The rise of the Sunbelt.

Glaeser, Edward L. ; Tobio, Kristina

1. Introduction

In the 1930s, the American South seemed trapped in the poverty and relative decline that had marked that region since the Civil War. The 11 states of the former Confederacy were depicted by William Faulkner as quaint remnants of a pre-modern world and by both Nazis and communists as embarrassing evidence of the limits to American freedom and economic opportunity. Since World War II, however, the South has been a great regional success. Figure 1 shows that the South's share of the U.S. population increased from 24% in 1950 to 30% today, with the South defined, as it is throughout this paper, as the 11 states of the former Confederacy. In 1950, the average Southern county had an income that was 76% of the average income across all counties. In 2000, the average Southern county's income was 94% of the county average. Between 1950 and 2000, the average housing price in Southern counties grew from 83 to 91% of the U.S. average.

[FIGURE 1 OMITTED]

The rise of the South is part of the general correlation between warmth and growth across the entire United States. The Sunbelt--all places, Southern or not, with warm winters and hot summers--has experienced a boom since the 1950s. Throughout this paper, we will consider three different proxies for Sunbelt status: high January temperatures (indicating warm winters), high July temperatures (indicating hot summers), and location in one of the states of the former Confederacy. Table 1 gives the cross-county correlations between our three proxies for Sunbelt status and the growth of population, income, and housing values decade by decade. As Table 1 shows, the correlation between population growth and warm winters was more than 10% in every post-war decade. Figure 2 shows the correlation across counties between population growth from 1950 to today and the presence of warm winters, indicated by a county's average January temperature over the 1970-2000 time period. Figure 3 shows the somewhat weaker relationship between income growth and warm winters over the same period. Figure 4 shows the relationship between housing price growth and warm winters and, although some of this relationship must be attributed to improving housing quality in the Sunbelt, this correlation is quite robust.

While there can be little doubt that the Sunbelt boomed in the decades after World War II, the causes of this boom are less clear. In section 2, we discuss three broad possible explanations for the success of the South and the Sunbelt: increasing productivity, rising demand for Sunbelt amenities, and a more flexible housing supply. Many authors, from McDonald (1961) to Caselli and Coleman (2001), have documented the remarkable economic performance of the South during the post-war era and have suggested that strong economic growth should lead to population growth. Other authors have proposed that population growth in the Sunbelt reflects increasing demand for Southern amenities due to sun-related technological change, such as the rise of air conditioning (Borts and Stein 1964; Graves 1980; Mueser and Graves 1995). A final hypothesis is that the South has grown not because it is more attractive or more productive, but because its housing supply is far more elastic, mainly due to a pro-development regulatory system (Glaeser, Gyourko, and Saks 2006).

Section 3 presents a framework based on Rosen (1979) and Roback (1982) that uses changes in population, income, and housing prices to assess the potential sources for Southern and Sunbelt growth. The model predicts that rising productivity will cause population, nominal income, and housing prices to rise. When productivity increases, income will rise faster than housing prices, and real incomes, defined as nominal income corrected for local prices, will also surge. Rising amenity levels or an increasing willingness to pay for the amenities of a location will cause population and housing prices to rise, but nominal and real incomes will fall. (1) An increase in housing supply will cause population to rise and both income and housing prices to fall.

This framework enables us to estimate the relative growth of productivity, amenities, and housing supply in the Sunbelt, along with the relative contribution of these forces to the growth of the region. The framework depends critically on the spatial equilibrium assumption that assumes that different "real incomes" across space offset different amenity levels. The most problematic aspect of our application of the Rosen-Roback spatial equilibrium framework is that we ignore the forward-looking aspects of housing prices, but we hope that future work will remedy this weakness.

In section 4, we estimate the relationship between our proxies for Sunbelt status and the growth of population, income, and housing. Since the changes in the quality of the housing stock in the South appear to be enormous, we use only the areas for which we have Office of Federal Housing Enterprise Oversight (OFHEO) repeat sales indices from 1980 onward. This limits our sample to 135 metropolitan areas and, as a result, our results differ from the correlations in Table 1, which includes all U.S. counties. We use census median housing values for years before 1980.

[FIGURE 2 OMITTED]

In univariate regressions across metropolitan areas, each of these variables predicts population growth in every decade since 1950, except for July temperature, indicating hot summers, which demonstrates almost no effect in the 1960s. However, across metropolitan areas, the correlation between population growth and January temperature, indicating warm winters, has declined since the 1970s. In contrast, the correlation between July temperature and growth has been rising since the 1970s. This change reflects the relative slowdown in the growth of California and the explosion of metropolitan areas such as Las Vegas, Houston, and Atlanta.

Using the same metropolitan area--level data, we find that the correlation between income growth and the South dummy is strongly positive between 1950 and 1980 but that the relationship has weakened since then. The correlations between income growth and the other two variables are less reliable. The correlation between the South dummy and housing prices is weak, except for the 1980s, when housing price growth strongly declines with the South dummy. The correlation between January temperature and housing price growth is also somewhat weak, except for the 1970s, when housing price growth strongly rises with warm winters. Housing price growth and July temperature correlations have been negative since 1960 and are most strongly negatively correlated in the 1980s.

[FIGURE 3 OMITTED]

These basic findings from metropolitan area aggregate census data are corroborated by the Census Individual Public Use Micro Sample (IPUMS) data, which allow us to control for increases in the education level in the Sunbelt. Controlling for individual attributes using the micro data does not change the basic trend of rising real incomes in the South since 1970. When we control for local cost of living using American Chamber of Commerce Research Association (ACCRA) indices, we also find that real incomes in the South have risen steadily since 1970. There is also an increasing correlation between hot summers and real income, but there is a declining correlation between warm winters and real income. These differences suggest, quite plausibly, that while the amenity flows associated with warm winters are rising, the amenity flows associated with hot summers are falling.

In section 5, we use the parameter estimates from these regressions to estimate the shocks to productivity, amenities, and housing supply. We estimate strong positive shocks to productivity in the South between 1960 and 1980 and somewhat weaker shocks after then. The association between productivity growth and warm winters is strong between 1950 and 1990, and only disappears in the 1990s. The association between productivity and hot summers is strong between 1950 and 1980 and weaker after then.

The biggest surprise in this paper is that we estimate declining amenity flows in the South and in all places with hot summers throughout the entire post-war period. A central insight of the Rosen-Roback model is that in a spatial equilibrium, higher amenity levels must be offset by lower real wages as people become willing to accept lower wages in return for enjoying attractive amenities. Declining amenity flows are therefore implied by rising real incomes, as people demand higher wages to compensate them for the lack of amenities. One objection to this inference is that real incomes may be rising because of improvements in human capital, but over much of the recent period, rising real incomes reflect minimal housing price growth rather than strong income growth. While the amenity flows associated with the South and with hot summers have uniformly fallen, the amenity value of a warm winter rose in the 1960s and 1970s. California, for example, appears to have had robust amenity flow increases as its real wages have generally fallen since 1970. However, most of the Sunbelt is not in California, and in most areas, housing prices have increased far more modestly than income.

[FIGURE 4 OMITTED]

We estimate dramatic relative housing supply growth in the South and in places with hot summers between 1970 and 1990. We infer housing supply increases because the stock of housing rose dramatically but housing prices did not. Putting these estimates together, we find that population growth in the Sunbelt was driven primarily by productivity increases between 1950 and 1980, but housing supply has played an increasingly important role in the growth of the Sunbelt since then.

The rise of Southern productivity has been well studied, but there has been far too little attention paid to the fact that the housing supply is growing so quickly in the South. In the penultimate section of the paper (section 6), we are unsuccessful in our attempt to find a simple explanation for the high level of Southern construction, using factors like lower initial land density and governmental fragmentation. Section 7 concludes.

2. Why Did the South Rise Again?

There are three natural explanations for the population growth in the South since 1950. First, the region may have become more economically productive. Second, the region may have become a more attractive place to live. Third, the region might be particularly good at producing new housing. We first discuss these three hypotheses, and in the next section, we present an empirical methodology that will allow us to apportion credit for the South's success to these three forces.

Many economists have offered different explanations for the rise in Southern income levels. Barro and Sala-I-Martin (1992) emphasize the greater accumulation of capital in once-backward places, and, as late as 1940, the South was certainly still backward. Caselli and Coleman (2001) present a related theory focused on the structural transformation out of agriculture into industry, which implies that since the South was much more agricultural in 1940, it transformed more quickly. Another view is that the Northern productivity edge came from better access to waterways and a dense railroad network, advantages that became increasingly irrelevant as transportation costs plummeted during the 20th century (Glaeser and Kohlhase 2004). (2)

Other authors connect the improvements in the Southern economy to changing political institutions. Besley, Persson, and Sturm (2005) emphasize the rise of political competition that came after the end of the Jim Crow South. Cobb (1982) also points to the end of the Jim Crow South and emphasizes the importance of the switch from leaders who cared about white supremacy to leaders who cared about subsidizing industry. Holmes (1997) presents particularly compelling evidence on the importance of Southern right-to-work laws, while Olson (1983) suggests that Southern pro-growth policies reflect a lack of anti-growth interest groups in the South.

In this paper, we will not distinguish between the many different theories of Southern economic development. Rather, we assess the importance of economic development to Southern growth relative to an alternative view that emphasizes consumption and amenities. The growth of the Sunbelt can be attributed to amenities either if Southern amenities improved or if unchanged Southern amenities became more attractive to people as their incomes increased. The amenity improvement hypothesis argues that Southern cities were relatively unpleasant places to live at the start of 20th century because of their oppressive summers, incidence of disease, and lack of clean water. Over time, public health improvements and technological changes, such as the introduction of air conditioning, have certainly made the South a far more attractive place to live. (3) The findings of Mueser and Graves (1995) suggest that unchanged amenities have driven Sunbelt growth, and, as such, when society as a whole got richer, demand increased for Sunbelt and Southern amenities such as mild winters and beautiful scenery. Both of these arguments, however, propose that the South's growth has been driven primarily by consumer interaction with amenities, not by productivity gains.

The third hypothesis is that the increases in Southern population reflect increases in neither Southern productivity nor Southern amenities, but rather Southern tolerance for new construction (Glaeser, Gyourko, and Saks 2006). Permitting is certainly abundant in some Southern metropolitan areas; in 2005, Atlanta, Dallas, and Houston were three of the five metropolitan areas that permitted most. In an extreme version of this view, all of America experienced rising housing demand, but the supply of housing was much more elastic in the South. Since housing supply was more flexible, more homes were built, and more people came to live in the South.

The South's initial low housing density and its lack of natural barriers, such as rivers, have likely contributed to its large supply of developable land, which in turn may have increased permitting. Alternatively, the difference in permitting behavior could be the result of different regulatory environments. High-permitting Houston, for example, famously lacks zoning, in contrast to the stringent regulatory environments that appear to have played a major role restricting growth in many areas outside of the South (Glaeser, Gyourko, and Saks 2006). Another explanation might be that the same political forces that gave the South a political regime that favored economic growth may have also given the South a regime that favored new construction.

Although this discussion has emphasized the growth of the South, the same three basic hypotheses can also explain the growth of places within the Sunbelt in general. Since the colder regions of the country developed first, productivity increases could have been faster in warmer places. There could certainly be an increased demand for the amenities of warm winters and hot summers, and it is certainly plausible that housing supply increased disproportionately in warm areas. Since it is more plausible that people developed a stronger taste for warm winters than for hot summers, we use both January and July temperature variables in our analysis. Since our three measures of the Sunbelt status are not identical, we estimate our decomposition for each measure separately.

