Lake Amenities,Environmental Degradation,and Great Lakes Regional Growth

Abstract

Regional migration and growth are increasingly associated with high-quality in situ natural amenities. However, most of the previous US research has focused on the natural amenities of the Mountain West or the South. The Great Lakes, with their abundant fresh water and natural amenities, would also appear well positioned to provide the foundation for this type of economic growth. Yet, while some parts of the western Great Lakes region are prime examples of amenity-led growth, other areas in the eastern Great Lakes may not have capitalized on their natural amenities, perhaps because of their strong industrial legacy. Using a unique county-level data set for the Great Lakes region (including Indiana, Illinois, Michigan, Minnesota, New York, Ohio, Pennsylvania, and Wisconsin), we test whether growth in the region is associated with proximity to lake amenities and whether there are offsetting industrial legacy or pollution effects. We also examine whether amenities have additional attraction value for those with high levels of human capital. Consistent with theory that suggests that natural amenities are normal or superior goods, we find that coastal areas in the region are positively associated with increases in shares of college graduates. However, we find little evidence that lake amenities contribute to broader household migration, especially after 2000. Based on these results, there may be opportunities to leverage Great Lake amenities to support economic growth in terms of attracting individuals with high levels of human capital who are most likely to make quality of life migration decisions.

Keywords

urban and regional economic development economic growth and development policy and applications other urban and regional issues urban and regional issues regional econometric models economic analysis methods population and employment distribution human spatial structure spatial structure natural resource management natural resources

Motivation and Literature Review

Regional migration and growth are increasingly associated with high-quality in situ natural amenities. The US Great Lakes, with their abundant fresh water and natural amenities, would seem to provide the foundation for amenity-linked economic growth. However, industrialization and environmental degradation, especially in the former rust belt cities of the eastern Great Lakes region, suggest the link between amenities and economic growth in this region may be less clear. At the same time, with persistent economic sluggishness and the ongoing economic restructuring that has eroded the region’s economy, finding new ways to capitalize on its assets and generate growth is paramount. Policy makers are interested in whether the Great Lakes are quality of life (QOL) enhancing assets that can attract households and support regional economic growth.

Underlying the research on amenities and economic growth is the Tiebout (1956) theoretical notion that people “vote with their feet” and sort to reside in places with particular bundles of economic and site-specific public goods and amenities. Indeed, the Tiebout approach is closely associated with the spatial equilibrium model first associated with Rosen (1979) and Roback (1982). In this, the utility that individuals get from a particular location depends on both the QOL-enhancing effects and the productivity-generating wages they receive (Tiebout 1956; Rosen 1979; Roback 1982; Rappaport and Sachs 2003). Households then move toward locations with higher levels of utility. Firm locations and movements, which affect household wages, depend on the relative benefits firms receive in terms of productivity, wages, and capital costs. Related to the spatial equilibrium approach is the environmental Kuznets curve in which environmental quality first declines and then rises with rising levels of economic development (Rupasingha et al. 2004), consistent with amenities being a normal or superior good.

A wealth of previous empirical work has demonstrated the link between population growth and natural amenities (e.g., Graves, 1976, 1980; Rappaport 2007, 2004; Rappaport and Sachs 2003; McGranahan 1999; McGranahan 2008; Deller et al. 2001; Partridge et al. 2008) and urban amenities (Glaeser, 1999; Glaeser and Maré 2001; Florida, 2002a, 2002b). In a review article, Partridge (2010) found that, since the middle of the twentieth century, growth patterns have been consistent with migration to areas with high levels of amenities. However, most recently, Partridge et al. (2012) found evidence that amenities may have become less important to migration and population change after 2000. Additionally, Banzhaf and Walsh (2008) found that environmental degradation associated with toxic releases is associated with decreases in population and income, supporting the idea that people sort away from environmental disamenities.

Since employment growth is determined by both household and firm migration, even if people are moving toward places with higher levels of QOL, this may not correspond with employment growth. Gabriel and Rosenthal (2004) and Chen and Rosenthal (2008) provided evidence that regional factors that are QOL enhancing and attract retirees may not be productivity enhancing. Conversely, areas with more pollution may, all else equal, be less attractive to households but be productivity enhancing for manufacturing and similar industries because of looser environmental regulations (Jeppesen, List, and Folmer 2002). While Monchuk and Miranowski (2007) found evidence that natural amenities are associated with employment growth in the Midwest, Partridge et al. (2008) found tremendous spatial variation in the effect of natural amenities on growth.

A concern for policy makers in the region is that high-amenity places are often converted into recreation or retirement destinations which may have a lot of low-paying hospitality-oriented jobs. However, an offsetting factor is that because amenities are normal, superior, or even luxury (Kerr 2011) goods, high-amenities areas may be most attractive to those with higher incomes and workers with higher skills (Roback 1988). The implication is that high-amenity areas will have a larger share of workers with high levels of human capital, which can be proxied by those with college degrees or higher levels of education. In this case, amenities can lead to new firms entering the region who demand high-skilled labor and thus bid up the wages (Deller et al. 2001; Partridge and Rickman 2003; Kim, Marcouiller, and Deller 2005), producing higher economic growth among firms demanding “knowledge workers.” Increasing numbers of high-skilled workers and the firms that hire them can lead to knowledge spillovers and other productivity benefits that can increase wages (Rosenthal and Strange 2001, 2004). Thus, these effects can help transform high-amenity regions into economically diverse regions that include higher-paying, higher-skilled jobs.

Access to amenities could also contribute to population and job growth over a wide geographic area. For example, Schmidt and Courant (2006) show that certain amenities, such as national monuments, can influence economic outcomes for hundreds of kilometers. Other papers have examined the geographical reach of site-specific attributes including access to cities, public goods, and natural amenities (Ferguson et al. 2007; Partridge et al. 2008; Irwin 2002). Thus, the influence of the Great Lakes may not just be on the immediately adjacent counties but also on counties that are “near” the lakes as people move to be close enough to the lakes to enjoy them in their leisure time.

In considering the potential for Tiebout-style sorting in response to Great Lakes amenities, we hypothesize that people who favor Great Lakes natural amenities will self-sort to live close to the water (all else constant). However, if there is also nearby environmental degradation in the form of existing industry, abandoned industrial sites, or declining water quality, individuals will choose to locate elsewhere. The question is do households value the lake amenities and can the region chart a new economic future that is based on the lakes as in situ natural resources rather than one based on industrialization and extraction?

To explore this question, this article examines whether proximity to lake amenities is associated with population and employment growth. We also test whether lake amenities have additional attraction value for individuals with high levels of human capital, building on the work by Roback (1988). Our analysis uses a unique data set that includes geographically defined variables related to natural amenities and environmental disamenities. While there is a plethora of research in the environmental literature that examines the effect of individual environmental disamenities on regional growth or local housing values, this is one of the first attempts to use detailed pollution and other environmental data to assess how environmental degradation can offset the benefits to regional growth from natural amenities.

We also separately control for whether a county is located on the coast of one of the Great Lakes and the distance to the nearest Great Lake; these measures help sort out whether positive lake amenity effects spillover into the region. Hence, our basic identification strategy will be whether proximity to the Great Lakes matters after conditioning on key factors that influence growth including initial industry composition, (other) natural amenities, and in some specifications, state-fixed effects. The empirical model asks whether proximity to lake amenities influences growth in a given county after conditioning on other factors that also influence economic growth.

We also include a measure for whether or not a county is in the eastern versus the western part of the region in order to control for historic economics factors that might make economic prospects in the east (those counties closer to Lakes Erie, Huron, or Ontario) different than that in the west (those counties closer to Lakes Michigan or Superior). As shown in Table 1, since 1990, the Western Great Lakes appear to be doing better economically, with anecdotal evidence that they are benefiting from amenity-driven growth. However, there has been little empirical examination of this potential “east–west” divide. If the difference in growth rates is due to the fact that eastern counties are more industrialized and have more pollution, then once we control for such disamenities any differences should disappear.

Table 1.

The Great Lakes Region, Descriptive Statistics.

