage districts or some other form of special governmental district may cross county or even state lines. There may also be quite distinct differences in allowable local government functions by kind of government, by classes of communities and counties. We must also add to this the probability of acquiring relevant data on all of our local government service territories. Some of these pitfalls are discussed in this section.

Part 4. Estimating and Evaluating the Equations
Once all of the data have been organized a system of equations needs to be set up that is consistent with the conceptual model (Part 1). The proper specification of the variables and the proper form of estimation are important considerations. Very strong justification exists for using 3-stage least squares when estimating the labor force and the fiscal impact variables (as described in the Johnson, et al, article). Accordingly, care must be taken to specify identities, instruments, and the endogenous variables properly. It is also important to realize that as the number of equations in the system expands, the more unstable the model may become.

Many analysts might find it useful to compile their original model prototype using 3-stage least squares for the labor force equations and OLS for the fiscal or economic indicators. Over time they can revise and re-specify their model consistent with theory. An OLS-based model is far superior, of course, to no model and, in our experience, only slightly less descriptive than a fully estimated 3-stage least squares model. This is especially true because the advantages to 3-stage least squares seem to accrue mostly to the tails of the distribution—accuracy is probably not significantly improved for the mean county or community.

This section discusses the procedures of model specification. In addition, significant attention is paid to common problems in construction, specification, and evaluation of the model and its equations.

Part 5. Creating the Impact Assessment Spreadsheet or Database
The pre-processed data and the coefficients generated need to be put into some system so that simulations can be made. Most models to date have used a Lotus 123 or an Microsoft Excel spreadsheet format. The models can also be compiled using traditional statistical programs such as SAS or SPSS. We recently have also created an interactive model using Microsoft Access (this was to facilitate translation into HTML for the posting of a web-based model).

The whole point of this section is to introduce the model builder to different approaches to developing a spreadsheet (or some other means of conveyance) for calculating the outcomes and displaying the results. Our original model was very transparent and consisted of three linked spreadsheets: one to report the results in a relatively pleasing format, one to compile the county level impacts, and another to compile the community level impacts. Our newest model is significantly revised, more interactive, way more pleasing to look at, and divided into functional segments. The output does not change regardless of the form, although the ease of use does.

Part 6. Using the Model
While most of us have a general idea of what we want to do with our community impact model, the simple fact is that each state applies their models somewhat differently and towards different ends. Here in Iowa, the model is almost exclusively used as an impact model that is linked to the results of an input-output analysis of local employment change. In addition, the typical client is the community not the county, since most economic development gains and losses accrue and are measured at the municipal level. The Wisconsin model appears to more readily describe expected changes in the commercial environment instead of the fiscal environment. The Missouri Show Me model is a county based system designed to be used to test different growth and change scenarios for participating counties. Western states models appear to be more sensitive to issues associated with resource extraction and with physical infrastructure costs. In this section of the report we compile a set of impact analysis descriptions and approaches.

Download manual (PDF)

| Department of Economics | Working Papers |

For more information, contact admin@econ.iastate.edu.