3. A Decomposition of the Sources of Sunbelt Growth

We now turn to a formal framework that will enable us to assess the relative importance of productivity growth, amenity growth, and housing supply growth to the expansion of the Sunbelt. We start with a simple spatial equilibrium model that assumes that firms and workers are indifferent across space at all points in time. This model is a descendant of Rosen (1979) and Roback (1982) and is similar to the framework used by Glaeser, Scheinkman, and Shleifer (1995), except that it introduces housing supply heterogeneity. As opposed to Mueser and Graves (1995), we simplify and do not worry about forward-looking behavior or forward--looking housing prices.

Every location in the United States is characterized by a location-specific productivity level of A and a firm output of [AN.sup.[beta]][K.sup.[gamma]][Z.sup.1-[beta]-[gamma]], where N represents the number of workers, K is traded capital, and Z is nontraded capital. Traded capital can be purchased anywhere for a price of 1. The location has a fixed supply of nontraded capital equal to [bar.Z]. Firms behave competitively, and this delivers a labor demand curve from the firms' first order conditions. This labor demand curve depends on nominal wages, not wages correcting for local prices, which suggests that nominal wages are our primary tool for learning about shocks to local productivity.

Consumers have Cobb-Douglas utility functions defined over tradable goods (sold at a fixed price of 1) and nontraded housing, denoted H, or [theta][C.sup.1-[alpha]][H.sup.[alpha]]. Optimizing behavior yields the indirect utility function [[alpha].sup.[alpha][(1 - [alpha]).sup.1-[alpha]][theta]W[p.sup.-[alpha].sub.H]. If a spatial equilibrium holds, then this welfare level must equal a reservation utility level, denoted [U.bar]. The spatial equilibrium assumption does not mean that wages corrected for local price (real wages) are equal across space, but that higher real wages in some places are offsetting lower amenity levels. (4) We will assume that this spatial equilibrium holds at every point in time, which does not mean that labor or capital needs to adjust perfectly at all points in time, but just that housing prices are sufficiently flexible to offset differences in wages and amenities.

Housing is produced competitively with height, denoted h, and land, denoted L. The total quantity of housing supplied equals hL. There is a fixed quantity of land in the location, denoted [bar.L], which will determine an endogenous price for land, denoted [p.sub.L], and housing, denoted [p.sub.H]. The cost of producing hL units of structure on top of L units of land is [c.sub.0][h.sup.[delta]]L. Given these assumptions, the profit to a developer for producing hL units of housing is [p.sub.H]hL-[c.sub.0][h.sup.[delta]]L - [p.sub.L]L, where [delta] > 1, and this must equal zero since there is free entry of developers. The first order condition for height then implies the area's housing supply.

Together, the firms' labor demand equation, the equality between indirect utility in the town and reservation utility, and the housing price equation are three equations with the three unknowns of population, income, and housing prices. Solving these equations for the unknowns gives us

Log(N) = [K.sub.N] + ([delta] + [alpha] - [alpha][delta])Log(A) + (1 - [gamma])([delta]Log([theta]) + [alpha]([delta] - 1)Log([bar.L]))/[delta](1 - [beta] - [gamma]) + [alpha][beta]([delta] - 1), (1)

Log(W) = [K.sub.W] + ([delta] - 1)[alpha]Log(A) - (1 - [beta] - [gamma])([delta]Log([theta]) + [alpha]([delta] - 1)Log([bar.L]))/[delta](1 - [beta] - [gamma]) + [alpha][beta]([delta] - 1), (2)

and

Log([p.sub.H]) = [K.sub.P] + ([delta] - 1)Log(A) + ([beta]Log([theta]) - [1 - [beta] - [gamma])Log([bar.L]))/([delta](1 - [beta] - [gamma]) + [alpha][beta]([delta] - 1), (3)

where [K.sub.N], [K.sub.W], and [K.sub.P] are constant terms that differ across cities, but not within a city, over time. These are the static equilibrium equations that serve as the basis for our growth regressions. To make these equations dynamic, we first assume that they hold at all points in time and then assume that changes in productivity, amenities, and housing supply are characterized by the growth equations Log([A.sub.t+l]/[A.sub.t]) = [K.sub.A] + [llambda].sub.A] S + [[mu].sub.A], Log([[theta].sub.t+1]/[[theta].sub.t]) = [K.sub.[theta]] + [[lambda].sub.[theta]]S + [[mu].sub.[theta]], and Log([[bar.L].sub.t+1]/[[bar.L.].sub.t]) = [K.sub.L] + [[lambda].sub.L]S + [[mu].sub.L]. In these equations, [K.sub.A], [K.sub.[theta]], and [K.sujb.L] are constants; [[lambda].sub.A], [[lambda].sub.[theta]], and [[lambda].sub.L] are coefficients; [[mu.sub.A], [[mu.sub.[[theta], and [[mu.sub.L] are error terms; and S is a variable reflecting Sunbelt status. Equations 1 3 then imply the following:

Log([N.sub. t + 1]/[N.sub.t]) = [K.sub.[DELTA]N] + [[chi].sup.-1](([delta] + [alpha] - [alpha]delta])[[lambda].sub.A] + (1 - [gamma])([delta][[lambda].sub.[theta]] + [alpha]([delta - 1)[[lambda].sub.L]))S + [[mu].sub.N], (4)

Log([W.sub. t + 1]/[W.sub.t]) = [K.sub.[DELTA]W] + [[chi].sup.-1](([delta] - 1)[alpha][[lambda].sub.A] - (1 - [beta] - [gamma])([delta][[lambda].sub.[theta]] + [alpha]([delta - 1)[[lambda].sub.L]))S + [[mu].sub.W], (5)

Log([P.sub. t + 1]/[P.sub.t]) = [K.sub.[DELTA]P] + [[chi].sup.-1](([delta] - 1) ([[lambda].sub.A] + [beta][[lambda].sub.[theta]] - (1 - [beta] - [gamma])([[lambda].sub.L]))S + [[mu].sub.P], (6)

where [chi] = ([delta](1 - [beta] - [gamma]) + [alpha][beta]([delta] - 1))).

Equations 4-6 suggest how to interpret regressions where changes in population, income, and price are regressed on a proxy for Sunbelt status. Once again, these proxies for Sunbelt status are January temperature (warm winters), July temperature (hot summers), or a South dummy. If [[??].sub.N], [[??].sub.W], and [[??].sub.P] represent the estimated coefficients on the Sunbelt measure for the population, wage, and price change regressions, then the model tells us that [[gamma].sub.A], the impact of Sunbelt status on productivity growth, equals a weighted average of the coefficients in the population and income regressions, (1 - [beta] - [gamma])[[??].sub.N] + (1 - [gamma]) [[??].sub.W]. The impact of Sunbelt status on amenity growth, [[gamma].sub.[theta]], equals [alpha][[??].sub.P] - [[??].sub.W], which subtracts the impact on wages from the impact on prices. This is a dynamic version of the Rosen-Roback method of inferring amenities from the weighted average of prices and income. The impact of Sunbelt status on land supply, [[gamma].sub.L], equals [[??].sub.N] + [[??].sub.W] - ([delta][[??].sub.P]/([delta] - 1)), which sums the coefficients in the population and income regressions and subtracts the coefficient from the price regression multiplied by a measure of housing supply inelasticity.

This framework does not include forces that might slow transitions to long-run equilibrium, such as costs to the rapid installation of capital or of new construction. These forces could indicate that rising population or wages do not reflect rising productivity, but instead reflect the slow movement of capital to the Sunbelt. As long as prices adjust quickly, relative to the decadal frequencies at which we examine the data, then these barriers to mobility of capital and construction will not change the fact that changes in real wages tell us about changes in amenities.

In the next section, we present our estimates of [[??].sub.N], [[??].sub.W], and [[??].sub.P], which we then use in section 5 to estimate [[lambda].sub.A], [[lambda].sub.[theta]], and [[lambda].sub.L], and thus to estimate the relative contribution of productivity, amenity, and housing supply to the population growth of the Sunbelt.

4. Population, Housing Price, and Income Growth in the Sunbelt

We now turn to the correlations between our three different proxies for Sunbelt status and population, housing price, and income growth. We restrict our sample to the set of 135 metropolitan areas for which OFHEO price indices are available since 1980. Changes in Southern housing quality have been dramatic enough that we believe these indices must be used, as they allegedly correct for changes in housing quality. Our results on population and income changes are robust to the inclusion of more metropolitan areas. The inclusion of more nonmetropolitan area counties does reduce the correlation between July temperature and growth, as demonstrated by a comparison of Table 1 to our regression tables.

When we turn to income and housing price growth, we use data from the Census Individual Public Use Micro Sample (IPUMS). With the metropolitan area-level data, we will be able to consider the entire post-war period from 1950 to 2000. With individual-level data, we will only consider the period between 1970 and 2000.

Population Growth

Table 2 examines population growth in the Sunbelt, with the change in the logarithm of population decade by decade regressed on our three proxies for Sunbelt status. The table follows the basic structure that we will repeat in Tables 3 and 4, where each column refers to a separate decade-level regression. The rows of the table represent results from four different regressions for each decade, one regression for each of our three proxies and one regression including all three proxies.

The first regression includes only an intercept and a South dummy variable that takes on a value of 1 if the metropolitan area is located in one of the 11 states of the former Confederacy. The second regression includes an intercept and a continuous variable capturing the average January temperature, in units of hundreds of degrees, which indicates warm winters. The third regression includes an intercept and a continuous variable capturing the average July temperature, again in units of hundreds of degrees, which indicates hot summers. The fourth regression includes all three proxies for Sunbelt status simultaneously and helps us to recognize the differences between our three different proxies for location in the Sunbelt.

In our decompositions, we will focus on the univariate regressions rather than the regressions including all three variables. The variables are highly correlated, and multicollinearity issues become significant when we include all three in the regressions. Moreover, our goal is to focus more on a single phenomenon--the rise of the Sunbelt--rather than trying to distinguish separately the impact of a number of different measures of Sunbelt status holding other measures constant.

The first column of the table characterizes population growth in the 1950s. During this period, warm winters predict population growth much more strongly than either of the other two Sunbelt proxies. The first regression shows that metropolitan areas in the South grew 0.041 log points faster than non-Southern metropolitan areas in the 1950s. This coefficient is positive, small, and statistically insignificant. The second regression shows that in the 1950s, 10 extra degrees of January temperature are associated with an 0.084 log point increase in population growth. Mean January temperatures explain more than one-fifth of the variation in metropolitan area growth rates in the 1950s. The third regression shows that a 10-degree increase in July temperatures is associated with an increase in the growth rate of about 5%. The relationship is not statistically significant and explains only 2% of the variation in metropolitan area growth.

The final regression combines all three variables. The estimated coefficients on January and July temperatures both increase in magnitude. The estimated coefficient on location in the South switches signs and becomes both significant and negative. This regression suggests that the immediate post-war growth of the South was actually associated with the success of warm places generally. The results from this regression suggest, if anything, that the South substantially underperformed relative to other warm places.

The second column gives results for the 1960s. All three proxies for Sunbelt status become weaker during this decade. The coefficient on July temperature flips signs. Only the mean January temperature variable remains robustly positive and continues to explain about 15% of the variation in population growth, but even the impact of this variable is only about 50% of its value in the 1950s. In the multivariate regression, July temperature and the South dummy have a negative effect.

The impact of the South dummy on population growth gets substantially stronger after 1970. The estimated coefficient in the third regression is 0.086, which implies that Southern areas grew almost 9% faster than their Northern equivalents during the 1970s. In columns 4 and 5, the South dummy variable has a significant coefficient of approximately 0.06 in the 1980s and 0.065 in the 1990s. The growth of the South is particularly a post-1970 phenomenon, which logically seems to support the idea that the Civil Rights era was a watershed in the history of the South. Related work, such as Donohue and Heckman (1991), finds a break in the pattern of African-American migration in the 1960s. They find that the first half of the decade is characterized primarily by out-migration and the second half of the decade, after the Civil Rights Act in 1964, is characterized primarily by in-migration.