	Eastern Great Lakes counties (n = 244)		Western Great Lakes counties (n = 384)		Difference between east and west?
	Mean	SD	Mean	SD	T-Statistic
Growth variables: 1990–2000
% Change in county-level population	5.56	8.43	8.56	10.93	3.65***
% Change in county-level employment	16.65	15.77	21.83	17.43	3.77***
% Change in county-level share of college graduates	23.19	11.49	26.95	13.19	3.65***
Growth variables: 2000–2007
% Change in county-level population	0.67	5.91	1.59	8.40	1.48
% Change in county-level employment	2.46	10.84	1.97	10.79	−0.56
% Change in county-level share of college graduates^a	16.01	12.34	17.30	24.87	0.75
	Eastern Great Lakes counties (n = 244)		Western Great Lakes counties (n = 384)		Total
	Mean	SD	Mean	SD	Mean	SD
Levels of growth variables
Population, 1990	130,038	232,730	73,594	281,350	95,524	264,769
Population, 2000	134,368	233,992	80,807	298,300	101,618	276,147
Employment, 1990	68,655	138,619	41,420	175,403	52,002	162,533
Employment, 2000	76,987	151,941	49,241	191,841	60,021	177,796
% College Graduates, 1990	13.05	5.77	13.00	5.40	16.22	6.72
% College Graduates, 2000	15.96	6.86	16.39	6.62	13.02	5.54
Explanatory variables (1990)
Distance to nearest Great Lake	136.94	93.68	199.56	134.05	175.23	123.74
Percentage forest area	42.60	25.15	26.82	25.84	32.95	26.69
Topography score	0.14	1.02	−0.53	0.61	−0.27	0.86
January temperature—Z-score	−0.57	0.32	−1.00	0.69	−0.83	0.62
July temperature—Z-score	0.47	0.39	−0.17	0.59	0.08	0.61
Percentage water (not including Great Lakes)	−0.21	0.46	−0.01	0.69	−0.08	−0.62
Average annual snowfall	30.36	17.07	25.83	17.25	27.59	17.30
Distance to the nearest Superfund site	30.59	22.07	43.05	32.74	38.83	29.53
Number of power plants	1.52	3.07	0.85	1.49	1.03	2.20
Total water emissions (tons)	38,805	223,023	27,469	273,389	32,322	258,697
Total air emissions (tons)	1,277,166	2,170,466	832,704	2,443,207	997,886	2,366,926
Disamenity index	0.02	2.14	−0.04	2.18	−0.02	−2.17
Distance to nearest metro	43.36	32.74	68.34	54.34	58.64	48.66
Incremental distance to metro > 250,000	20.13	30.72	40.88	56.21	32.82	48.97
Incremental distance to metro > 500,000	14.84	26.86	40.63	57.77	30.61	49.76
Incremental distance to metro > 2.5 million	67.18	85.18	17.22	35.72	36.63	64.69
% College graduates	13.05	5.77	13.00	5.40	13.02	5.54
% Some college	19.93	4.66	22.60	3.65	21.56	4.27
% High school graduates	40.76	5.91	38.89	4.84	39.62	5.35
% White	95.86	5.13	96.23	5.29	96.08	5.22
% Married	61.24	4.32	62.59	4.52	62.07	4.49
% Population female	51.15	1.13	50.83	1.25	50.95	1.22
Population density (population per square mile)	228.29	408.91	139.72	427.51	174.13	422.27
% Population under 18	25.20	2.27	25.66	2.26	25.48	2.27
% Population over 65	13.61	2.65	14.88	3.36	14.39	3.16
% Population over 16 that is employed	55.62	6.24	57.77	6.75	56.94	6.64
% Manufacturing employment 1970	27.08	10.47	19.06	11.72	22.18	11.90
% Recreational homes	7.58	12.67	8.42	12.75	8.09	12.71
% nonfarm proprietor employment	16.39	4.19	17.36	4.56	16.98	4.44
% wage and salary workers, leisure	8.33	3.34	8.88	4.48	8.66	4.08
% wage and salary workers, manufacturing	24.02	10.69	20.45	11.37	21.84	11.24
% wage and salary workers, government	19.12	7.39	20.79	8.09	20.14	7.86
% wage and salary workers, agriculture	1.61	1.55	3.20	2.75	2.58	2.48

Note: ^a The share of college graduates in 2007 is actually the American Community Survey five-year estimate from 2005 to 2009.

Despite the growing body of literature demonstrating the positive growth effects of natural amenities, this is one of the only recent studies to focus specifically on the Great Lakes region. While some of the earliest investigations into the relationship between natural amenities and growth were for the areas around Lakes Michigan and Superior (Wehrwein and Johnson 1943), much of the recent research has instead focused on the natural amenities of the Mountain west or the South. Given the spatial heterogeneity of amenity effects (Partridge et al. 2008), national studies are not always instructive in explaining how amenities would affect growth in this region.

The results reveal that, consistent with a story that high-skilled workers should be more attracted to places with higher levels of amenities, coastal areas in the Great Lakes region are positively associated with increases in shares of college graduates, relative to other parts of the region. However, after controlling for industrial disamenities, there is no statistical difference in this effect for places in the eastern part of the region versus the west.

At the same time, there is only weak evidence that lake amenities are associated with overall population or employment growth in the Great Lakes region. And after 2000, population and employment growth declined region wide. Since proximity to the lakes was associated with rising rents in the 1990s, this suggests increasing household value and/or firm productivity from lake amenities. The overall decline in firm and household migration after 2000 could also be due to the value of amenities being capitalized into rents and wages; providing little reason for additional movement after 2000. Another possible explanation is that the economic downturn of the 2000s may have reduced demand for access to lake amenities because they are normal, superior, or even luxury goods. Finally, there is evidence that the environmental quality in the lakes worsened after the late 1990s and this could explain a devaluing of the lake amenity post 2000.

Based on these results, it appears that Great lakes communities may be able to leverage their proximity to lake amenities to support economic growth, especially in terms of attracting individuals within the region with high levels of human capital who are most likely to make migration decisions based on QOL measures. We caution that additional analysis is needed to identify which policies would be most effective for specific portions of the region and to assess their costs. Further, if the recent deterioration of lake water quality in the region becomes widely known, it could act as a disamenity and repel those households most likely to consider migrating to the region. Such patterns would support the need to maintain or enhance environmental quality in order to maximize economic growth.

The remainder of this article is organized into the following sections: the second section outlines the theoretical framework used for estimation; the third section presents the empirical specifications; the fourth section discusses the unique data set used in this analysis; the fifth section discusses the key results; and, finally, the sixth section summarizes and concludes.

Theoretical Framework

Our empirical analysis is based on a two-equation spatial equilibrium framework (from Roback 1982; Partridge et al. 2010; Jeanty, Partridge, and Irwin 2010) in which firms maximize profits and households maximize utility. In this model, the representative household chooses amounts of a composite traded good (Y), land ( $L^{h}$ ), and site-specific characteristics (s) to maximize utility subject to a budget constraint:

max U_{i} (Y_{i}, L_{i}^{h} | s_{i}) s . t . w_{i} = Y_{i} + r_{i} L_{i}^{h},

where wage and rental payments are w and r and the price of the composite good is normalized to 1. The s vector includes all characteristics that make regions heterogeneous, including natural and urban amenities and environmental disamenities.

In spatial equilibrium, because households can sort to the location with the highest utility, wages and rents will adjust so that the indirect utility is the same in all regions, and is equal to $\overline{V}$ , otherwise, some households would have an incentive to move.

\overline{V} = V_{i} (w_{i}, r_{i} | s_{i}) .

The representative firm produces Y using a constant returns-to-scale production function $Y_{i} = f (L_{i}^{f}, N_{i} | s_{i})$ where N is the number of workers. Here the area characteristics act as profit shifters. For example, a higher share of college graduates may raise the productivity of the firm.

Again, under the assumption of perfect mobility, in spatial equilibrium, wages and rents will adjust so that unit costs are equalized across regions and are equal to 1. We utilize the unit cost function because of the constant returns to scale assumption.

C_{i} (w_{i}, r_{i} | s_{i}) = 1.

Given a partial adjustment process, in the long run, spatial equilibrium will be reached when utility and cost differentials are eliminated across all regions. In equilibrium, because the value of amenities is capitalized into wages and rents, Roback (1982) and others have shown that regional differences in wages and rents can be used to value location-specific attributes.

However, as shown by Rappaport (2004), spatial equilibrium is the long-run steady state of a growth model and migration toward equilibrium (i.e., being out of the steady state) can be persistent. Even small frictions to labor and capital mobility, productivity, or QOL can draw this adjustment process out for decades. The literature includes a number of papers that have found ongoing migration toward equilibrium in the United States (Graves and Mueser 1993; Rappaport 2007; Greenwood et al. 1991; Partridge et al. 2012).

Thus, we assume there will be ongoing movements toward equilibrium. And, since regional utility and cost differentials are the main drivers of firm and household relocations across regions, the movement of households μ _i and the movement of firms η _i are functions of these differentials:

μ_{i} = f {[\overline{V} - V_{i} (w_{i}, r_{i} | s_{i})] θ_{i}}

η_{i} = f {[C_{i} (w_{i}, r_{i} | s_{i}) - 1] τ_{i}},

where $θ_{i}$ reflects frictions to household movement, such as moving costs and imperfect information, and $τ_{i}$ is an adjustment factor related to firm movement, such as moving costs and barriers to entry.

The model can also be expanded to include two types of workers in the spirit of Roback (1988), high-human capital workers and low-human capital workers. In that case, there would be two household migration equations with each type responding to different long-run spatial equilibrium utility levels.

Empirical Specifications

Following Roback (1982) and a host of other empirical papers that have subsequently used the Roback model (e.g., Partridge et al. 2010; Jeanty, Partridge, and Irwin 2010), we thus consider how population and employment changes are affected by natural amenities, especially access to the Great Lakes, and environmental disamenities (equations 6 and 7, below). Population changes are affected by migration of households, related to equation (4). While we are interested in assessing movement of households or migration due to utility differentials, Rappaport (2007) and Faggian, Olfert, and Partridge (2012) show that population change is a good proxy for household migration and reveals the representative household’s assessment of where his or her well-being is improved. Employment changes are determined in the labor market by the movement of both households and firms, equations (4) and (5) above. Household movements affect labor supply and firm movements affect labor demand; and they jointly determine a region’s employment level.

To better understand who is moving, and assuming amenities are normal or superior (or luxury) goods, we also investigate whether higher-educated individuals or those with higher levels of human capital are choosing to locate near amenities in the region. Specifically, we consider the change in the share of college graduates in the region as demonstrated in equation (8). This allows us to test Roback’s (1988) extension of the original Roback (1982) model.