The effect of warm winters is strong in both the 1970s and 1980s, explaining 34% of the variation in population growth in the 1970s and 38% of the variation in population growth in the 1980s. The coefficients of 0.73 and 0.64 tell us that a 10-degree increase in January temperature is associated with a population increase of about 7% in each decade. In the 1990s, the coefficient on January temperature declined to 0.31 and the R2 of the regression declined to 14%. Warm winters became a less potent predictor of population after 1990.

The opposite pattern holds for hot summers, which became a more significant predictor of growth in the 1990s than it had been before then. A 10-degree increase in July temperature was associated with an approximately 4% increase in growth in the 1970s and 1980s, but with an approximately 5% increase in growth in the 1990s. The R2 of July temperature increased between the 1980s and 1990s. Though hot summers remain a less important predictor of population growth than warm winters, the gap between the two variables is closing.

The fourth multivariate regression shows that the South never grows when we control for warmth using the temperature variables. As such, we can only examine the disproportionate growth of the South without controlling for warmth. The coefficients on January and July temperature show the same patterns as the univariate regressions, with warm winters becoming much less significant in the 1990s and hot summers becoming more significant over time. We will look for this pattern when we turn to other variables.

Income Growth

To examine the correlation between income growth and location in the Sunbelt, we use two related methodologies. First, with the same sample of metropolitan areas we used in the population growth regressions, we use metropolitan area-level census data to regress the decade by decade difference in the logarithm of median income on our three measures of Sunbelt status. The results of these regressions are found in Table 3. While these regressions enable us to look at the entire post-war period from 1950 on, they do not enable us to control for the changes in the human capital of the South, changes in the returns to human capital, or changes in the quality of human capital (Card and Krueger 1992).

In order to control for observable measures of human capital, we use individual-level IPUMS data to run regressions of the form

Log(Annual Income) = Year Dummies + Sunbelt Status

+ Sunbelt Status * Year Dummies

+ Human Capital Measures

+ Human Capital Measures * Year Dummies. (7)

By controlling for human capital measures, which include age and race dummies as well as a fourth order polynomial in years of schooling, we can control for human capital differences across the United States. The interactions between these variables and the year dummies allow the returns to human capital to change over time. Furthermore, these interactions allow the quality of human capital nationwide to change over time but do not control for any disproportionate increase in human capital quality in the South (as in Card and Krueger 1992). Though the IPUMS data allow us to control for human capital measures, we can only use data for the 1970-2000 time period.

We run regressions including only males between 25 and 55 years of age who work 40 or more weeks per year, who work 35 hours or more per week, and who have earnings above 50% of the earnings of someone who works full time over the year and earns the minimum wage. We have also run a set of regressions using a more restricted set of metropolitan areas for which we have ACCRA indices of local costs of living.

Table 3 presents our basic income growth regressions using the metropolitan area-level data. Just as in the case of Table 2, the columns give results for regressions for each decade. The rows give results for our different proxies for Sunbelt status. In the 1950s, income in the South grew by almost 3% more than income in the rest of the country. In the 1960s and 1970s, Southern incomes truly soared, growing by almost 8% and more than 6% in the two decades, respectively. During these decades, the South dummy explains between 19 and 27% of the variation in income growth across metropolitan areas. In the 1980s and 1990s, income growth in the South was far more modest. In the 1980s, there is almost no correlation between the South dummy and growth, while in the 1990s, there was about 1.8% higher growth in the South.

The temporal pattern for income growth is almost the exact opposite of the temporal pattern for population growth. The connection between population growth and location in the South increased after 1980, while the connection between income growth and location in the South decreased after that date. Only in the 1970s did Southern location strongly predict both income and population growth. This timing mismatch is a clue that different processes might have been driving Southern success in the early post-war period versus the more recent time periods.

The second set of regressions in Table 3 looks at the generally weak correlation between income growth and warm winters. In the 1950s, 1970s, and 1980s, a 10-degree increase in January temperature is associated with income growth between 0.01 and 0.014 log points. In the 1960s and the 1990s, warmer winters are negatively associated with income growth. These findings suggest that the strong correlation between warm winters and population growth is probably not the result of a connection between warm winters and productivity growth.

The third set of regressions uses hot summers as the proxy for Sunbelt status. The pattern for the mean July temperature variable is similar to the pattern for the South dummy. Between 1950 and 1980, hot summers positively predict income growth. A 10-degree increase in July temperature is associated with an income growth of about 2% in each decade. After 1980, there is a small and negative correlation between hot summers and growth. Since the correlation between population growth and hot summers is strongest for the later period, when the correlation between income growth and hot summers is weakest, we are again led to the view that different forces may be driving the early and late time periods in our data.

The last set of regressions includes all three variables. In these regressions, the partial effect of location in the South is always positive but much smaller in the 1950s and 1980s than it is in the other three decades. The South's strongest decade was the 1960s, where incomes grow about 12% more than the rest of the country. The effects of warm winters and hot summers follow a similar pattern as the univariate regressions.

Table 5 produces our results for the individual-level income regressions. The first regression uses data for individuals living in the 86 MSAs that appear in the IPUMS data for all four decades. The first regression shows the coefficient on location in the South and the changes in the correlation with income growth over the decades, using census wages adjusted for national, not local, prices using the Consumer Price Index. The coefficient on Southern location of -0.107 implies that incomes were 0.107 log points lower in the South than in the rest of the country in 1970, even while controlling for education, age, and race. The interaction between Southern location and a post-1980 dummy variable tells us that Southern incomes rose by 0.041 log points in the 1970s. During the 1970s, over one-third of the wage difference between the South and the rest of the United States disappeared. After that point, wage gains for the South were uneven. The South closed ground with the rest of the United States in the 1970s but has essentially plateaued since then, suggesting that productivity growth in the South may have tapered off after 1980.

A comparison of these regressions with the metropolitan area--level income change regressions suggests that controlling for individual human capital measures does not change our results substantially. The individual-level regressions suggest somewhat less income growth than the straight income change regressions in the 1970s and somewhat more income growth in the 1990s. Both approaches show no income growth in the 1980s.

The next two regressions show results for real wage changes, adjusted for local prices using ACCRA data. In the second regression, we include results for the entire four-decade period. In this case, we are restricted to the 22 metropolitan areas for which we have ACCRA data for the entire period. In the third regression, we provide results only for 1990 and 2000, which allows us to increase the sample to the 68 metropolitan areas for which we have 1990 and 2000 ACCRA data.

The second regression suggests that real incomes, controlling for human capital, were actually higher in the South than outside the South in 1970. Lower Southern wages were more than offset by lower Southern prices (as shown by Bellante 1979 and DuMond, Hirsch, and Macpherson 1999). The Rosen-Roback framework infers from this fact that the South had lower amenities in 1970, which is certainly plausible. The coefficient on the interaction between the South dummy and the dummy for years after 1980 is 0.028, which suggests a modest rise in real income in the South in the 1970s. However, the coefficient on the interaction between the South dummy and the dummy for years after 1990 is slightly lower at 0.020. The interaction between the South dummy and the year 2000 is 0.081, which suggests that real incomes rose even more steeply in the 1990s.

In the third regression, with its much larger sample, we estimate a basic coefficient of 0.032 on Southern location. This means that the real wage gap between South and non-South was smaller in this more inclusive sample than in the more restricted sample. The coefficient on the interaction between the South dummy and the year 2000 dummy is 0.087, which is quite similar to the 0.081 coefficient estimated in the second regression. Both specifications suggest that real incomes soared in the South in the 1990s, driven both by rising nominal incomes and declining price levels relative to the non-South.

These regressions imply that the rise of the South has been accompanied by declining, not rising, amenity levels, and though there may certainly be infra-marginal migrants who are drawn to the South for its amenities, the marginal migrant is presumably drawn by something else. Higher real wages in the South in 1970 suggest that amenities were initially lower, and the continuing rise in real wages also suggests the willingness to pay for Southern amenities declined relative to the willingness to pay for amenities in the non-South. These findings of real wage increases are quite compatible with the view of Mueser and Graves (1995), who proposed that rising incomes would increase the demand for pre-existing amenities, such as attractive locales or mild winters. These are Southern amenities that have remained unchanged through the years. However, these results are not compatible with the view that the growth of the South was driven by an increasing willingness to pay for improved Southern amenities since, if that were the case, we would have found that real wages were falling in the South.

The middle panel of Table 5 examines the relationship between warm winters and census wages. The first regression in this panel uses wages adjusted by the national Consumer Price Index and shows the coefficient on January temperature and its interactions with three time dummies. The coefficient on January temperature itself, -0.188, indicates that in 1970, as temperature increased by 10 degrees, wages fell by about 2%. Thirty years ago, places with warm winters had slightly lower wages.

The coefficient on the interaction between January temperature and the post-1980 dummy variable is -0.014, which means that as the January temperature increased by 10 degrees, wages fell by 14%. This wage change is very small and negative and is quite different from the income growth associated with January temperature in the 1970s in Table 3. The best explanation for this discrepancy is that income growth in places with warm winters in the 1970s reflected significant human capital upgrading. The coefficient on January temperature and the post-1990 dummy variable is 0.039, which means that as the January temperature increased by 10 degrees, wages increased by nearly 4%. The coefficient on January temperature on the year 2000 dummy is negative, but small and insignificant. These regressions show that lower incomes are no longer associated with warm winters.

The second regression in the middle panel gives our income results for the 22 metropolitan areas for which we have ACCRA cost of living data since 1970. The raw coefficient on January temperature is positive, suggesting that real wages were slightly higher in these warmer places in 1970. A 10-degree increase in January temperature was associated with a 1.25% increase in wages. The interaction between January temperature and the post-1980 variable produces a coefficient of 0.138, which means that a 10-degree increase in January temperature increased wages by 1.38%. This result suggests that the positive connection between real wages and warm winters actually rose in the 1970s.

The interaction between January temperature and the post-1990 dummy is -0.377, which means that real wages in places with warm winters declined substantially in the 1980s. The interaction between January temperature and the year 2000 dummy produces a coefficient of -0.383. Real wages were relatively declining in places with warm winters between 1980 and 2000, which suggests that the value placed on warm winters rose substantially over the past 20 years.

The third regression in the middle panel includes those 68 metropolitan areas for which we have ACCRA data for 1990 and 2000. The baseline coefficient on January temperature indicates that a 10-degree increase in January temperature in 1990 was associated with a 1.89% decrease in real incomes during that year. The interaction between January temperature and the 2000 dummy indicates that a 10-degree increase in January temperature is associated with 1.83% slower real income growth during the 1990s. This regression confirms the results of the second regression in that real wages have been declining in places with warm winters. Because of these wage declines, the Rosen-Roback model then implies that the amenity flows associated with warm winters have been rising as people are willing to accept lower wages in return for living in a place with a more pleasant climate.

The bottom panel in Table 5 looks at hot summers and census wages. In our first regression, which adjusts wages using the Consumer Price Index, the coefficient on July temperature of -0.685 suggests that a 10-degree increase in July temperature was associated with an almost 7% reduction in wages in 1970. The interaction between July temperature and the post-1980 dummy is 0.184, which suggests that about one-fourth of the connection between July temperature and low incomes disappeared during the 1970s. This coefficient is lower than the coefficient on July temperature in the 1970s income growth regression reported in Table 3, which implies that the income growth in places with hot summers, like those with warm winters, had something to do with human capital upgrading.

The interactions between July temperature and both the year 2000 dummy and the post-1990 dummy are small and statistically insignificant. When we hold human capital constant, we find that there has been little change in incomes in places with hot summers over the past 20 years. Places with hot summers continue to have lower wages than the rest of the country.