For each county i in state s, the reduced form estimation equations are given by:

{Δ P O P}_{i s} = α^{P} + β_{1}^{P} {A M E N I T Y}_{i s} + β_{2}^{P} X_{i s}^{P} + σ_{s}^{P} + ϵ_{i s}^{P},

{Δ E M P}_{i s} = α^{E} + β_{1}^{E} {A M E N I T Y}_{i s} + β_{2}^{E} X_{i s}^{E} + σ_{s}^{E} + ϵ_{i s}^{E},

{Δ C O L L G R A D}_{i s} = α^{C} + β_{1}^{C} {A M E N I T Y}_{i s} + β_{2}^{C} X_{i s}^{C} + σ_{s}^{C} + ϵ_{i s}^{C} .

AMENITY includes the natural amenities and industrial disamenities, X is a vector of control variables (described in the DATA section), $σ$ ’s are state-fixed effects that account for common state-specific factors such as regulatory regime and tax structure, and the ∊’s are error terms. State fixed effects will control for any specific state policies that might lead to higher or lower growth rates. As explained below, we examine changes in both the 1990s and from 2000 to 2007.¹ We specifically choose to cut off our analysis in 2007 in order to avoid including the most recent recession in our analysis. We consider the two decades separately because other research has found some evidence that in recent years the effect of amenities may be changing while the role of economic migration has declined (Partridge et al. 2012). Using 2000 as our split between the time periods also relates to the availability of data in the year 2000. Perhaps the structural change takes place in a nearby year, but due to data limitations, we cannot accurately test for that. Nevertheless, when considering the entire 1990–2007 period, we find that results are similar to those in the 1900s, suggesting that not splitting the data would mask the potential changes in the effect of amenities after 2000. To minimize endogeneity, we use beginning period values for the explanatory variables contained in X; that is, for 1990–2000, we include 1990 explanatory variables, and for 2000–2007, we include explanatory variables from 2000.

These models allow us to test whether lake amenities are positively associated with changes in population, employment, and the share of the population with high human capital in the Great Lakes region. They also provide a means of testing whether there is an offsetting impact from pollution and industrial disamenities and whether there are differences in the eastern versus the western parts of the region. As seen in Table 1, eastern counties have experienced slower population growth, are closer to federally designated hazardous waste (Superfund) sites, have more power plants, and have more water and air pollution than those in the west.

Given that population and employment growth declined after 2000, and because of evidence that the effect of amenities may be changing (Partridge et al. 2012), we are also interested in whether there was a change in preference for lake amenities or a change in productivity associated with being closer to one of the lakes. Thus, we consider the change in the value of the amenity from the 1990s to the post-2000 time period using a first differencing approach.² This approach involves subtracting the change in the measures of growth for population, employment, and college graduates in the 1990s from the change in that same measure in the post-2000 period. This method allows us to difference out the county-level fixed effects and to minimize omitted variable bias. The time-invariant amenity measures are assumed to have different coefficient values in each time period, and the state fixed effects and the east dummy variable (explained below) are eliminated from the estimation equation. The resulting equations also combine the constants and error terms into common terms. The new vector Z includes the time-varying amenity and disamenity measures and the variables previously contained in X in equations (6) –(8).

The reduced form estimation is as follows, where Y is population, employment, or college graduates:

\begin{aligned} Δ Y_{i s 00} - Δ Y_{i s 90} = α^{Y} + (β_{1, 00}^{Y} - β_{1, 90}^{Y}) {A M E N I T Y}_{i s} + β_{2, 00}^{Y} Z_{i s 00}^{Y} - β_{2, 90}^{Y} Z_{i s 90}^{Y} + ϵ_{i s}^{Y} \\ = α^{Y} + Δ β_{1}^{Y} {A M E N I T Y}_{i s} + β_{2, 00}^{Y} Δ Z_{i s}^{Y} + Δ β_{2}^{Y} Z_{i s 90}^{Y} + ϵ_{i s}^{Y} . \end{aligned}

For all models (6–9), we adjust for both standard heteroscedasticity and any clustering of the standard errors based on the Bureau of Economic Analysis (BEA) economic areas. Primo, Jacobsmeier, and Milyo (2007) show that not properly accounting for this correlation can lead researchers to overstate the statistical significance of coefficient estimates.

We also consider the possibility of general spatial error correlation, spatial correlation of the dependent variables, and spatial spillovers of the explanatory variables. We conduct Moran’s I and LM error and LM lag tests using an inverse distance weight matrix and report the results of the robust LM error and robust LM lag tests for the most complete specifications of equations (6) – (8) (excluding those with spatial spillovers) in Tables 2 –4. In some cases, we find no evidence of spatial effects. In other cases, based on the results of these tests, we also estimate spatial models. If both the LM error and the LM lag tests are statistically significant, we estimate either a spatial autoregressive model (SAR) or a spatial error model (SEM) depending on which of the robust LM tests is more significant. However, even in the presence of spatial effects, we find our estimates for our key variables using the designated spatial model as generally being very similar to ordinary least squares (OLS), and thus we report and discuss the OLS results. The results of the extra spatial models are reported in the Appendix. Additionally, as a further control for spatial effects, we estimate models including spatial spillovers of our key economic variables, which we describe in the discussion of the empirical results.

Table 2.

Population Changes in the Great Lakes Region.

	1990–2000						2000–2007
	Model 1		Model 2^a		Model 3		Model 1		Model 2^b		Model 3
Distance to nearest Great Lake	−0.021	(0.007)***	−0.023	(0.005)***	−0.019	(0.005)***	−0.011	(0.007)	−0.009	(0.007)	−0.010	(0.006)
Great Lakes coastal county	−4.077	(1.382)***	−3.734	(1.448)**	−4.080	(1.374)***	−2.588	(1.382)*	−2.389	(1.372)*	−2.525	(1.459)*
% forest area	0.060	(0.039)	0.006	(0.025)	0.003	(0.023)	0.021	(0.021)	0.023	(0.016)	0.023	(0.018)
Topography score	−0.664	(0.887)	−0.049	(0.539)	0.078	(0.534)	0.385	(0.487)	0.249	(0.354)	0.138	(0.329)
January temperature—Z-score	−1.528	(0.984)	−1.402	(1.101)	−1.301	(1.150)	−1.982	(0.680)***	−1.006	(0.524)*	−1.152	(0.401)***
July temperature—Z-score	1.158	(1.562)	0.058	(1.004)	0.076	(1.116)	−1.148	(1.091)	−0.949	(0.665)	−1.008	(0.597)*
% water (not including Great Lakes)	2.005	(0.904)**	1.185	(0.620)*	1.312	(0.607)**	0.442	(0.639)	0.166	(0.564)	0.152	(0.555)
Disamenity index	−0.599	(0.188)***	−0.154	(0.133)	−0.058	(0.126)	−0.288	(0.133)**	−0.073	(0.076)	−0.101	(0.063)
Distance to nearest metro	0.006	(0.019)	0.017	(0.011)	0.002	(0.011)	−0.015	(0.015)	−0.007	(0.009)	−0.001	(0.013)
Incremental distance to metro > 250,000	−0.051	(0.015)***	−0.031	(0.008)***	−0.035	(0.008)***	−0.044	(0.009)***	−0.023	(0.005)***	−0.019	(0.006)***
Incremental distance to metro > 500,000	−0.058	(0.010)***	−0.042	(0.008)***	−0.043	(0.008)***	−0.038	(0.005)***	−0.025	(0.005)***	−0.024	(0.005)***
Incremental distance to metro > 2.5 million	−0.034	(0.010)***	−0.023	(0.006)***	−0.023	(0.007)***	−0.021	(0.006)***	−0.017	(0.004)***	−0.017	(0.005)***
Located within a metro area	5.528	(1.383)***	3.355	(0.880)***	3.144	(0.670)***	5.432	(1.405)***	2.468	(0.810)***	2.011	(0.606)***
East control	−6.110	(1.599)***	−9.557	(1.615)***	−9.017	(1.693)***	−2.592	(0.985)**	−3.949	(1.665)**	−4.270	(1.807)**
East × Distance to nearest Great Lake			0.028	(0.008)***	0.028	(0.007)***			0.018	(0.009)**	0.017	(0.008)**
East × Coastal			3.892	(1.698)**	3.882	(1.757)**			3.278	(1.334)**	3.556	(1.593)**
Other controls?
Demographic	N		Y		Y		N		Y		Y
Industry	N		Y		Y		N		Y		Y
State fixed effects	N		N		N		N		N		N
Other economic controls	N		N		Y		N		N		Y
Adjusted R ²	0.298		0.574		.590		0.288		0.566		0.575
Highest VIF	4.94		7.99		8.17		4.88		9.02		9.4
Number of observations (n)	628		628		624		628		628		624

Note: Demographic includes initial (1990 or 2000) variables representing education levels (% college graduates, % some college, % high school graduates), population density, age (% under 18 and % over 65), race (% white), % married, and % female.

Industry includes initial (1990 or 2000) % of population that was employed, % nonfarm proprietor employment, % manufacturing employment, % agriculture employment, % government employment, and % wage and salary leisure employment.

Other economic controls include 1970 manufacturing employment share and spatially lagged 1990 employment share variables.

(Values in italics and in parentheses are the clustered standard errors using Bureau of Economic Analysis [BEA] economic areas as clusters.)

^a Robust LM Lag Test: 40.61, p value: 0.000; Robust LM Error Test: 9.15, p value: 0.003.

^b Robust LM Lag Test: 1.07, p value: 0.302; Robust LM Error Test: 4.77, p value: 0.029.