The second regression in the bottom panel controls for local price levels for those 22 metropolitan areas for which we have ACCRA indices for four decades. The basic coefficient on July temperature is 0.777, which suggests that a 10-degree increase in July temperature is associated with almost an 8% increase in real incomes. Places with hot summers in 1970 had low nominal incomes but high real incomes, as low housing costs more than offset the low wages in that decade.

The coefficient on the interaction of July temperature and the post-1980 dummy is small and positive. A 10-degree increase in July temperature is associated with an approximately 2% increase in real wages during the 1970s. The interaction between July temperature and the post-1990 dummy is much larger. In the 1980s, a 10-degree increase in July temperature was associated with a 6% increase in real wages. The interaction between July temperature and the year 2000 dummy is stronger yet. A 10-degree increase in July temperature was associated with more than an 11% real income growth in the 1990s.

The third regression in the bottom panel looks at the 68 metropolitan areas for which we have ACCRA price data in 1990 and 2000. In this broader sample, the baseline coefficient on July temperature is 0.404, which suggests that people needed 4% higher real wages in 1990 to compensate for a 10-degree increase in July temperature. In this sample, the interaction between July temperature and the year 2000 dummy is 0.614, which suggests that by 2000, people required 6% higher real wages to compensate them for living in places where their summers were hotter by 10 degrees.

Our real income data show striking differences between different measures of Sunbelt status. Warm winters were associated with lower real wages in 1990, and this effect just got stronger throughout the 1990s. Hot summers and location in the states of the former Confederacy were associated with higher real wages in 1990, and those effects also got stronger throughout the 1990s. This pattern is quite understandable. High incomes in society as a whole should make us willing to pay more for mild winters, but we would be unlikely to be willing to pay more for very hot summers, even with air conditioning. This implies both that the real wage premium required to live through hot summers is rising and that the real wage premium required to live through cold winters is also rising.

Housing Price Growth

Though we have already begun to address housing prices through the ACCRA indices, which include housing costs, in this subsection we turn directly to the correlations between housing prices and our proxies for Sunbelt status. In Table 4, we repeat the basic structure of Tables 2 and 3, using the metropolitan area--level data to regress housing price growth on our three proxies for location in the Sunbelt. In the first three decades, we use median housing prices from the U.S. Census. These prices do not control for housing quality, which is changing dramatically in the South during this time period. For the 1980s and 1990s, we use OFHEO price indices, which are meant to correct for changes in housing quality.

The top rows of the table show the correlation between location in the South and housing price growth. The first two columns show small positive correlations between location in the South and housing price growth. Location in the South was associated with a 4% added increase in housing prices in the 1950s and a 0.6% increase in price growth in the 1960s. However, if housing quality was indeed rising faster in the South during this period, then real prices might have actually been falling in the South even during these early decades. In the 1970s, housing prices in the South fell by 3.5% relative to the rest of the nation.

The OFHEO real price index for the 1980s and 1990s shows that real prices declined spectacularly in the South in the 1980s. Southern location was associated with about 17% lower housing price appreciation between 1980 and 1990. In the 1990s, there was no correlation between Southern location and faster price growth.

The second set of regressions in Table 4 shows the correlation between housing price growth and warm winters. In four out of five decades, this warmth variable is almost uncorrelated with housing price growth. In the 1950s, 1960s, and 1980s, a 10-degree increase in January temperatures was associated with extra housing price growth of between 1 and 2%. In the 1990s, a 10-degree increase in January temperature was associated with 4% less growth in housing prices. Only in the 1970s did warm winters strongly predict faster housing price appreciation. In that decade, when California became far more expensive, a 10-degree increase in January temperatures was associated with 8.3% faster price growth. January temperature explained almost 30% of the variation across metropolitan areas in price growth in the 1970s.

The third set of regressions looks at hot summers. In the 1950s, there was almost no correlation between hot summers and housing price growth. In the 1960s, a 10-degree increase in July temperature was associated with an approximately 4.5% decrease in housing price appreciation. In the 1970s, a 10-degree increase in July temperature correlates with an approximately 5% decrease in housing price appreciation. If housing quality was also increasing more quickly in places with hot summers during this time period, then the cost of housing was dropping even more quickly during these decades.

Our results in the fourth column, where we use OFHEO price indices for the 1980s, show an even more spectacular relative decline in housing prices in the 1980s. A 10-degree increase in July temperature is associated with about 17% lower housing price growth between 1980 and 1990. This is part of the large rise in real incomes in places with hot summers during that decade. The last column shows that hot summers continued to be associated with lower housing price appreciation in the 1990s, but the estimated coefficient is much smaller.

The final set of regressions includes all three proxies for Sunbelt status in each regression. The patterns are quite similar to the other regressions. Warm winters are positively associated with price growth for every decade except for the 1990s, when housing prices in California fell significantly. The connection between warm winters and price growth was extremely strong in the 1970s and still quite robust in the 1980s. Hot summers are always negatively associated with housing price growth, and the effect is strongest in the 1980s. The South dummy positively predicts price growth in the 1950s, 1960s, and 1990s and negatively predicts price growth in the 1970s and the 1980s. Holding temperature constant, housing prices in the South appear either to have been quite weak or declined in most of the post-war period.

Table 6 once again uses the IPUMS to look at housing price growth, and using the individual-level data allows us to control for observable housing characteristics. We are able to control for basic features of the house, such as number of rooms and age, but we are not able to control for more subtle aspects of housing quality. As such, we consider these price regressions to have worse quality controls than those that are based on the OFHEO price indices, but better quality controls than those using raw median housing values.

In the first regression, found in the top panel of Table 6, we look at the South dummy and its interaction with different year dummies. The raw coefficient on the South dummy is -0.318, which suggests that Southern housing costs are about 30% less than non-Southern housing in 1970. Assuredly, some portion of this gap reflects unmeasured quality differences between the South and the rest of the nation. The interaction between the South and the 1980 dummy suggests that this difference did not change at all in the 1970s. The interaction between the South dummy and the 1990 year dummy produces a coefficient of -0.147, which suggests that Southern prices continued to fall relative to the non-South in the 1980s. Finally, the coefficient on the interaction between the South dummy and the year 2000 dummy is 0.02, which suggests that Southern prices increased quite modestly in the 1990s.

In the middle panel of Table 6, we look at warm winters. The raw coefficient on January temperature is slightly and insignificantly negative, suggesting that in 1970 places with warm winters were not more expensive than places with cold winters. The interaction between January temperature and the year 1980 dummy is positive and suggests that a 10-degree increase in January temperatures was associated with an 8% increase in prices in the 1970s.

In Table 6, the interaction between January temperature and the year 1990 is positive but insignificant and the interaction between January temperature and the year 2000 is negative and insignificant. The IPUMS confirms the basic fact that, on average, housing prices were not rising faster in places with warm winters over the last 20 years.

The last regression in the bottom panel of Table 6 looks at warm summers. The raw impact of July temperature is negative and significant. None of the interactions with year dummies are significant, although they are estimated reasonably precisely. In this case, there is a slight divergence between Tables 4 and 6. The OFHEO indices suggest that housing prices are getting significantly cheaper in places with hot summers in the 1970s, while the IPUMS results suggest a somewhat smaller decrease in price. We are prone to put more weight on the OFHEO measures, since quality levels were increasing significantly in places with hot summers. The IPUMS and OFHEO results are somewhat similar for the 1980s in that they both show declining prices, but the IPUMS decline is smaller.

5. Productivity, Amenity, and Housing Supply Changes

We now take our regression estimates from section 4 and estimate the productivity, amenity, and housing supply shocks that hit the Sunbelt. We return to the model and its predictions about the link between regression coefficients and the underlying parameters of the model. First, we discuss the results for the South, and then we turn to our two temperature measures.

The South

The model suggests that the connection between Sunbelt status and productivity growth ([[lambda].sub.A] equals (1 - [beta] - [gamma])[[??].sub.N] + (1 - [gamma])[[??].sub.W], or a weighted sum of the coefficients in the population growth and the income regressions, where [[??].sub.N] and [[??].sub.W] are the coefficients on the South dummy for the population and wage change regressions, respectively. The weight on the population regression is the share of production associated with immobile capital. The weight on the income regression is the share of production associated with labor plus immobile inputs. We take 0.6 for labor's share of input costs ([beta]) and 0.3 for the share of mobile capital in inputs ([sigma]). Our estimates do change with different parameter estimates, but the changes are modest.

Each column in Table 7 shows results for different decades. For the first two decades, we can only use the income change regressions for our estimates of [??].sub.W], since we only have metropolitan area--level data for this time period. For the 1970s, 1980s, and 1990s, we also show results using the coefficients estimated in the individual income regressions. We give results for the correlation between productivity growth and our three Sunbelt proxies separately. As such, we provide six estimates of productivity shocks for the later two decades. In all cases, we provide bootstrapped standard errors for the estimates.

The first two rows give the results for the correlation between location in the South and productivity growth. In the 1950s, 1960s, and 1970s, productivity growth in the South was dramatic. In the 1950s, we estimate that productivity grew by 3% more in the South than outside the South. In the 1960s, the Southern edge in productivity growth was over 6%. These results confirm the findings of writers like Hulten and Schwab (1984), Chinitz (1986), and Wright (1987), who document the post-war productivity increases in the South. However, these results do not confirm the view that Southern wages rose during this period because of Southern emigration.

These estimates do not control for individual characteristics and thus will confuse the upgrading of human capital with increases in actual productivity. In the 1970s, we are able to compare results estimated from mean income data with results estimated from individual data controlling for human capital levels. Using the mean income data, we estimate a productivity increase of 0.068 log points in the South in the 1970s. Using the IPUMS data, we estimate a smaller productivity increase of 0.05. Controlling for human capital decreases the estimated productivity surge, but the South still seems to have become significantly more productive in the 1970s.

In the 1980s, we estimate little productivity improvement in the South relative to the rest of the nation. The IPUMS estimates show a slight productivity decrease. The aggregate data show a slight productivity increase. In the 1990s, the South again appears to be gaining productivity, but the effect is smaller than in the 1970s or 1980s. The IPUMS and metropolitan area data suggest a productivity increase of approximately 3%.

In the third and fourth rows of the table, we look at our estimates of [lambda][theta] the connection between Sunbelt status, and rising amenities, which equals [alpha][[??].sub.P] - [[??].sub.W], where [alpha] is the share of expenditure going toward housing, and [[??].sub.P] is the coefficient from the price change regression. The term [alpha][[??].sub.P] - [[??].sub.W] also equals the decrease in real wages. We assume that [alpha], the share of expenditure going on housing, equals 0.3, which is based loosely on the Consumer Expenditure Survey. Our results are not particularly sensitive to this assumption. In the 1950s, we estimate a slight negative amenity decline of -0.017 log points. In the 1960s, the amenity decline is far more severe at -0.076 log points. This pattern of declining amenity levels appears over the next three decades as well. The metropolitan area--level data suggest values of -0.07, -0.05, and -0.02 for the 1970s, 1980s, and 1990s, respectively. The IPUMS regressions suggest values of -0.05, -0.02, and -0.03 for the three decades.

These estimates corroborate the real income results shown in Table 5. Over the past 30 years, real incomes have been rising in the South, even controlling for individual characteristics. Real incomes also appear to have been rising between 1950 and 1970, although we only have metropolitan area--level data for this time period and are therefore unable to control for individual human capital levels in these estimates. If amenities were booming in the South, then housing prices should have been going up more quickly than incomes in the South. However, our results are not compatible with this view, and we do not find evidence that the amenity flows in the South have been increasing relative to the rest of the country over the past 50 years.

To us, this result may be the biggest surprise of our empirical work. Despite all of the obvious improvements in Southern amenities, including air conditioning and the mitigation of disease, on the margin, people appear to have valued Southern amenities less in 2000 than they did in 1950. This result does not mean that air conditioning didn't matter. Without it, Southern amenity levels would surely have fallen further. It does mean that Southern growth does not seem to be caused by Southern amenity levels rising faster than the rest of the nation.

In the final rows of Table 7, we look at the connection between housing supply growth and Sunbelt status, [[lambda].sub.L], which equals [[??].sub.N] + [[??].sub.W] - ([delta][[??].sub.P]/([delta] - 1)), where 8 reflects the elasticity of housing supply. The value of 8 strongly impacts our parameter estimates. We use values of 1.5, which implies that the elasticity of price with respect to density is 0.5, and 3, which implies that the elasticity of price with respect to density is 2. The results are sensitive to different values of 5, but the basic finding of rising housing supply in the South after 1970 is not.

In the 1950s, there were no significant differences in housing price growth between the South and the non-South. When we let 8 equal 1.5, Southern housing seems to be growing more slowly than housing supply elsewhere. When we let [delta] equal 3, Southern housing seems to be growing more quickly than housing supply elsewhere. In neither case is the result significant. In the 1960s, housing supply does appear to be increasing significantly faster in the South than in the rest of the country, and the effect lies between 0.077 and 0.086, depending on the value of [delta].

These estimates are economically quite modest. However, in the 1970s and 1980s, the housing supply growth is substantially higher in the South. In the 1970s, our four estimates of [[lambda].sub.L] range from 0.18 to 0.25. These effects are more economically meaningful and statistically significant. In the 1980s, the estimates range from 0.29 to 0.57, which are quite sizeable. The estimates of [[lambda].sub.L] in the 1990s are in a narrow range around 0.065 to 0.087, similar to the estimates for the 1960s. As such, we find that the housing supply in the South really boomed relative to the rest of the country between 1970 and 1990.

We now ask how important housing supply is in explaining economic growth, relative to economic productivity. Equation 4 shows that the overall coefficient on the South dummy in a growth regression equals ([delta] + [alpha] - [alpha] [delta])[[lambda].sub.A] + (1 - [lambda])[delta][[lambda].sub.[theta]] + [alpha]([delta] - 1) [[lambda].sub.L]) times a constant. To compare the relative importance of productivity and housing supply, we must multiply the productivity effect by [1/(1 - [lambda])] {[delta]/[[alpha]([alpha] - 1)] - 1 }. If we let 8 equal 1.5, this expression equals 12.8. If [alpha] equals 3, the value of this multiplier falls to 5.7.

With multipliers in this range, it is clear that rising productivity, not rising housing supply, drove the South's boom in the first two post-war decades. Our estimate of [[lambda].sub.A] is 0.032 in the 1950s, and our estimate of [[lambda].sub.L] ranges from -0.05 to 0.01. With a multiplier of 5.7 or more, the productivity effect dwarfs the housing supply effect. In the 1960s, our estimate of [[lambda].sub.A] is 0.066, and our estimate of [[lambda].sub.L] ranges from 0.077 to 0.086. Given the range of estimates for [1/(1 [lambda])]{[delta]/[[alpha]([delta] - 1)] - 1}, the productivity effect is between 4 and 12 times as important as the housing supply effect.

In the 1970s and 1980s, however, housing supply becomes more important relative to productivity growth. For the 1970s, our estimates of [[lambda].sub.A] range from 0.05 to 0.068, and our estimates of [[lambda].sub.L] range from 0.18 to 0.25. Given a multiplier of 5.7, the productivity and housing supply effects are roughly similar in magnitude. Higher values of the multiplier will mean that productivity growth was still more important than housing supply growth. In the 1980s, our estimates of [[lambda].sub.A] range from -0.011 to 0.011, and our estimates of [[lambda].sub.L] range from 0.29 to 0.57. During this decade, relative housing supply increases in the South were extremely large and relative productivity growth was negligible.

The 1990s represent something of a reversion to the earlier time periods where productivity growth seems slightly more important than housing supply growth. During this period, our estimates of [[lambda].sub.A] range from 0.027 to 0.038, and our estimates of [[lambda].sub.L] range from 0.065 to 0.087. With a lower multiplier of 5.7, productivity seems about twice as important as housing supply growth to the growth of the South in the 1990s.

Overall, the pattern suggests that the relative population growth of the South was never driven by amenities. Prior to 1970, population growth was driven by rising productivity. In the 1970s and 1990s, housing supply growth and productivity growth were about equally important in driving Southern population growth. Housing supply growth was almost completely responsible for Southern population growth in the 1980s.

January and July Temperatures

We now turn to Tables 8 and 9, our parameter estimates for January and July temperature. Both temperature measures are strongly associated with productivity growth in the 1950s and 1970s. The productivity growth effect in the 1970s is weaker when we use the individual-level IPUMS estimates than when we use the metropolitan area level data for both temperature measures. This reflects the fact that some part of this rising productivity is also reflecting rising human capital levels in disproportionately warm places. Warmer places also had positive productivity growth in the 1960s, but the effect is smaller and not statistically significant. Since we are unable to control for human capital changes during that decade, this increase may just reflect improvements in education.

The two temperature measures diverge in the 1980s and 1990s. In the 1980s, places with warm winters saw their productivity increase, but places with hot summers did not. This was, after all, a boom era for California. In the 1990s, places with hot summers saw their productivity levels increase more quickly than places with warm winters. While the 1950-1980 period saw a roughly parallel pattern of productivity growth for the entire Sunbelt, during the 20 years since then, places with mild winters and places with hot summers have increasingly diverged.

Perhaps unsurprisingly, the connection between temperature and amenity growth has always been quite different for the two different warmth measures. Hot summers have been negatively associated with amenity growth for every one of the post-war decades. This may just reflect the fact that hot summers are always unpleasant, with or without air conditioning, and as people have become richer, they have become increasingly willing to pay to avoid those hot summers. Warm winters are, however, positively associated with amenity growth in the 1960s and especially the 1970s, when January temperature strongly predicted housing price growth.

In the 1950s, warm winters had a slightly negative impact on amenity growth. In the 1980s, the results are mixed. Using the aggregate data, we find a slight amenity downturn. Using the individual-level data, we find a slight amenity upturn. In the 1990s, we find a weak negative correlation between warm winters and amenity growth. These results are quite sensitive to the booms and busts of the California housing cycles, so perhaps it is unwise to read too much into them. The general finding is that warm winters have sometimes been associated with rising amenity values, but hot summers have never been associated with rising amenity values.

The connection between temperature and housing supply growth also differs between the two measures. Hot summers are always strongly associated with housing supply expansion, relative to the rest of the county. The effect of hot summers on housing supply growth was weakest in the 1950s and strongest in the 1980s. In all three of the post-1970 decades, places with hot summers have had extremely high levels of housing supply expansion.

The effect of warm winters on housing supply growth has been far more mixed. In the 1950s and 1990s, warm winters predicted housing supply growth. In the 1970s, when California became much more anti-development, warm winters were associated with less housing supply growth. In the 1960s and 1980s, warm winters were weakly positively associated with housing supply growth.

How important were amenities, productivity, and housing supply in explaining the relative growth of warm places? Rising amenities had nothing to do with the rise of the places with hot summers, since our estimates suggest that the amenity flows in these places have been declining. During the 1950s, rising productivity was more important than rising housing supply in explaining the growth of places with both warm winters and hot summers. In the 1960s, if the multiplier [1/(1 - [lambda])]{[delta]/[[alpha]([delta] - 1)] - 1} equals 5.7, the productivity growth was more important than housing supply growth in places with warm winters but less important than housing supply growth in places with hot summers. If the multiplier is higher, then productivity growth may have been more important in both types of areas.

In the 1970s, productivity was certainly much more important than housing supply growth in explaining the success of places with warm winters, since housing supply did not disproportionately rise with January temperature. When we look at the effect of hot summers in the 1970s, our estimates of [[lambda].sub.A] range from 0.224 to 0.3, and our estimates of [[lambda].sub.L] range from 1.37 to 2.27. If [delta] equals 1.5, then housing supply growth is much more important than productivity growth. If 8 equals 3, then productivity growth is more important than housing supply growth for places with hot summers.

In the 1980s, productivity growth was more important than housing supply growth for places with warm winters, but housing supply growth was more important than productivity growth for places with hot summers. In the 1990s, housing supply growth was more important than productivity growth in places with hot summers if [delta] equals 1.5, but productivity growth was more important than housing supply growth if [delta] equals 3. Housing supply growth was certainly more important than productivity growth in places with warm winters in the 1990s.

The growth of high temperature areas has generally not been driven by an increasing willingness to pay for their amenities. Instead, rising populations were the result of rising productivity during most of the post-war period and, since 1970, were the results of large increases in housing supply, relative to the rest of the nation.

6. Why Is Housing Supply Increasing Rapidly in the Sunbelt?

While there has been a great deal written about rising productivity in the South, much less has been written about why housing supply appears to be increasing so much more quickly in that region. As a final exercise, we will see whether any simple factor can explain why new housing has been so much more abundant in the Sunbelt.

In Table 10, we look at the relationship between our three different proxies for Sunbelt status and the logarithm of the number of housing permits in a county between 2000 and 2005, which we use as our measure of new residential construction in a county. We examine only those counties with more than 50,000 inhabitants in 2000. In all regressions, we include the logarithm of total land area in the county so that the regression can be understood as trying to explain the logarithm of new housing per acre. Our basic methodology is to first regress the logarithm of new housing on one of three proxies for Sunbelt status and the logarithm of new acreage.

We then include four other variables. Two of these variables, the logarithms of income and median housing values in 2000, are meant to proxy for demand. The other two variables, the logarithm of housing density in 2000 and the number of governmental units in the county, are meant to proxy for supply conditions. Greater housing density in 2000 should capture the possibility that Southern places build more because they begin with less housing. The number of government units may decrease development if small jurisdictions are more prone to restrict development because they fail to internalize the benefits to the entire community provided by increased building. This hypothesis suggests that small bedroom communities are likely to associate new development with increased congestion and increased public costs, while large cities are able to better understand the benefits of new development, such as the creation of lower costs for businesses.

In the first regression, we see the basic impact of the South dummy on the logarithm of new residential construction after 2000. The coefficient suggests that those counties in the South have roughly a 20% greater propensity to permit than counties outside of the South when holding only land area constant. In the second regression, we see that controlling for the other four characteristics not only fails to explain the greater Southern propensity to develop, but also increases the size of the coefficient to approximately 0.7. There is a remarkable correlation between location in the South and permitting that is only made stronger by controlling for density, prices, and income.

In the third column, we look at the correlation between warm winters and permitting. A 10-degree increase in January temperature is associated with approximately 25% more permitting. Controlling for price and income in the fourth regression does essentially nothing to this coefficient. Again, the warmth-permitting correlation is not explained by land density, price, income, or the number of governmental units.

The fifth and sixth regressions look at hot summers. The raw impact of July temperature is that a 10-degree increase in July temperature increases permitting by about 36%. When we control for the other variables, this coefficient increases, so that a 10-degree increase in July temperatures increases permitting by about 65%. The seventh and eighth regressions include all three of our Sunbelt proxies. In regression 8, we see that when we use the full roster of controls, the South dummy and July temperature both strongly predict permitting, while January temperature does not. Thus, the regression results found in Table 10 show that a number of simple theories cannot explain the Sunbelt predilection for construction.

7. Conclusion

The traditional view of the post-war South emphasizes the economic convergence of this region. This view is certainly correct for the period between 1950 and 1980 and even in the 1990s, when productivity grew more quickly in the states of the former Confederacy. Moreover, our estimates suggest that the increasing population of the Sunbelt prior to 1970 was driven almost entirely by the increasing association between warmth and economic productivity.