*Significance at the 90 percent level. **Significance at the 95 percent level. ***Significance at the 99 percent level.

Table 3.

Employment Changes in the Great Lakes Region.

	1990–2000						2000–2007
	Model 1		Model 2^a		Model 3		Model 1		Model 2^b		Model 3
Distance to nearest Great Lake	−0.012	(0.013)	−0.012	(0.012)	−0.003	(0.010)	−0.001	(0.008)	−0.004	(0.008)	−0.005	(0.007)
Great Lakes coastal county	−4.460	(1.805)**	−3.273	(1.987)	−3.045	(1.596)*	−2.000	(1.406)	0.462	(1.285)	0.607	(1.318)
Percentage forest area	0.132	(0.068)*	0.036	(0.055)	0.064	(0.054)	0.021	(0.025)	0.009	(0.028)	0.015	(0.029)
Topography score	−1.691	(1.477)	−1.079	(1.098)	−1.213	(1.029)	1.742	(0.638)***	0.921	(0.583)	0.848	(0.658)
January temperature—Z-score	−2.949	(1.451)**	−1.477	(2.044)	−1.628	(1.936)	−4.684	(0.759)***	−3.212	(0.787)***	−3.157	(0.775)***
July Temperature—Z-score	1.952	(2.672)	0.682	(2.221)	1.122	(1.898)	0.047	(1.586)	−0.319	(1.374)	−0.354	(1.351)
Percentage water (not including Great Lakes)	3.128	(1.388)**	1.358	(0.937)	1.688	(0.775)**	1.301	(0.848)	0.649	(0.583)	0.697	(0.561)
Disamenity index	−1.116	(0.312)***	−0.393	(0.208)	−0.262	(0.206)	0.050	(0.165)	0.382	(0.134)***	0.336	(0.122)***
Distance to nearest metro	0.027	(0.024)	0.020	(0.018)	−0.006	(0.018)	−0.023	(0.012)*	−0.019	(0.009)**	−0.005	(0.011)
Incremental distance to metro > 250,000	−0.055	(0.018)***	−0.032	(0.013)**	−0.047	(0.013)***	−0.048	(0.010)***	−0.025	(0.009)***	−0.020	(0.009)**
Incremental distance to metro > 500,000	−0.073	(0.016)***	−0.051	(0.016)***	−0.059	(0.013)***	−0.042	(0.007)***	−0.031	(0.007)***	−0.028	(0.008)***
Incremental distance to metro > 2.5 million	−0.061	(0.016)***	−0.045	(0.011)***	−0.052	(0.011)***	−0.014	(0.007)*	−0.009	(0.008)	−0.009	(0.009)
Located within a metro area	7.581	(1.723)***	4.436	(1.602)***	4.220	(1.235)***	6.139	(1.511)***	3.225	(0.875)***	2.832	(0.914)***
East control	−6.753	(2.834)**	−9.205	(2.595)***	−7.609	(2.316)***	−1.869	(1.296)	−2.426	(2.499)	−3.442	(2.394)
East × Distance to nearest Great Lake			0.033	(0.020)	0.032	(0.018)*			0.019	(0.012)	0.021	(0.012)*
East × Coastal			2.040	(2.464)	1.682	(2.279)			0.780	(1.738)	1.322	(1.655)
Other controls?
Demographic	N		Y		Y		N		Y		Y
Industry	N		Y		Y		N		Y		Y
State-fixed effects	N		N		N		N		N		N
Other economic controls	N		N		Y		N		N		Y
Adjusted R ²	0.188		0.432		0.440		0.176		0.424		0.434
Highest VIF	4.94		7.99		8.17		4.88		9.02		9.4
Number of observations (n)	628		628		624		628		628		624

Other economic controls include 1970 manufacturing employment share and spatially lagged 1990 employment share variables.

(Values in italics and in parentheses are the clustered standard errors using Bureau of Economic Analysis [BEA] economic areas as clusters.)

^a Robust LM Lag Test: 27.30, p value: 0.000; Robust LM Error Test: 8.09, p value: 0.004.

^b Robust LM lag test: 0.02, p value: .891; robust LM error test: 0.11, p value: .738.

*Significance at the 90 percent level. **Significance at the 95 percent level. ***Significance at the 99 percent level.

Table 4.

Changes in Shares of College Graduates in the Great Lakes Region.

	1990–2000						2000–2007^a
	Model 1		Model 2		Model 3^b		Model 1		Model 2		Model 3^c
Distance to nearest Great Lake	0.020	(0.007)***	0.019	(0.007)**	0.018	(0.007)**	0.023	(0.015)	0.028	(0.017)	0.023	(0.019)
Great Lakes coastal county	4.172	(1.412)***	4.814	(1.730)***	5.517	(2.006)***	4.501	(1.750)**	5.217	(2.822)*	6.143	(2.942)**
Percentage forest area	0.011	(0.033)	0.009	(0.033)	0.034	(0.028)	−0.024	(0.056)	−0.013	(0.059)	−0.055	(0.070)
Topography score	0.886	(0.714)	0.754	(0.783)	1.102	(0.687)	−1.169	(1.395)	−1.013	(1.424)	−0.254	(1.437)
January temperature—Z-score	−2.449	(1.105)**	−2.416	(1.075)**	−0.753	(1.593)	4.297	(1.953)**	4.148	(1.953)**	2.766	(2.055)
July temperature—Z-score	1.604	(1.241)	1.369	(1.285)	1.865	(1.066)*	−1.342	(2.761)	−1.258	(2.743)	−1.085	(2.759)
Percentage water (not including Great Lakes)	2.283	(0.722)***	2.292	(0.724)***	2.206	(0.953)**	2.708	(1.323)**	2.641	(1.316)*	2.836	(1.468)*
Disamenity index	−0.351	(0.155)**	−0.346	(0.155)**	−0.034	(0.196)	0.566	(0.415)	0.547	(0.416)	0.714	(0.473)
Distance to nearest metro	0.014	(0.017)	0.015	(0.017)	0.018	(0.017)	0.023	(0.028)	0.018	(0.027)	0.006	(0.025)
Incremental distance to metro > 250,000	−0.029	(0.011)**	−0.028	(0.010)***	−0.018	(0.012)	−0.001	(0.013)	−0.004	(0.013)	−0.018	(0.014)
Incremental distance to metro > 500,000	−0.015	(0.013)	−0.015	(0.012)	−0.012	(0.012)	0.037	(0.022)	0.035	(0.022)	0.028	(0.023)
Incremental distance to metro > 2.5 million	−0.034	(0.010)***	−0.032	(0.009)***	−0.031	(0.008)***	0.011	(0.010)	0.007	(0.010)	0.011	(0.014)
Located within a metro area	4.735	(1.192)***	4.677	(1.203)***	7.510	(1.406)***	−6.047	(2.273)**	−6.015	(2.241)***	−0.063	(1.796)
East control	−2.523	(1.434)*	−2.831	(2.655)	−0.073	(2.760)	2.609	(1.741)	5.640	(2.788)**	6.874	(3.395)**
East × Distance to nearest Great Lake			0.005	(0.018)	−0.005	(0.016)			−0.020	(0.013)	−0.029	(0.018)
East × Coastal			−1.276	(2.751)	−3.864	(3.273)			−2.040	(3.254)	−6.057	(3.698)
Other controls?
Demographic	N		N		Y		N		N		Y
Industry	N		N		Y		N		N		Y
State fixed effects	N		N		N		N		N		N
Adjusted R ²	0.092		0.089		0.216		0.047		0.045		0.074
Highest VIF	4.94		6.44		7.99		4.88		6.51		9.02
Number of observations (n)	628		628		628		628		628		628

(Values in italics and in parentheses are the clustered standard errors using Bureau of Economic Analysis (BEA) economic areas as clusters.)

^aThe share of college graduates in 2007 is actually the American Community Survey five-year estimate from 2005 to 2009.

^b Robust LM Lag Test: 4.78, p value: 0.029; Robust LM Error Test: 3.85, p value: 0.050.

^c Robust LM Lag Test: 0.64, p value: 0.424; Robust LM Error Test: 0.06, p value: 0.802.

*Significance at the 90 percent level. **Significance at the 95 percent level. ***Indicates significance at the 99 percent level.

Sensitivity analysis will further test if the results are robust to the model specifications.

Data

We have constructed a unique data set that consists of observations for the counties in the eight states in the US Great Lakes region—Illinois, Indiana, Michigan, Minnesota, New York, Ohio, Pennsylvania, and Wisconsin. Counties within 100 miles of the Atlantic Ocean are excluded from this sample. Counties are the most appropriate geographical level for our study because we are looking for a geographical area large enough to capture a significant share of the population in terms of both place of residence and place of work—that is, having strong features of a functional economic area—in order to measure the effects of household and firm location. Indeed, there is an extensive literature that models how amenities affect both firm and household location at the county level, finding statistically significant effects (e.g., Deller et al. 2001; Rupasingha et al. 2004; Partridge et al. 2008, 2009, 2010; Rappaport 2007). Alternatively, if we are not trying to measure the effects of household and firm location, but rather the valuation of amenities, then we would likely want to use a hedonic model to look at homes that are very proximate to the lakes. Nonetheless, like all geographical units, counties also have their shortcomings.

As discussed in detail below, data are collected from a variety of sources, including the US Census, the BEA, the Department of Housing and Urban Development, the Environmental Protection Agency (EPA), the Department of Energy (DOE), the US Geological Service (USGS), the US Department of Agriculture Economic Research Service (USDA/ERS), and others. ArcGIS is also used to construct a number of specialized variables using data from these sources. This data set allows us to control both for access to lake amenities and the intervening effects of industrial disamenities, while also controlling for other factors that would be expected to explain regional growth. Figure 1 contains a map of the counties in this region and illustrates that there was significant variation in growth during the time period in our analysis. Table 1 shows the full list of variables and some descriptive statistics.

Figure 1.

Great Lakes region, county-level population change, 1990–2007.

Dependent Variables

The analysis uses the following dependent variables:

Percent change in population, 1990–2000 and 2000–2007,³ using data from the BEA (equation 6).

Percent change in employment, 1990–2000 and 2000–2007, using data from the BEA (equation 7).

Percent change in college graduate share, 1990–2000 and 2000–2007, using data from the US Census (equation 8).⁴

Difference in growth rates = (Percentage change in population, employment, or college share growth, 2000–2007) − (Percentage change in population, employment, or college share growth, 1990–2000; equation 9).

Explanatory Variables

The AMENITY variables can be grouped into two categories—natural amenities and industrial disamenities. Natural amenities include measures related to the Great Lakes and other natural amenities that may be important to households. The value of the Great Lakes is measured by (1) distance to the nearest Great Lake in kilometers and (2) a coastal measure that indicates whether or not a county is located on the coast of a Great Lake. Other natural amenities included in the model are measures of relative values of January temperature, July temperature, and topography.⁵ We also create a measure of interior water area that does not include the water area in the Great Lakes. Additionally, we have a measure of the percentage of the county that is in forest cover. These natural amenity data are time invariant and are from USDA/ERS, USGS, or constructed using ArcGIS.⁶ The Great Lakes region also has “lake-effect snow,” with there being higher annual snowfalls in counties bordering the Great Lakes than in noncoastal counties (44.5 vs. 24.9 in.). Thus, in sensitivity analysis, we include average annual snowfall from the National Outdoor Recreation Supply Information System.

Industrial disamenities data include number of power plants, constructed using data from DOE’s Energy Information Administration; and from the EPA, distance to the nearest federally designated toxic waste “Superfund” site and measures of total air and water pollution (in pounds) released in the county (measures of industrialization) using Toxic Release Inventory (TRI) data. Because of high levels of multicollinearity among the disamenity measures, we create disamenity indices for both 1990 and 2000, which are comprised of the sum of the z-scores for the four disamenity measures in each year. This approach is similar to that used by McGranahan (1999) to construct his well-known index of natural amenities. This allows us to assess the offsetting impact of disamenities in the models while avoiding the multicollinearity that arises when the measures are included individually.

In deciding what variables to include in the X vector, we follow the literature in proxying for the other forces that influence household and firm migration such as demographic composition, industry composition, urban proximity, and amenities (Deller et al., 2001; Glaeser, Scheinkman, and Schleifer 1995; Partridge et al. 2008, 2009). Conditioning on other factors that affect household and firm migration is important to ensure that the results are not affected by omitted variable bias. All time-varying explanatory variables are from the initial time period in order to mitigate any endogeneity. Alternative model specifications are considered as robustness tests.

Because the attractiveness of a region is affected by access to urban amenities or urban agglomeration that may increase productivity, we control for urban access and agglomeration through a dummy variable for counties in metropolitan areas; measures of distance (in kilometers) to the nearest metropolitan area, and incremental distances to metro areas with 250,000, 500,000, and 2.5 million people (based on the 2000 Census), which are generated using ArcGIS; and population density, which is measured in 1990 and 2000, using US Census data.⁷ By including measures of proximity to cities of various sizes, we are controlling for whether a community is near a small metropolitan area like Springfield, Illinois, with about 200,000 people, or a large one like Chicago, with over nine million people, which provides access to the full range of urban amenities and services for businesses.

The educational attainment of the population can increase the attractiveness of a region to firms due to productivity benefits and many endogenous growth theories rely on a pool of human capital skills to generate growth. However, for some areas with limited employment prospects, a higher initial share of educated individuals may actually be inversely related to growth. Finally, college graduates may be attracted to areas with higher levels of amenities. Partridge et al. (2008) found that the relationship between college graduate share and growth was generally negative in nonmetropolitan counties in the Midwest; but that there is tremendous heterogeneity in the relationship between college graduates and growth. From the US Census, for 1990 and 2000, we include measures of educational attainment, including percentage of the population aged twenty-five and older with only high school diplomas (or equivalent), those with some college, and college graduates (i.e., bachelors, graduate, professional, and doctorate degrees).

Other demographic measures from the US Census, available for both 1990 and 2000, include racial composition, gender composition, age composition, and percentage of the population that is married. As shown in much of the migration literature, the demographic composition of a region can affect its relative attractiveness to both people and firms. For example, young, single men may be more attracted to a place with a higher percentage of young, unmarried women. Similarly, firms would be more attracted to a place with working age people of both genders. Since people tend to be attracted to places with similar types of people, racial composition may affect migration.

To control for initial economic conditions, we include the initial percentage of the population over sixteen that is employed, which is a measure of the efficiency of the local labor market, and data on industrial composition. From BEA and EMSI⁸ county-level industry employment data, we construct shares of county employment in various industries in 1990 and 2000. Specifically, we use initial percentage of total employment in manufacturing, agriculture, and government; and percentage of wage and salary employment in leisure companies (North American Industry Classification Systems (NAICS) sectors 71 and 72). These employment shares control for economic opportunities and the industrial mix. By including the share of leisure employment, we are controlling for whether there is a high level of employment in industries that are associated with a tourist destination.

The share of 1970 manufacturing employment, from the BEA, controls for historic industrial legacy effects that may prevent a county from taking advantage of lake-related amenity development. Industrial legacy effects include both abandoned factories and other abandoned industrial sites that are not fully accounted for in the disamenity index and a dependence on an industry composition of declining industries; both of which would be expected to contribute to lower growth.

In sensitivity analysis, we include the initial share of recreational homes from the US Census, to control for areas that have a history of being tourist areas. Additionally, if retirees choose to convert these summer homes into retirement homes, this may have long-term effects on population growth. However, we caution that recreational homes may be endogenous, as people may choose to buy recreational homes in places where they expect the economy to do well in order to improve the resale value.

The east–west effect is proxied by a dummy variable, East, that is equal to 1 if the nearest Great Lake is Erie, Huron, or Ontario. We also interact the east variable with distance to the nearest Great Lake and the coastal variable to test whether there is a difference in the effect of lake amenities in the eastern versus the western portions of the Great Lakes region after controlling for other factors.

Since we are interested in the Great Lakes as drivers of growth, ideally we would have good data on water quality. However, as noted by Gyourko, Kahn, and Tracy (1997), since water quality is monitored at the state level and the standards vary across states, consistent measures of water quality for the entire region are unavailable.

To avoid direct endogeneity, the explanatory variables are measured at the beginning of the period. For example, if the dependent variable measures a change from 1990 to 2000, the explanatory variables are from 1990.

Results and Discussion

Because of multicollinearity concerns, we first estimate several parsimonious models that only include the natural amenities measures, the disamenity index, and combinations of the state fixed effects, the East indicator dummy variable, and measures of urban proximity, to assess if there were any relationships between the variables of interest—proximity to amenities and disamenities and the “east” variables—and our growth measures. We then add demographic and industrial control variables as well as measures of spatial spillovers and other economic controls. When state fixed effects are included in the models, there is additional evidence of multicollinearity, although including them does not significantly change the results. Thus, we report estimates without state fixed effects.

Additionally, including both the initial and changes in the explanatory variables when estimating changes in population, employment, and college graduate share growth between the decades (equation 9) introduces multicollinearity into the models. Thus, we report only the models that include the beginning (1990) levels of time-varying explanatory variables in the results, which also minimizes endogeneity. Doing so does not change the key results.

Population Changes

As shown in Table 2, overall proximity to the Great Lakes does not seem to be a major driver of population growth. In the 1990s, being on the coast was associated with roughly 4 percent lower population growth than the rest of the region. This result holds up as we move from model 1, which includes only controls for urban and natural amenities and disamenities and whether or not a county is in the east, through model 3, which includes a full set of demographic and industrial controls. At the same time, however, there is a slightly positive relationship between being closer to one of the Great Lakes and population growth, roughly 0.2 percent for every 10 kilometers from one of the lakes. Since the population-weighted centroid of the average coastal county is 10 kilometers from a Great Lake, the average coastal county in the region has 3.8 percent lower population growth than other counties in the region. One possibility for this result may be that coastal counties have higher population densities; in 1990, coastal counties had an average of 174 people per square mile, while noncoastal counties had only 186 people per square mile, a statistically significant difference (t-statistic = 4.92). This suggests perhaps there are limits to growth along the coasts where population densities are high though many coastal counties are relatively sparsely populated with considerable room for more dense development. However, since our models control for population (or population density), the growth we observe is after accounting for that factor.

Another possible explanation for the negative coastal effect may be due to higher housing prices along the coast. While in 1990, coastal and noncoastal areas had statistically similar housing prices, by 2000, housing prices in coastal counties had jumped so that the average median housing value was $93,639 in coastal counties versus $85,690 in noncoastal counties (difference in means t-statistic = 2.65). Thus, over the decade, rising housing prices may have discouraged people from wanting to live in coastal counties.

We also considered whether the amount of snowfall may be driving this negative coastal effect. However, controlling for snowfall does not affect our results in the 1990s nor is snowfall statistically related to population growth (results not shown).