We found little evidence to support the view that the growth of the Sunbelt had much to do with sun-related amenities. Though we do estimate rising amenity flows in places with warm winters during the 1960s and 1970s, this is the only evidence we find of rising amenity flows in the Sunbelt. The positive relationship between real incomes and July temperatures indicates amenity flows are falling for places in the Sunbelt with hot summers. Additionally, real incomes in the South also appear to have been steadily rising, which suggest that amenity flows are also falling in the South. The negative amenity flows we estimate for the South and hot summers corroborate these findings. However, our results do not mean that amenity improvements in the South such as air conditioning or clean water were irrelevant, as we suspect that amenity flows would have been far lower without them. Our results do suggest that over time, the marginal resident required more and more compensation for living in the South.

The final significant fact about Sunbelt growth is that since 1970, the greater expansion of Southern housing supply has been particularly important. We estimate that housing supply was increasing by at least 20% more in the South than elsewhere in the country in the 1970s and 1980s. Ten extra degrees of July temperature was similarly associated with 20% or more housing supply growth. Over the past three decades, our estimates suggest that faster housing supply growth in the South has been at the very least almost an equally important factor as economic productivity in driving the rise of Sunbelt population. The causes of greater housing supply in the Sunbelt remain a pressing topic for future research.

Appendix

MSA-level data on population, income, and housing value are from the U.S. Census Bureau. Income is median family income. Housing value is the median value of specified owner-occupied housing units. These data are used in the regressions found in Tables 2-4.

Income and housing value data are adjusted using the Consumer Price Index at: http://www.bls.gov/cpi/home.htm.

Census housing values for the 1980s and 1990s are replaced with Office of Federal Housing Enterprise Oversight (OFHEO) house price index data, found at www.ofheo.gov/download.asp. The yearly changes are determined by calculating the changes between the first quarters of each year. OFHEO price indices are available for only 135 MSAs, so we restrict our sample to these 135 MSAs for all our regressions (population, income, and housing value).

The individual-level income and housing value data, along with all our control variables (age, race, and education for the income regressions and age of home, kitchen details, and number of rooms for the housing regressions), are from the Integrated Public Use Microdata Series (IPUMS) at http:l/usa.ipums.org/usal. The full citation for these data is:

Ruggles, Steven, Matthew Sobek, Trent Alexander, Catherine A. Fitch, Ronald Goeken, Patricia Kelly Hall, Miriam King, and Chad Ronnander. 2004. Integrated public use microdata series: Version 3.0 [Machine-readable database]. Minneapolis, MN: Minnesota Population Center [producer and distributor].

The income variable is INCWAGE, or wage and salary income. The housing value variable is VALUEH, or value of the housing unit.

We use the 1% samples for 1970, 1980, 1990, and 2000.

For our income regressions found in Table 5, we use a sample of full-time, full-year workers, which consists of males 25-55 years old who work 40 or more weeks a year and 35 or more hours per week. We remove any individuals earning less than half the minimum wage for a full-year worker (that is, individuals with a yearly salary less than 0.5 minimum wage * 1,400 working hours per year). We adjust the top-coded earnings by multiplying them by 1.5.

For Table 5, regression 1, wages are adjusted by deflating by the Consumer Price Index at: http://www.bls.gov/cpi/ home.htm. Regression 1 contains data for the 86 MSAs, which are represented in all four decades in the IPUMS data.

For Table 5, regressions 2 and 3, wages are adjusted by deflating by the ACCRA Cost of Living Index at: http:// www.coli.org/. Regression 2 contains data for the 22 MSAs that are available for all four decades in both the IPUMS data and the ACCRA data. Regression 3 contains data for the 68 MSAs that are available for 1990 and 2000 in both the IPUMS data and the ACCRA data.

For our housing regressions found in Table 6, we removed any homes without complete indoor plumbing, homes that were built before 1940, homes that had only one room, and homes that had nine or more rooms. Home values were deflated by the Consumer Price Index.

Sample contains data for the 86 MSAs, which are represented in all four decades in the IPUMS data.

The county-level data set with population, income, and housing value variables was put together using data from the U.S. Census County Data Books found at ICPSR 2896. This data set is used for the housing value and income ratios found in Figure 1; Figures 2-4; and the correlations in Table 1. The full citation for the data is:

Haines, Michael R. 2005. Historical, demographic, economic, and social data: The United States, 1790-2000. Inter-university Consortium for Political and Social Research. hd1:1902.2/02896 http://id.thedata.org/ hdl%3A1902.2%2F02896.

Income is median family income. Housing value is median value of specified owner-occupied housing units.

Temperature data are from the National Climatic Data Center at: http://cdc.noaa.gov/CDO/cdo.

The data we used in our housing permit regressions, found in Table 10, are from a variety of sources. The sample size for this data set is the 912 counties that have populations over 50,000.

The permit data for 2000-2005 are from Residential Construction Data files, from the U.S. Census Bureau, at: http://www.census.gov/const/www/permitsindex.html.

The county-level income and housing value data for 2000 are from the U.S. Census Bureau at: http:// factfinder.census.gov/.

The county-level housing density and land area data for 2000 are from the U.S. Census Bureau at: http:// www.census.gov/population/www/censusdata/density.html.

Temperature data are from the National Climatic Data Center at: http://cdo.ncdc.noaa.govlCDO/cdo.

The number of local governments is from the Census of Governments, Preliminary Report No. 1 The 2002 Census of Governments, Table 16, at: http://www.census.gov/govs/ wwwlcog2002.html.

We use the state-level population data for Figure 1. Our historical data (1950-1990) on state-level population come from the U.S. Census Bureau at: http://www.census.gov/ population/censusdatalurpopOO90.txt.

Our 2000 data on state-level population come from the U.S. Census Bureau at: http://factfinder.census.gov/ servlet/GCTTable?_bm=y&-geo_id=01000US&-_box_head_nbr=GCT-PHl&-ds_name =DEC_2000_SF1_U&-redoLog=false&-format=US-9&-mt_name=PEP_2006_EST_GCTT 1R_US9S.

References

Barro, Robert, and Xavier Sala-I-Martin. 1992. Convergence. Journal of Political Economy 100:223 51.

Bellante, Don. 1979. The North-South differential and the migration of heterogeneous labor. The American Economic Review 69:166-75.

Besley, Timothy, Torsten Persson, and Daniel Sturm. 2005. Political competition and economic performance: Theory and evidence from the United States. NBER Working Paper No. 11484.

Borts, George, and Jerome Stein. 1964. Economic growth in a free market. New York: Columbia University Press.

Card, David, and Alan B. Krueger. 1992. School quality and black-white relative earnings: A direct assessment. The Quarterly Journal of Economics 107:151-200.

Caselli, Francesco, and Wilbur Coleman. 2001. The U.S. structural transformation and regional convergence: A reinterpretation. Journal of Political Economy 109:584-616.

Chinitz, Benjamin. 1986. The regional transformation of the American economy. American Economic Review 76:300-3.

Cobb, James C. 1982. The selling of the South. Baton Rouge: Louisiana State University Press.

Donohue, John J. III, and James Heckman. 1991. Continuous versus episodic change: The impact of civil rights policy on the economic status of blacks. Journal of Economic Literature 29:1603-43.

DuMond, Michael J., Barry T. Hirsch, and David A. Macpherson. 1999. Wage differentials across labor markets and workers: Does cost of living matter? Economic Inquiry 37:577-98.

Garrett, Martin. 1968. Growth in manufacturing in the South, 1947-1958: A study in regional industrial development. Southern Economic Journal 34:352-64.

Glaeser, Edward, Joseph Gyourko, and Raven Saks. 2006. Urban growth and housing supply. Journal of Economic Geography 6:71-89.

Glaeser, Edward, and Janet Kohlhase. 2004. Cities, regions and the decline in transport costs. Papers in Regional Science 83:197-228.

Glaeser, Edward, Jose Scheinkman, and Andrei Shleifer. 1995. Economic growth in a cross-section of cities. Journal of Monetary Economics 36:117-43.

Graves, Philip. 1980. Migration and climate. Journal of Regional Science 20:227-38.

Holmes, Thomas. 1997. The effect of state policies on the location of manufacturing: Evidence from state borders. Journal of Political Economy 106:667 705.

Hulten, Charles, and Robert Schwab. 1984. Regional productivity growth in U.S. manufacturing: 1951-1978. American Economic Review 74:152-62.

McDonald, Stephen. 1961. On the South's recent economic development. Southern Economic Journal 28:30-40.

Mueser, Peter, and Philip Graves. 1995. Examining the role of economic opportunity and amenities in explaining population redistribution. Journal of Urban Economics 37:176-200.

Newman, Robert J. 1982. Dynamic patterns in regional wage differentials. Southern Economic Journal 49:246-54.

Olson, Mancur. 1983. The South will fall again: The South as leader and laggard in economic growth. Southern Economic Journal 49:917-32.

Roback, Jennifer. 1982. Wages, rents, and the quality of life. Journal of Political Economy 90:1257-78.

Rosen, Sherwin. 1979. Wage-based indexes of urban quality of life. In Current issues in urban economics, edited by Peter Mieszkowski and Mahlon Straszheim. Baltimore: Johns Hopkins University Press, pp. 74-104.

Wright, Gavin. 1987. The economic revolution in the American South. Journal of Economic Perspectives 1:161 78.

(1) This claim is certainly true in our model and in most simple models of spatial equilibrium. However, if human capital were heterogeneous or if firms were attracted by consumption amenities, then nominal incomes might not fall with rising amenity levels.

(2) The long and distinguished literature on the economic development of the post-war South includes McDonald (1961), Botts and Stein (1964), Garrett (1968), Newman (1982), Hulten and Schwab (1984), Chinitz (1986), and Wright (1987).

(3) Southern cities, since they are newer, may also have an advantage at adapting to the automobile.

(4) Our use of the term "real wage" is in line with much of the spatial equilibrium literature, but it is different from DuMond, Hirsch, and Macpherson (1999), who include amenities in their definition of real wages. Using their definition, real wages would be utility levels in our model, and the spatial equilibrium framework implies that these will be equal across space.

Edward L. Glaeser * and Kristina Tobio [(dagger)]

* Kennedy School of Government, Harvard University and NBER, Taubman-344, 79 JFK Street, Cambridge, MA 02138-5801, USA.

([dagger]) Kennedy School of Government, Harvard University, Taubman-348, 79 JFK Street, Cambridge, MA 02138-5801, USA; E-mail kristina_tobio@ksg.harvard.edu; corresponding author.

This paper was written as an invited association lecture for the Southern Economics Association. The Taubman Center for State and Local Government provided generous financial support.

Edward Glaeser is the Fred and Eleanor Glimp Professor of Economics in the Faculty of Arts and Sciences at Harvard University, where he has taught since 1992. He is Director of the Taubman Center for State and Local Government and Director of the Rappaport Institute of Greater Boston. He teaches urban and social economics and microeconomic theory. He has punished dozens of papers on cities, economic growth, and law and economics. In particular, his work has focused on the determinants of city growth and the role of cities as centers of idea transmission. He also edits the Quarterly Journal of Economics. He received his Ph.D. from the University of Chicago in 1992. This was the Association Lecture delivered at the Southern Economic Association meetings, Charleston, South Carolina, on November 11, 2006.