Perhaps people in this region are just generally interested in having access to the lakes for recreational purposes, but may not care whether or not they directly live on the lake, consistent with a story that natural amenities can have significant distance effects. There is also the possibility that disamenities that are located directly on one of the lakes, like power plants, may keep people from wanting to live directly on the lake.

Overall, eastern counties had lower total population growth in the 1990s. However, eastern coastal counties had slightly higher population growth than other counties in the eastern part of the region. At the same time, for counties in the eastern part of the region, there is a small positive relationship between being farther away from one of the Great Lakes and population growth. The net result is that, in the 1990s, it appears that the eastern part of the region saw almost no population growth effect due to lake amenities. An alternative explanation is that because regression provides the average effect, it may mask the positive gains of a few key coastal counties.

Table 2 also shows that after 2000 the correlation between proximity to one of the Great Lakes and population growth weakens considerably. Again, these results are robust to the inclusion of additional explanatory variables, in moving from model 1 to model 3. As expected from the lack of statistical difference between average eastern and western population growth in that decade (see Table 1), all else equal, both parts of the region experience similar levels of growth and the negative association between being in the east and total population change is smaller. This is also consistent with evidence that amenity-driven migration is declining in recent years. Yet, coastal counties (in the entire region) continue to be negatively associated with population growth. And, as shown in model 3, after 2000, eastern coastal counties continue to see higher population growth (1 percent) than other eastern counties in the region. Interestingly, after 2000, when we control for snowfall, any negative impact on population growth from being on the coast disappears and there is a weak negative statistical relationship between snowfall and population growth (results not shown).

Looking closely at whether there is an offsetting effect of proximity to industrial and environmental disamenities, we see some evidence of a negative disamenity effect. In both decades, higher levels of the disamenity index are associated with statistically significant lower population growth. However, once we control for the industrial mix in model 3 (in both decades), the statistical significance disappears. We were initially concerned that disamenities could be associated with employment opportunities, so we replaced the disamenity measure for 2000 with the measure for 1990 in the post-2000 models (results not shown). The results are qualitatively similar to those using the 2000 disamenity index and we conclude that the measure is not endogenous. This is further verified by the lack of correlation between changes in employment and the disamenities index; −0.16 in the 1990s and −0.02 after 2000.

Employment Changes

Next, we consider employment changes. Overall, there is only a weak, negative relationship between proximity to the Great Lakes and employment growth.

In the 1990s, as Table 3 shows, being on the coast of one of the Great Lakes was weakly associated with less employment growth. This is consistent with the lower population growth observed in those counties. Again, as expected, we also find evidence that, overall, being in the east was associated with lower 1990s employment growth. However, unlike with population growth, eastern coastal counties are not doing better than other eastern counties when it comes to job growth, and this result is robust from model 1 to model 3 as we add explanatory variables. Additionally, in model 3, we observe a (weak) negative relationship between being in the east and closer to one of the Great Lakes and employment growth.

After 2000, as Table 3 illustrates, for the broader region there appears to be no statistically significant relationship between proximity to the Great Lakes and job growth. And consistent with the lack of statistical difference between average eastern and western employment growth in that decade (see Table 1), all else equal, both parts of the region experience similar levels of growth.⁹

We also consider whether there is a relationship between disamenities and employment growth. In the 1990s, the disamenity index appears to be weakly negatively associated with employment growth. This suggests that job growth in the region is probably not due to gains from manufacturing and other businesses which might find proximity to power plants and other sources of pollution to be attractive or productivity enhancing. We see further evidence of this from the positive and statistically significant coefficient on the share of wage and salary leisure employment in the 1990s employment change equation.¹⁰ However, after 2000, the statistically significant correlation between the share of initial leisure employment and overall employment growth disappears and there is now a positive and statistically significant relationship between the level of disamenities and employment growth. This suggests that jobs are being created in different industries after 2000, somewhat supportive of the notion that the lakes became less of an amenity post 2000.

College Graduate Share Growth

To explore whether those with high levels of human capital are more likely to be attracted to high levels of amenities, we consider the growth in the share of college graduates. As Table 4 illustrates, in both decades, coastal counties were associated with higher levels of growth of college graduates, which is consistent with the theoretical models of Roback (1988). Additionally, the statistically significant relationship between percentage of water (excluding the Great Lakes) and the growth in the share of college graduates provides further evidence that college graduates are attracted to natural amenities. For the 1990s, however, away from the coastal counties, being farther from one of the Great Lakes is statistically significant and positively associated with higher growth in college graduates. This suggests that there is a split among college graduates. Some relocate to areas immediately adjacent to one of the lakes and others locate where access to the Great Lakes is not a factor. For example, recreational amenities may draw college graduates and those in creative occupations to places like Traverse City, Michigan. However, the availability of jobs is likely the main driver in relocations to places like Columbus, Ohio, where proximity to Lake Erie, which is more than 100 miles away, is not likely a factor.

We also examined whether there is a difference between the east and west. In the 1990s, there is no statistically significant difference between the east and west in terms of growth of college graduates. But after 2000, the eastern counties actually seem to be doing better, overall, in terms of increasing their share of college graduates. And, in both decades, the eastern coastal counties appear to enjoy the same boost (or even more of one) in terms of attracting college graduates as their counterparts in the west. This suggests that the disamenity index is adequately controlling for industrialization effects and that the relatively poor performance of the eastern region is related to these disamenities. Finally, when we control for annual snowfall, we find virtually no difference from our original results in either decade.

Differences between the 1990s and 2000s

In Table 5, we explore whether the attraction value of lake amenities is changing between the 1990s and post-2000 period by looking at the difference across decades. This also allows us to difference out the unobservable fixed effects (see equation (9)).

Table 5.

Differences between the 2000s and the 1990s^ain how Amenities and Disamenities Affect Growth.

	Population change		Employment change		Change in share of college graduates
Distance to nearest Great Lake	0.011	(0.005)**	0.014	(0.014)	0.015	(0.015)
Great Lakes coastal county	1.185	(0.992)	3.453	(1.984)*	3.683	(1.881)*
Percentage forest area	−0.023	(0.022)	−0.075	(0.068)	−0.078	(0.073)
Topography score	0.790	(0.404)*	2.822	(1.307)**	−0.284	(1.608)
January temperature—Z-score	−0.490	(0.702)	−1.874	(1.708)	4.025	(1.761)**
July temperature—Z-score	−0.430	(0.916)	−0.001	(3.178)	0.588	(2.519)
Percentage water (not including Great Lakes)	−0.890	(0.527)*	−0.317	(1.105)	3.048	(1.211)**
Distance to nearest metro	−0.024	(0.010)**	−0.041	(0.024)*	0.007	(0.022)
Incremental distance to metro > 250,000	0.001	(0.008)	0.001	(0.014)	−0.016	(0.015)
Incremental distance to metro > 500,000	0.010	(0.006)	0.018	(0.018)	0.028	(0.024)
Incremental distance to metro > 2.5 million	0.012	(0.006)*	0.045	(0.014)***	0.016	(0.012)
Located within a metro area	0.088	(0.705)	−0.647	(1.886)	−1.496	(1.908)
Other controls?
Demographic	Y		Y		Y
Industry	Y		Y		Y
Adjusted R ²	0.300		0.189		0.062
Highest VIF	5.08		5.08		5.08
Number of observations (n)	628		628		628

Note: Demographic includes 1990 variables representing education levels (% college graduates, % some college, % high school graduates), population density, age (% under 18 and % over 65), race (% white), % married, and % female.

Industry includes 1990 % of population that was employed, % nonfarm proprietor employment, % manufacturing employment, % agriculture employment, % government employment, and % wage and salary leisure employment.

(Values in italics and in parentheses are the clustered standard errors using BEA economic areas as clusters.)

^aDifference between Population, Employment, and College Graduate Share growth from 2000 to 2007 and 1990 to 2000

*significance at the 90 percent level. **significance at the 95 percent level. ***Significance at the 99 percent level.

There is a positive change in the association between distance to the nearest Great Lake and population growth from the 1990s to the 2000s, suggesting that the value of being closer to the lake declined after 2000 (which is consistent with the results in Table 2). One explanation is that if amenities are normal or superior goods, then the sluggish economy post 2000 may have reduced the demand for lake-based amenities. Another possible reason is that the environmental quality of the lakes worsened after the late 1990s and this may explain the reversal of migration patterns post 2000.

We also difference across decades to see if there is a change in the value of proximity to the Great Lakes relative to employment growth between the 1990s and the post 2000. As shown in Table 5, there is evidence of a weak increasing coastal effect on employment growth across the decades, suggesting that perhaps there is an increasing productivity benefit from being on the coast.

Finally, we consider the difference in growth in college graduate shares across decades. Again, this allows us to difference out the fixed effects and assess whether the value of proximity to the lakes in terms of increasing college graduate share is changing across decades. The positive and statistically significant coefficient for the coastal variable suggests that amenity-based migration for the college graduate population is becoming more important post 2000. And, other natural amenities, as evidenced by the positive and statistically significant coefficient on non-Great Lakes water, also appear to be increasing in value to college graduates. This contrasts with the negative change related to lake amenities we observe for the total population, but is consistent with the positive change for total employment. This is also evidence that the total population changes, which are averages across the entire population, may be masking the heterogeneity in the effect of being closer to one of the lakes or on the coast. It is also consistent with a story that, after 2000, only the highly skilled were able to live near the lakes, possibly due to increasing housing prices. Additionally, these results suggest that perhaps it is the increase in high-skilled workers in coastal counties that is driving the productivity improvements that have led to increases in employment. However, we caution that the low explanatory value of these models, as evidenced by the small values of the adjusted R ², suggests more analysis may be needed.