Table 1. Correlation of Change in County-Level Population, Income,
and Housing Values with South Dummy, January Temperature,
and July Temperature

 South Dummy

 Change in Change Change in
 Population in Income Housing
 (%) (%) Value (%)

1950s -3 32 -1
1960s 4 38 22
1970s 14 33 8
1980s 19 11 25
1990s 19 0 25

 January Temperature
 (Warm Winters)

 Change in Change in Change in
 Population Income Housing
 (%) (%) Value (%)

1950s 11 38 10
1960s 18 21 34
1970s 28 31 13
1980s 33 5 33
1990s 29 -9 32

 July Temperature
 (Hot Summers)

 Change in Change in Change in
 Population Income Housing
 (%) (%) Value (%)

1950s -5 22 -6
1960s -1 30 19
1970s -1 37 -19
1980s 5 -2 4
1990s 4 4 5

Sample is all counties. Population, income, and housing value data
from historical U.S. Census County Data Books, found in Haines,
Michael R.; Inter-university Consortium for Political and Social
Research, 2005-02-25, "Historical, Demographic, Economic, and
Social Data: The United States, 1790-2000," hdl:1902.2/02896 http://
id.thedata.org/hdl%3A1902.2%2F02896 Inter-university Consortium
for Political and Social Research [distributor(DDI)]. Temperature
data from National Climatic Data Center, U.S. Department of
Commerce, at http://cdo.ncdc.noaa.gov/.

Table 2. Population Growth in the Sunbelt

 Log Change in Population

 (1) (2)

 1950s 1960s

With South dummy
 Intercept 0.288 (0.013) 0.198 (0.013)
 South dummy 0.041 (0.029) 0.017 (0.029)
 [R.sup.2] 0.006 0.003
 Observations 135 135
With mean January temperature
(indicating warm winters)
 Intercept -0.005 (0.052) 0.039 (0.035)
 Mean January temperature 0.843 (0.137) 0.455 (0.094)
 [R.sup.2] 0.221 0.150
 Observations 135 135
With mean July temperature
(indicating hot summers)
 Intercept -0.081 (0.223) 0.229 (0.148)
 Mean July temperature 0.502 (0.296) -0.036 (0.196)
 [R.sup.2] 0.021 0.000
 Observations 135 135
With South dummy and mean January
and July temperatures
 Intercept -0.454 (0.231) 0.094 (0.161)
 South dummy -0.157 (0.051) -0.058 (0.036)
 Mean January temperature 1.05 (0.155) 0.574 (0.108)
 Mean July temperature 0.542 (0.304) -0.114 (0.212)
 [R.sup.2] 0.274 0.181
 Observations 135 135

 Log Change in Population

 (3) (4)

 1970s 1980s

With South dummy
 Intercept 0.141 (0.014) 0.115 (0.012)
 South dummy 0.086 (0.03) 0.059 (0.025)
 [R.sup.2] 0.057 0.039
 Observations 135 135
With mean January temperature
(indicating warm winters)
 Intercept -0.104 (0.033) -0.103 (0.027)
 Mean January temperature 0.731 (0.088) 0.644 (0.071)
 [R.sup.2] 0.341 0.384
 Observations 135 135
With mean July temperature
(indicating hot summers)
 Intercept -0.132 (0.155) -0.160 (0.128)
 Mean July temperature 0.386 (0.206) 0.383 (0.17)
 [R.sup.2] 0.026 0.037
 Observations 135 135
With South dummy and mean January
and July temperatures
 Intercept -0.222 (0.153) -0.324 (0.12)
 South dummy -0.042 (0.034) -0.071 (0.027)
 Mean January temperature 0.786 (0.103) 0.733 (0.081)
 Mean July temperature 0.143 (0.201) 0.270 (0.158)
 [R.sup.2] 0.349 0.417
 Observations 135 135

 Log Change in Population

 (5)

 1990s

With South dummy
 Intercept 0.126 (0.009)
 South dummy 0.065 (0.02)
 [R.sup.2] 0.074
 Observations 135
With mean January temperature
(indicating warm winters)
 Intercept 0.028 (0.025)
 Mean January temperature 0.312 (0.066)
 [R.sup.2] 0.142
 Observations 135
With mean July temperature
(indicating hot summers)
 Intercept -0.224 (0.099)
 Mean July temperature 0.484 (0.131)
 [R.sup.2] 0.093
 Observations 135
With South dummy and mean January
and July temperatures
 Intercept -0.23 (0.112)
 South dummy -0.004 (0.025)
 Mean January temperature 0.27 (0.076)
 Mean July temperature 0.365 (0.148)
 [R.sup.2] 0.188
 Observations 135

Population data are MSA-level data from the U.S. Census. Sample is
restricted to the 135 MSAs for which there are OFHEO data available.
OFHEO data are from http://www.ofheo.gov/download.asp. Temperature
units are in hundreds of degrees.

Table 3. Income Growth in the Sunbelt

 Log Change in Income

 (1) (2)

 1950s 1960s

With South dummy
 Intercept 0.359 (0.006) 0.280 (0.005)
 South dummy 0.029 (0.013) 0.078 (0.011)
 [R.sup.2] 0.039 0.268
 Observations 135 135
With mean January temperature
(indicating warm winters)
 Intercept 0.331 (0.017) 0.300 (0.017)
 Mean January temperature 0.095 (0.044) -0.012 (0.045)
 [R.sup.2] 0.034 0.001
 Observations 135 135
With mean July temperature
(indicating hot summers)
 Intercept 0.207 (0.063) 0.154 (0.064)
 Mean July temperature 0.211 (0.084) 0.189 (0.085)
 [R.sup.2] 0.045 0.035
 Observations 135 135
With South dummy and mean January
and July temperatures
 Intercept 0.228 (0.075) 0.445 (0.061)
 South dummy 0.008 (0.017) 0.119 (0.014)
 Mean January temperature 0.060 (0.051) -0.204 (0.041)
 Mean July temperature 0.152 (0.099) -0.134 (0.081)
 [R.sup.2] 0.065 0.393
 Observations 135 135

 Log Change in Income

 (3) (4)
 1970s 1980s

With South dummy
 Intercept -0.181 (0.005) 0.282 (0.01)
 South dummy 0.064 (0.011) -0.001 (0.022)
 [R.sup.2] 0.193 0.000
 Observations 135 135
With mean January temperature
(indicating warm winters)
 Intercept -0.207 (0.016) 0.232 (0.029)
 Mean January temperature 0.109 (0.043) 0.138 (0.076)
 [R.sup.2] 0.047 0.024
 Observations 135 135
With mean July temperature
(indicating hot summers)
 Intercept -0.375 (0.061) 0.507 (0.11)
 Mean July temperature 0.276 (0.081) -0.301 (0.146)
 [R.sup.2] 0.081 0.031
 Observations 135 135
With South dummy and mean January
and July temperatures
 Intercept -0.228 (0.068) 0.521 (0.129)
 South dummy 0.059 (0.015) 0.004 (0.029)
 Mean January temperature -0.006 (0.046) 0.186 (0.087)
 Mean July temperature 0.067 (0.09) -0.408 (0.17)
 [R.sup.2] 0.197 0.075
 Observations 135 135

 Log Change in Income

 (5)
 1990s

With South dummy
 Intercept 0.076 (0.005)
 South dummy 0.018 (0.011)
 [R.sup.2] 0.020
 Observations 135
With mean January temperature
(indicating warm winters)
 Intercept 0.119 (0.014)
 Mean January temperature -0.110 (0.037)
 [R.sup.2] 0.061
 Observations 135
With mean July temperature
(indicating hot summers)
 Intercept 0.098 (0.056)
 Mean July temperature -0.024 (0.074)
 [R.sup.2] 0.001
 Observations 135
With South dummy and mean January
and July temperatures
 Intercept 0.237 (0.061)
 South dummy 0.057 (0.014)
 Mean January temperature -0.194 (0.041)
 Mean July temperature -0.132 (0.08)
 [R.sup.2] 0.176
 Observations 135

Income data are MSA-level data from the U.S. Census. Sample is
restricted to the 135 MSAs for which there are OFHEO data available.
OFHEO data from http://www.ofheo.gov/download.asp. Temperature
units are in hundreds of degrees.

Table 4. Housing Value Growth in the Sunbelt

 Log Change in Housing Value

 (1) (2)
 1950s 1960s

With South dummy
 Intercept 0.238 (0.010) 0.099 (0.010)
 South dummy 0.040 (0.021) 0.006 (0.022)
 [R.sup.2] 0.027 0.001
 Observations 135 135
With mean January temperature
(indicating warm winters)
 Intercept 0.190 (0.027) 0.063 (0.029)
 Mean January temperature 0.157 (0.073) 0.103 (0.077)
 [R.sup.2] 0.034 0.013
 Observations 135 135
With mean July temperature
(indicating hot summers)
 Intercept 0.178 (0.107) 0.431 (0.108)
 Mean July temperature 0.092 (0.142) -0.440 (0.144)
 [R.sup.2] 0.003 0.066
 Observations 135 135
With South dummy and mean January
and July temperatures
 Intercept 0.250 (0.126) 0.531 (0.127)
 South dummy 0.028 (0.028) 0.040 (0.028)
 Mean January temperature 0.116 (0.085) 0.119 (0.085)
 Mean July temperature -0.067 (0.166) -0.641 (0.167)
 [R.sup.2] 0.042 0.115
 Observations 135 135

 Log Change in Housing Value

 (3) (4)
 1970s 1980s

With South dummy
 Intercept 0.347 (0.018) 0.522 (0.027)
 South dummy -0.035 (0.038) -0.172 (0.060)
 [R.sup.2] 0.006 0.058
 Observations 135 135
With mean January temperature
(indicating warm winters)
 Intercept 0.043 (0.043) 0.417 (0.081)
 Mean January temperature 0.831 (0.113) 0.196 (0.215)
 [R.sup.2] 0.290 0.006
 Observations 135 135
With mean July temperature
(indicating hot summers)
 Intercept 0.743 (0.191) 1.810 (0.291)
 Mean July temperature -0.536 (0.253) -1.760 (0.386)
 [R.sup.2] 0.033 0.135
 Observations 135 135
With South dummy and mean January
and July temperatures
 Intercept 0.340 (0.168) 1.465 (0.338)
 South dummy -0.170 (0.037) -0.143 (0.075)
 Mean January temperature 1.205 (0.113) 0.663 (0.228)
 Mean July temperature -0.527 (0.221) -1.578 (0.445)
 [R.sup.2] 0.482 0.190
 Observations 135 135

 Log Change in Housing Value

 (5)
 1990s

With South dummy
 Intercept 0.333 (0.017)
 South dummy 0.006 (0.037)
 [R.sup.2] 0.000
 Observations 135
With mean January temperature
(indicating warm winters)
 Intercept 0.478 (0.047)
 Mean January temperature -0.400 (0.124)
 [R.sup.2] 0.072
 Observations 135
With mean July temperature
(indicating hot summers)
 Intercept 0.573 (0.186)
 Mean July temperature -0.318 (0.247)
 [R.sup.2] 0.012
 Observations 135
With South dummy and mean January
and July temperatures
 Intercept 0.883 (0.21)
 South dummy 0.129 (0.047)
 Mean January temperature -0.562 (0.412)
 Mean July temperature -0.498 (0.277)
 [R.sup.2] 0.125
 Observations 135

Housing value data are MSA-level data from the U.S. Census and
OFHEO. Sample is restricted to the 135 MSAs for which there are
OFHEO data available. The 1980s and 1990s regressions units are
in use these OFHEO data, while the 1950s, 1960s, and 1970s
regressions use Census data. OFHEO data from
http://www.ofbeo.gov/download.asp. Temperature hundreds of degrees.