Sensitivity Analysis and the Role of Recreational Homes

Because of the role that recreational homes can play in economic growth in high-amenity areas (Deller, Marcouiller, and Green 1997), we reestimated each of our core models but included the percentage of recreational homes in that county as a control (results not shown). While we are concerned that recreational homes may be endogenous and may be highly correlated with proximity to the lake, we find that there appears to be a positive and statistically significant relationship between the percentage of recreational homes and overall population growth in both decades. The result of including recreational homes in our models is that the coastal dummy variable in the population change between 1990 and 2000 model loses its statistical significance, although the coefficient remains large and negative. However, results for population change after 2000 remain the same. At the same time, there is no evidence of a statistical relationship between the percentage of recreational homes and either employment growth or growth in the share of college graduates in either decade. Additionally, including a control for recreational homes in our model leaves the other key results virtually unchanged.

Conclusion and Future Research

With the economic recession and ongoing economic restructuring, there is interest in the Great Lakes region in whether the lakes themselves can be drivers of growth. There is an increased sense of urgency in finding new ways to generate growth, especially for the former Rustbelt cities and the entire eastern part of the region, which has experienced a decline in population since the 1970s.

To assess whether the Great Lakes are associated with growth in the region, we examine changes in population, employment, and the share of college graduates. Given the access to fresh water and recreational opportunities that the Great Lakes offer, could the region reinvent itself with amenity-driven growth? Or, will the industrial legacies of the Rust Belt create a repulsion effect that overwhelms the attraction of QOL-enhancing amenities? The evidence is mixed.

There is evidence that being in a county on one of the Great Lakes is associated with growth at the high end of the human capital spectrum, as seen by the relationship with growth in the share of college graduates. This is consistent with work by Roback (1988) that suggests that amenities are more important to those with higher incomes or higher human capital and by Kerr (2011) that shows that amenities are luxury goods. Additionally, Great Lakes coastal counties in the eastern part of the region, with their industrial legacy, seem to be doing just as well as their western counterparts in terms of increasing shares of college graduates, and this trend continues past 2000. It appears that amenity-based migration by those with higher levels of education is happening throughout the region and did not trail off after 2000.

At the same time, overall population changes in the Great Lakes region do not appear to be driven by a strong amenity effect. In the 1990s, there is some evidence that people may be interested in locating close enough to one of the lakes to enjoy their recreational benefits but not necessarily on one of the lakes. This is consistent with studies that show proximity benefits from natural amenities. It may also be that other factors such as housing prices affect exactly where they choose to live and that higher rents near the lake may lead them to choose to be close enough to enjoy the benefits of the lakes, but just far enough away where they do not have to pay the amenity premium in their housing costs. But, between the 1990s and the 2000s, the relationship between proximity to the lakes and population growth declines. Since positive amenities are normal or superior goods, the housing crisis and the two recessions of the 2000s may have made households, in general, less concerned with amenities and disamenities. Alternatively, it could be that the decline in environmental quality of the lakes starting in the late 1990s has made the lakes less attractive after 2000.

One possibility for the relatively weak results for overall migration is that, using county-level data, we are not able to distinguish between those households that live directly on or within a short distance of one of the Great Lakes and those that live within a coastal county but farther from the lake. Thus, in future work we will use micro-level data at the household level to try to distinguish true coastal households from those who simply live “near” one of the lakes.

At first glance, it also appears that the Great Lakes do not seem to play much of a role in overall employment changes in the region, with almost no discernible relationship between proximity to the lakes and employment changes after 2000. However, when we look at whether the value of proximity to the lakes is changing between the decades, we see evidence that being on the coast is becoming more important to employment growth. Perhaps this is due to the increasing numbers of college graduates who are moving into the coastal counties thus creating productivity benefits for employers.

Finally, we note that both nationally and in this region, there is a dramatic decrease since 2000 in both population and employment growth. This could be because rising housing costs may be offsetting amenity benefits for the broader population and resulting in smaller disequilibrium adjustments toward equilibrium in terms of population and employment changes after 2000. In this region, if the higher housing prices mean that only the higher-skilled, higher-income workers and households can afford to live near the lakes, this could have important distributional or welfare effects not explored in this article.

At the same time, if coastal areas in the Great Lakes region can attract high-skilled workers that could have tremendous benefits for the region in terms of transforming it into one that is economically diverse and includes higher-paying, higher-skilled jobs. Previous work has shown that as high-skilled labor enters a market, this can also attract new firms entering the region who demand this high-skilled labor. Through knowledge spillovers and productivity increases as well as bidding between firms, wages can increase (Deller et al., 2001; Partridge and Rickman 2003b; Kim, Marcouiller, and Deller 2005; Rosenthal and Strange 2001, 2004). However, this strategy could also crowd out lower-skilled, lower-income workers from these areas.

There also could be tremendous heterogeneity in terms of the ability of specific communities to implement an economic development strategy focused on attracting high-skilled workers with lake amenities. It may be that a combination of both access to amenities and jobs in specific industries are important to attracting high-skilled workers; thus, it would be interesting to investigate which types of jobs the college graduates are taking. Finally, we note that our results do not suggest that natural amenities in this region are more attractive than those in California, Florida, or the Mountain West. However, they do imply that there may be a way that communities in this region can capitalize on their natural amenities even if they cannot replicate the levels of growth seen in other high-amenity locations. Future work should also consider other measures of growth, such as the change in the share of self-employment or employment in specific industries or occupations.

The real and perceived quality of amenities varies throughout the region and changes over time. For example, there is evidence that many would be tourists think of Lake Erie as dirty (Ohio Sea Grant 2005), while there are positive perceptions of Lake Superior’s water quality. Even when environmental restoration improves quality, changes in perceptions may take time. As the same time, water quality has declined for many parts of the region in recent years and this may or may not be known to potential in-migrants. Overall, this research suggests that for those counties and subregions close to the Great Lakes there may be economic benefits to preserving or restoring the quality of the lakes as a QOL-enhancing amenity.

Footnotes

Appendix

Spatial Models

	SAR		SEM		SAR		SAR
	Population change, 1990–2000		Population change, 2000–2007		Employment change, 1990–2000		Change in college graduate share, 1990–2000
Distance to nearest Great Lake	−0.018	(0.005)***	−0.009	(0.004)**	−0.010	(0.009)	0.016	(0.008)*
Great Lakes coastal county	−3.717	(1.325)***	−2.252	(1.044)**	−3.726	(2.583)	5.138	(2.320)**
Percentage forest area	0.003	(0.021)	0.022	(0.016)	0.026	(0.042)	0.031	(0.037)
Topography score	0.114	(0.485)	0.120	(0.386)	−0.638	(0.947)	1.140	(0.840)
January temperature—Z-score	−0.515	(0.827)	−0.961	(0.692)	0.079	(1.618)	−0.298	(1.480)
July temperature—Z-score	0.108	(0.983)	−0.828	(0.763)	0.687	(1.916)	1.855	(1.704)
Percent water (not including Great Lakes)	1.201	(0.491)**	0.182	(0.376)	1.355	(0.956)	2.179	(0.850)***
Disamenity index	−0.114	(0.141)	−0.078	(0.107)	−0.348	(0.275)	−0.036	(0.245)
Distance to nearest metro	0.017	(0.009)*	−0.008	(0.007)	0.020	(0.018)	0.018	(0.016)
Incremental distance to metro > 250,000	−0.026	(0.007)***	−0.024	(0.006)***	−0.025	(0.014)*	−0.016	(0.013)
Incremental distance to metro > 500,000	−0.036	(0.006)***	−0.026	(0.005)***	−0.042	(0.013)***	−0.010	(0.011)
Incremental distance to metro > 2.5 million	−0.016	(0.005)***	−0.019	(0.005)***	−0.031	(0.011)***	−0.027	(0.010)***
Located within a metro area	3.434	(0.814)***	2.361	(0.633)***	4.518	(1.587)***	7.461	(1.412)***
East control	−7.615	(1.385)***	−3.761	(1.144)***	−7.308	(2.687)***	0.185	(2.387)
East × Distance to nearest Great Lake	0.023	(0.007)**	0.017	(0.006)***	0.029	(0.013)**	−0.004	(0.012)
East × Coastal	3.569	(1.935)*	3.127	(1.500)*	2.159	(3.770)	−3.652	(3.357)
ρ	0.903	(0.092) ***	(N/A)		0.862	(0.127)	0.426	(0.379)
λ	(N/A)		0.758	(0.226)***	(N/A)		(N/A)
Other controls?
Demographic	Y		Y		Y		Y
Industry	Y		Y		Y		Y
Wald test of ρ = 0
Chi-square statistic (p value)	95.79	(0.000)			45.93	(0.000)	1.27	(0.261)
Wald test of λ = 0
Chi-square statistic (p value)			11.31	(0.001)

Note: SAR = spatial autoregressive model; SEM = spatial error model.

(Values in italics and in parentheses are the clustered standard errors using Bureau of Economic Analysis [BEA] economic areas as clusters.)

Demographic includes initial (1990 or 2000) variables representing education levels (% college graduates, % some college, % high school graduates), population density, age (% under 18 and % over 65), race (% white), % married, and % female.

*Significance at the 90 percent level. **Significance at the 95 percent level. ***Significance at the 99 percent level.

Authors’ Note

Previous versions of this article were presented at the 2011 AAEA Meeting in Pittsburgh, PA, and the 2011 RSAI Meeting in Miami, FL, as well as at seminars at The Ohio State University; University of Akron; New Mexico State University; California State University, Long Beach; Washington State University; University of Alaska Anchorage; we thank all participants for their feedback.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Ohio Sea Grant [project R/ME-032].

Notes

References

Banzhaf

H. S.

Walsh

R. P.

. 2008. “Do People Vote with Their Feet? An Empirical Test of Tiebout’s Mechanism.” The American Economic Review 98:843–63.

Chen

Rosenthal

S. S.

. 2008. “Local Amenities and Life-cycle Migration: Do People Move for Jobs or Fun?” Journal of Urban Economics 64:519–37.

Deller

Steven C.

Marcouiller

David W.

Green

Gary P.

. 1997. “Recreational Housing and Local Government Finance.” Annals of Tourism Research 24:687–705.

Deller

S. C.

Tsai

T.-H.

Marcouiller

D. W.

English

D. B. K.

. 2001. “The Role of Amenities and Quality of Life in Rural Economic Growth.” American Journal of Agricultural Economics 83:352–65.

Dorfman

Partridge

M. D.

Galloway

. 2011. “Are High-tech Employment and Natural Amenities Linked: Answers from a Smoothed Bayesian Spatial Model.” Spatial Economic Analysis 6:397–422.

Faggian

Olfert

M. R.

Partridge

M. D.

. 2012. “Inferring Regional Well-being from Individual Revealed Preferences: The “Voting with your Feet” Approach.” Cambridge Journal of Regions, Economy and Society 5:163–80.

Ferguson

Ali

Olfert

M. R.

Partridge

M. D.

. 2007. “Voting with their Feet: Jobs Versus Amenities.” Growth and Change 38:77–110.

Florida

2002a. The Rise of the Creative Class: And How it’s Transforming Work, Leisure, and Everyday Life. New York: Basic Books.

Florida

2002b. “The Economic Geography of Talent.” Annals of the Association of American Geographers 92:743–55.

10.

Gabriel

S. A.

Rosenthal

S. S.

. 2004. “Quality of the Business Environment versus Quality of Life: Do Firms and Households Like the Same Cities?” Review of Economics and Statistics 86:438–44.

11.

Glaeser

1999. “Are Cities Dying?” Journal of Economic Perspectives 12:139–60.

12.

Glaeser

Mare

. 2001. “Cities and Skills.” Journal of Labor Economics 19:316–42.

13.

Glaeser

E. L.

Scheinkman

J. A.

Schleifer

. 1995. “Economic Growth in a Cross-section of Cities.” Journal of Monetary Economics 36:117–43.

14.

Graves

P. E.

1976. “A Reexamination of Migration, Economic Opportunity and the Quality of Life.” Journal of Regional Science 16:107–12.

15.

Graves

P. E.

1980. “Migration and Climate.” Journal of Regional Science 20:227–37.

16.

Graves

P. E.

Mueser

. 1993. “The Role of Equilibrium and Disequilibrium in Modeling Regional Growth and Decline: A Critical Reassessment” Journal of Regional Science 33:69–84.

17.

Greenwood

M. J.

Hunt

G. L.

Rickman

D. S.

Treyz

G. I.

. 1991. “Migration, Regional Equilibrium, and the Estimation of Compensating Differentials.” The American Economic Review 81:1382–90.

18.

Gyourko

Kahn

Tracy

. 1999. “Quality of Life and Environmental Comparisons.” In Handbook of Regional and Urban Economics, edited by Cheshire

P. C.

Mills

E. S.

, Vol. 3, Chapter 37, 1413–54. Amsterdam, The Netherlands: Elsevier.

19.

Irwin

E. G.

2002. “The Effects of Open Space on Residential Property Values.” Land Economics 78:465–81.

20.

Jeanty

P. W.

Partridge

M. D.

Irwin

E. G.

. 2010. “Estimation of a Spatial Simultaneous Equation Model of Population Migration and Housing Price Dynamics.” Regional Science and Urban Economics 40:343–52.

21.

Jeppesen

List

J. A.

Folmer

. 2002. “Environmental Regulations and New Plant Location Decisions: Evidence from a Meta-analysis.” Journal of Regional Science 42:19–49.

22.

Kerr

2011. “The Effect of Amenities on Local Wage Distribution.” Working Paper. http://www.ewi-ssl.pitt.edu/econ/files/faculty/wp/Craig_CRAIG%20KERR%20The%20Effect%20of%20Amenities%20on%20Local%20Wage%20Distributions.pdf

23.

Kim

K.-K.

Marcouiller

Deller

S. C.

. 2005. “Natural Amenities and Rural Development: Understanding Spatial and Distributional Attributes.” Growth and Change 36:273–97.

24.

McGranahan

D. A.

1999. “Natural Amenities Drive Rural Population Change.” AER-781, Economic Research Service, U.S Department of Agriculture.

25.

McGranahan

D. A.

2008. “Landscape Influence on Recent Rural Migration in the U.S.” Landscape and Urban Planning 85:228–40.

26.

Monchuk

D. C.

Miranowski

J. A.

. 2007. “Amenities and Non-farm Employment Growth in the U.S. Midwest: The Impact of Recreational Amenities in Neighboring Counties.” The Review of Regional Studies 37:120–45.

27.

Ohio Sea Grant. 2005. “Lake Erie Coastal Ohio: An Exploratory Research Study.” Survey by the Ohio Sea Grant Program. http://lakeerie.ohio.gov/Portals/0/Closed%20Grants/large%20grants/lepf04-14.pdf

28.

Partridge

M. D.

2010. “The Dueling Models, NEG vs. Amenity Migration in Explaining U.S. Engines of Growth.” Papers in Regional Science 89:513–36.

29.

Partridge

M. D.

Rickman

D. S.

. 2003. “The Waxing and Waning of U.S. Regional Economies: The Chicken-egg of Jobs Versus People.” Journal of Urban Economics 53:76–97.

30.

Partridge

M. D.

Rickman

D. S.

Ali

Olfert

M. R.

. 2008. “The Geographic Diversity of U.S. Nonmetropolitan Growth Dynamics: A Geographically Weighted Regression Approach.” Land Economics 84:241–66.

31.

Partridge

M. D.

Rickman

D. S.

Ali

Olfert

M. R.

. 2009. “Agglomeration Spillovers and Wage and Housing Cost Gradients Across the Urban Hierarchy.” Journal of International Economics 78:126–40.

32.

Partridge

M. D.

Rickman

D. S.

Ali

Olfert

M. R.

. 2010. “Recent Spatial Growth Dynamics in Wages and Housing Costs: Proximity to Urban Production Externalities and Consumer Amenities.” Regional Science and Urban Economics 40:440–51.

33.

Partridge

M. D.

Rickman

D. S.

Olfert

M. R.

Ali

. 2012. “Dwindling U.S. Internal Migration: Evidence of a Spatial Equilibrium or Structural Shifts in Local Labor Markets?” Regional Science and Urban Economics 42:375–88.

34.

Primo

D. M.

Jacobsmeier

Milyo

. 2007. “Estimating the Impact of State Policies and Institutions with Mixed-level Data.” State Politics and Policy Quarterly 7:446–59.

35.

Rappaport

2004. “Why are Population Flows so Persistent?” Journal of Urban Economics 56:554–80.

36.

Rappaport

2007. “Moving to Nice Weather.” Regional Science and Urban Economics 37:375–98.

37.

Rappaport

Sachs

J. D.

. 2003. “The United States as a Coastal Nation.” Journal of Economic Growth 8:5–46.

38.

Roback

1982. “Wages, Rents, and the Quality of Life.” Journal of Political Economy 90:1257–78.

39.

Roback

1988. “Wages, Rents, and Amenities: Difference among Workers and Regions.” Economic Inquiry 26:23–41.

40.

Rosen

1979. “Wage-based Indexes of Urban Quality of Life.” In Current Issues in Urban Economics, edited by Mieszkowski

Straszheim

, 74–104. Baltimore, MD: Johns Hopkins University Press.

41.

Rosenthal

S. S.

Strange

W. C.

. 2001. “The Determinants of Agglomeration.” Journal of Urban Economics 50:191–229.

42.

Rosenthal

S. S.

Strange

W. C.

. 2004. “Evidence on the Nature and Sources of Agglomeration Economies.” In Handbook of Regional and Urban Economics, edited by Henderson

J. V.

Thisse

J. F.

, Vol. 4, 2119–71. Amsterdam, The Netherlands: Elsevier.

43.

Rupasingha

Goetz

S. J.

Debertin

Pagoulatos

. 2004. “The Environmental Kuznets Curve for US Counties: A Spatial Econometric Analysis with Extensions.” Papers in Regional Science 83:407–24.

44.

Schmidt

Courant

P. N.

. 2006. “Sometimes Close is Good Enough: The Value of Nearby Environmental Amenities.” Journal of Regional Science 4 6:931–51.

45.

Tiebout

C. M.

1956. “A Pure Theory of Local Expenditures.” Journal of Political Economy 64:416–24.

46.

Wehrwein

G. S.

Johnson

H. A.

. 1943. “A Recreation Livelihood Area.” The Journal of Land & Public Utility Economics 19:193–206.

47.

Wooldridge

J. M.

2002. Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT Press.