Table 5. Income Regressions Using IPUMS Data

 (1) (2)

 Census ACCRA-Deflated
 Wage Income Wage Income

With South dummy
 South -0.107 (0.004) 0.058 (0.007)
 South * 1980 dummy 0.041 (0.005) 0.028 (0.009)
 South * 1990 dummy -0.028 (0.004) 0.020 (0.007)
 South * 2000 dummy 0.031 (0.004) 0.081 (0.006)
 Observations 685,944 277,112
With January temperature (indicating warm winters)
 January -0.188 (0.117) 0.125 (0.056)
 January * 1980 dummy -0.014 (0.160) 0.138 (0.138)
 January * 1990 dummy 0.039 (0.170) -0.377 (0.185)
 January * 2000 dummy -0.026 (0.161) -0.383 (0.317)
 Observations 685,944 277,112
With July temperature (indicating hot summers)
 July -0.685 (0.165) 0.777 (0.254)
 July * 1980 dummy 0.184 (0.25) 0.242 (0.703)
 July * 1990 dummy -0.074 (0.242) 0.598 (0.770)
 July * 2000 dummy 0.043 (0.226) 1.170 (0.785)
 Observations 685,944 277,112

 (3)

 ACCRA-Deflated
 Wage Income

With South dummy
 South 0.032 (0.003)
 South * 1980 dummy
 South * 1990 dummy
 South * 2000 dummy 0.087 (0.004)
 Observations 285,535
With January temperature (indicating warm winters)
 January -0.189 (0.096)
 January * 1980 dummy
 January * 1990 dummy
 January * 2000 dummy -0.183 (0.234)
 Observations 285,535
With July temperature (indicating hot summers)
 July 0.404 (0.234)
 July * 1980 dummy
 July * 1990 dummy
 July * 2000 dummy 0.614 (0.508)
 Observations 285,535

1970-2000 income and control data from IPUMS, at Steven Ruggles,
Matthew Sobek, Trent Alexander, Catherine A. Fitch, Ronald Goeken,
Patricia Kelly Hall, Miriam King, and Chad Ronnander. Integrated
Public Use Microdata Series: Version 3.0 [Machine-readable database].
Minneapolis, MN: Minnesota Population Center [producer and
distributor], 2004. Sample is restricted to males aged 25-55 who
work "full-time, full-year" (35+ hours per week, 40+ weeks per year)
and who earn a yearly income more than half of that of a worker
earning a yearly income at the minimum wage. Data controls are age,
education, and race. MSA sample for regression (1) is the 86 MSAs
that are available for all four decades in the IPUMS data. MSA
sample for (2) is for the 22 MSAs that are available for all four
decades in the ACCRA data. MSA sample for (3) is the 68 MSAs that
are available for 1990 and 2000 in the ACCRA data. Temperature
units are in hundreds of degrees.

Table 6. Housing Value Regressions Using IPUMs Data

 (1)

 Housing Value Variable
With South dummy
 South -0.318 (0.002)
 South * 1980 dummy 0.001 (0.003)
 South * 1990 dummy -0.147 (0.003)
 South * 2000 dummy 0.020 (0.003)
 Observations 2,030,019
With January temperature (indicating warm winters)
 January -0.164 (0.348)
 January * 1980 dummy 0.853 (0.593)
 January * 1990 dummy 0.307 (0.842)
 January * 2000 dummy -0.554 (0.879)
 Observations 2,030,019
With July temperature (indicating hot summers)
 July -2.399 (0.378)
 July * 1980 dummy -0.089 (0.724)
 July * 1990 dummy -1.035 (1.066)
 July * 2000 dummy 0.017 (1.166)
 Observations 2,030,019

1970-2000 income and control data from IPUMs, at Steven Ruggles,
Matthew Sobek, Trent Alexander, Catherine A. Fitch, Ronald Goeken,
Patricia Kelly Hall, Miriam King, and Chad Ronnander. Integrated
Public Use Microdata Series: Version 3.0 [Machine-readable database].
Minneapolis, MN: Minnesota Population Center [producer and
distributor], 2004. Sample is restricted to homes built after 1940,
with complete plumbing, and with between two and eight rooms. Data
controls are presence of a kitchen, number of rooms, and built year.
MSA sample is for the 86 MSAs that are available in IPUMS for all
4 years. Temperature units are in hundreds of degrees.

Table 7. Decomposition: South Dummy Regressions

 Using MSA Coefficients Only

 1950s 1960s 1970s 1980s 1990s

[[lambda].sub.A] 0.032 0.066 0.068 0.011 0.027
(productivity) (0.017) (0.013) (0.010) (0.021) (0.009)

[[lambda].sub.[delta]] -0.017 -0.076 -0.074 -0.050 -0.016
(amenities) (0.013) (0.010) (0.012) (0.014) (0.006)

[[lambda].sub.L] if -0.050 0.077 0.254 0.574 0.065
 [delta] = 1.5 (0.075) (0.066) (0.064) (0.115) (0.070)
 (housing supply)

[[lambda].sub.L] if 0.010 0.086 0.202 0.316 0.074
 [delta] = 3 (0.060) (0.038) (0.030) (0.054) (0.036)
 (housing supply)

 Using MSA and IPUMS Coefficients

 1970s 1980s 1990s

[[lambda].sub.A] 0.050 -0.011 0.038
(productivity) (0.004) (0.003) (0.003)

[[lambda].sub.[delta]] -0.051 -0.024 -0.029
(amenities) (0.005) (0.004) (0.004)

[[lambda].sub.L] if 0.230 0.547 0.078
 [delta] = 1.5 (0.005) (0.004) (0.004)
 (housing supply)

[[lambda].sub.L] if 0.178 0.289 0.087
 [delta] = 3 (0.005) (0.004) (0.004)
 (housing supply)

MSA-level data are from the U.S. Census. IPUMS data from Steven
Ruggles, Matthew Sobek, Trent Alexander, Catherine A. Fitch,
Ronald Goeken, Patricia Kelly Hall, Miriam King, and Chad
Ronnander. Integrated Public Use Microdata Series: Version 3.0
[Machine-readable database]. Minneapolis, MN: Minnesota Population
Center [producer and distributor], 2004. MSA coefficients from
regressions in Tables 2, 3, and 4. IPUMS coefficients from
Table 5 regression (1).

Table 8. Decomposition: January Temperature (Warm Winter) Regressions

 Using MSA Coefficients Only

 1950s 1960s 1970s 1980s 1990s

[[lambda].sub.A] 0.245 0.082 0.233 0.239 -0.025
(productivity) (0.061) (0.050) (0.044) (0.070) (0.039)

[[lambda].sub.[delta]] -0.048 0.042 0.140 -0.079 -0.010
(amenities) (0.043) (0.051) (0.056) (0.045) (0.026)

[[lambda].sub.L] if 0.466 0.134 -1.653 0.195 1.401
[delta] = 1.5 (0.228) (0.233) (0.358) (0.573) (0.284)
(housing supply)

[[lambda].sub.L] if 0.702 0.289 -0.407 0.493 0.802
[delta] = 3 (0.195) (0.150) (0.190) (0.267) (0.136)
(housing supply)

 Using MSA and IPUMS Coefficients

 1970s 1980s 1990s

[[lambda].sub.A] 0.135 0.160 0.041
(productivity) (0.014) (0.007) (0.010)

[[lambda].sub.[delta]] 0.263 0.020 -0.094
(amenities) (0.017) (0.009) (0.012)

[[lambda].sub.L] if -1.776 0.095 1.485
[delta] = 1.5 (0.017) (0.011) (0.012)
(housing supply)

[[lambda].sub.L] if -0.529 0.389 0.210
[delta] = 3 (0.017) (0.010) (0.012)
(housing supply)

MSA-level data are from the U.S. Census. IPUMS data from Steven
Ruggles, Matthew Sobek, Trent Alexander, Catherine A. Fitch,
Ronald Goeken, Patricia Kelly Hall, Miriam King, and Chad
Ronnander. Integrated Public Use Microdata Series: Version 3.0
[Machine-readable database]. Minneapolis, MN: Minnesota Population
Center [producer and distributor], 2004. MSA coefficients from
regressions in Tables 2, 3, and 4. IPUMS coefficients from
Table 5 Regression (1). Temperature units are in hundreds of degrees.

Table 9. Decomposition: July Temperature (Hot Summer) Regressions

 Using MSA Coefficients Only

 1950s 1960s 1970s 1980s 1990s

[[lambda].sub.A] 0.269 0.144 0.298 -0.164 0.078
(productivity) (0.124) (0.088) (0.084) (0.125) (0.070)

[[lambda].sub.[delta]] -0.184 -0.320 -0.437 -0.227 -0.071
(amenities) (0.081) (0.089) (0.102) (0.081) (0.051)

[[lambda].sub.L] if 0.439 1.471 2.269 5.363 1.414
[delta] = 1.5 (0.473) (0.364) (0.646) (0.810) (0.554)
(housing supply)

[[lambda].sub.L] if 0.576 0.812 1.465 2.723 0.937
[delta] = 3 (0.408) (0.250) (0.295) (0.365) (0.289)
(housing supply)

 Using MSA and IPUMS Coefficients

 1970s 1980s 1990s

[[lambda].sub.A] 0.224 0.018 0.131
(productivity) (0.032) (0.027) (0.028)

[[lambda].sub.[delta]] -0.344 -0.454 -0.138
(amenities) (0.040) (0.034) (0.036)

[[lambda].sub.L] if 2.177 5.590 1.481
[delta] = 1.5 (0.040) (0.036) (0.039)
(housing supply)

[[lambda].sub.L] if 1.373 2.950 1.004
[delta] = 3 (0.040) (0.035) (0.037)
(housing supply)

MSA-level data are from the U.S. Census. IPUMS data from Steven
Ruggles, Matthew Sobek, Trent Alexander, Catherine A. Fitch,
Ronald Goeken, Patricia Kelly Hall, Miriam King, and Chad Ronnander.
Integrated Public Use Microdata Series: Version 3.0 [Machine-readable
database]. Minneapolis, MN: Minnesota Population Center [producer
and distributor], 2004. MSA coefficients from regressions in
Tables 2, 3, and 4. IPUMS coefficients from Table 5 regression (1).
Temperature units are in hundreds of degrees.

Table 10. Housing Permit Regressions

 Log of the Ratio of the Sum of Housing
 Permits to Land Area for 2000-2005

 (1) (2) (3) (4)

Constant 7.374 -24.387 7.005 -26.259
 (0.354) (1.598) (0.343) (1.668)
South dummy 0.219 0.720
 (0.088) (0.068)
January temperature 2.497 2.331
 (indicating warm winters) (0.342) (0.256)
July temperature (indicating
 hot summers)
Log of number of governmental 0.062 -0.029
 units (0.038) (0.035)
Log 2000 income 1.493 2.266
 (0.229) (0.239)
Log 2000 housing density 0.762 0.740
 (0.031) (0.033)
Log 2000 housing value 0.563 0.024
 (0.134) (0.132)
Log total land area -0.840 -0.033 -0.899 -0.091
 (0.052) (0.046) (0.050) (0.049)
[R.sup.2] 0.242 0.737 0.279 0.729

 Log of the Ratio of the Sum of Housing
 Permits to Land Area for 2000-2005

 (5) (6) (7) (8)

Constant 4.652 -29.055 6.023 -28.381
 (0.695) (1.650) (0.817) (1.632)
South dummy -0.472 0.457
 (0.126) (0.086)
January temperature 3.316 -0.467
 (indicating warm winters) (0.507) (0.376)
July temperature (indicating 3.617 6.475 1.551 5.314
 hot summers) (0.745) (0.525) (1.020) (0.710)
Log of number of governmental -0.024 0.069
 units (0.032) (0.037)
Log 2000 income 1.243 1.074
 (0.227) (0.258)
Log 2000 housing density 0.714 0.705
 (0.031) (0.031)
Log 2000 housing value 0.861 1.017
 (0.138) (0.161)
Log total land area -0.828 -0.060 -0.943 -0.068
 (0.051) (0.046) (0.053) (0.047)
[R.sup.2] 0.256 0.747 0.290 0.755

Sample is restricted to counties with populations of 50,000 or more.
Number of Governmental Units from the 2002 Census of Governments at
http://www.census.gov/govs/www/cog2002.html. Income, land area, and
housing density data are from the 2000 Census. Temperature data are
from the National Climatic Data Center, U.S. Department of Commerce,
at http://cdo.ncdc.noaa.gov/. Temperature units are in hundreds of
degrees.