Sunbelt Growth and the Knowledge Economy: An Exploratory Approach

JRAP 38(2): 176-188. © 2008 MCRSA. All rights reserved.



Sunbelt Growth and the Knowledge Economy:
An Exploratory Approach


Dennis L. Hoffman and Timothy D. Hogan
Arizona State University - USA



Abstract. Focusing on the narrower concept of a knowledge-economy-based growth strategy, this
paper explores whether a strong link between a college-educated population and a region’s
economic performance was an important ingredient in the growth experience of the Sunbelt
during the 1990s. The issue is addressed through analysis of two different datasets. First, the
education and income characteristics of the people moving to the Sunbelt region are examined
using migration data from the 2000 census. Then we look at the link between the knowledge-
economy metric of the share of college educated adults and economic growth in the Sunbelt in
the 1990s using data for 116 Sunbelt MSAs. The results of our analysis provide little evidence
that a college educated workforce was a major factor promoting economic growth in Sunbelt ci-
ties during that period.



1. Introduction
ment, the challenge facing policy makers is to attract

high value-added economic activity and to create con-
Policy makers in the Sunbelt have recognized that
ditions conducive to high productivity and sustained
the traditional approach to economic development – a
productivity growth.
strategy of offering a lower cost of doing business to
A popular way to characterize this approach to
attract business relocation and using job growth as the
economic development is in terms of the “knowledge
metric of success – must be replaced in a world charac-
economy” or “knowledge-based economy” The con-
terized by globalization and rapid technological
cept of the “knowledge-based economy” recognizes
change. While the fundamental goal for economic de-
the importance of knowledge and technology in eco-
velopment has always been prosperity, the operational
nomic growth (OECD 1996). A knowledge-based eco-
goal has shifted from providing jobs to increasing liv-
nomic growth strategy often is defined broadly to in-
ing standards. Dollar-denominated metrics – most of-
corporate innovation, research and development activ-
ten wages and/or per capita income – have sup-
ities, and non-education aspects of human capital
planted or at least supplemented job growth as the
(Raspe and Van Oort 2006), but much of the discussion
target variables of regional development.
of the knowledge economy has focused on the link
In many cases, the initial change in strategy was to
between economic growth and the stock of human
shift the focus to attracting high-wage jobs – often cha-
capital measured in terms of the college-educated
racterized as “high tech” manufacturing. Over time,
population. Based on national datasets, empirical re-
this approach has been broadened to recognize that
search clearly links economic growth (prosperity) and
the standard of living is fundamentally determined by
college-educated population (see for example, Glaeser
the productivity of the regional economy and that
and Saiz 2003 and Moretti 2004). And one study ac-
high productivity and productivity growth come from
tually asserts “the percentage of adults with a college
producing higher value products and services and by
degree is the single most important driver of economic
increasing efficiency in producing those goods and
growth” (Weissbourd and Berry 2004).
services. Based on this concept of economic develop-

---

Sunbelt Growth

177
Over the past fifty years, the southeastern and
sophisticated empirical analysis identified what were
southwestern regions of the U.S. have experienced
termed “three large trends that determined the recent
much more rapid growth than the rest of the nation.
growth of cities” – human capital, movement to war-
This portion of the U.S. is popularly referred to as the
mer, drier places, and reliance on autos. The regres-
Sunbelt.1 The Sunbelt’s share of the national popula-
sion analysis in the 2003 study, which included a bat-
tion jumped from 28 percent in 1950 to 40 percent by
tery of control variables and regional fixed effects,
2000. At the beginning of the 21st century, the popula-
found a strong link between the share of college-
tion of the Sunbelt nearly equals the combined popula-
educated adults and population growth using primari-
tion of the nation’s traditional Northeast and Midwest
ly census data for a national set of MSAs for the 1970–
“core” regions – 110 million versus 118 million (Lang
2000 period. The results also showed positive effects of
and Rengert 2001). While “mainstream” economic ana-
warm and dry climate measures. Of particular interest
lyses identified lower costs of doing business and a
to the current project were the results of two sets of
less unionized labor force as key factors leading to a
regression models in the study:
shift of economic activity in the U.S. to the south and

west (see for example, Olson 1983, Chinitz 1986, and
1. The analysis included two simple models of popu-
Wright 1987), more recently other researchers have
lation growth as a function of the share of college
argued that the non-economic factor of climate has
graduates – one using the entire MSA sample and
been a significant determinant of this growth pattern
the other using only MSAs that had average Janu-
(see for example, Gallup, Sachs, and Mellinger 1999
ary temperatures of 40+ degrees. The regression
and Glaeser, et. al. 2001). Some analysts have specu-
coefficient and R2 for the “warm climate” sub-
lated that climate has historically been a substitute for
sample were much smaller than for the national
human capital development (Quan and Beck 1987;
sample. Glaeser and Saiz interpreted these results
Glaeser and Saiz 2003).
to mean skills do not matter in warm cities.
Focusing on the narrower concept of a knowledge-
2. Models that included interaction effects between
economy-based growth strategy, this paper explores
education and climate also found a weaker link be-
whether a strong link between a college-educated
tween education and both MSA growth and MSA
population and a region’s economic performance was
wage levels in warmer areas.
an important ingredient in the growth experience of

the Sunbelt during the 1990s. The issue is addressed
Empirical results from an earlier study of regional
through analysis of two different datasets. First, the
growth also tangentially addressed the link between
education and income characteristics of the people
the knowledge economy and economic growth in the
moving to the Sunbelt region are examined using mi-
Sunbelt. Quan and Beck (1987) looked at the link be-
gration data from the 2000 census. Then we look at the
tween education and state economic growth, but their
link between the knowledge-economy metric of the
analysis focused on the relationships between per ca-
share of college educated adults and economic growth
pita income, wages and employment and public ex-
in the Sunbelt in the 1990s using data for 116 Sunbelt
penditures on K-12 and higher education. They found
MSAs.
positive links between education spending and the

economic variables for Northern states, but little evi-
2. A selective review of the literature
dence (with some results actually showing a negative

relationship) of any link for Sunbelt states.
The 2003 Glaeser and Saiz study of the link be-
In neither case do the results offer support for the
tween education and urban growth served as the ini-
hypothesis that human capital was an important factor
tial motivation for the research on which this paper is
in explaining the economic success of the Sunbelt in
based. Their 2003 analysis built upon an earlier study
the late 20th century.
by Glaeser and Shapiro (2001) in which somewhat less


3. Growth and educational attainment
1 The geographic area defined as the Sunbelt for this study is based

on U.S. Census Bureau definition. It is composed of 13 states (North
Some areas of the Sunbelt have achieved great eco-
Carolina, South Carolina, Georgia, Florida, Alabama, Mississippi,
nomic success along with sheer aggregate growth. Me-
Louisiana, Tennessee, Arkansas, Oklahoma, Texas, New Mexico,
tro areas like Atlanta and Charlotte are obvious exam-
and Arizona) plus Clark County, NV (Las Vegas), and a nine-county
region of Southern California (Imperial, Kern, Los Angeles, Orange,
ples. Other areas have experienced explosive popula-
Riverside, San Bernardino, San Diego, Santa Barbara, and Ventura).
tion and job growth but have not done as well in rais-
Some parts of the analysis are based on state-level data. In those
ing the living standards of their residents (at least in
cases the Sunbelt region is defined to include the 13 states plus the
comparison with the national average). For example,
entire states of California and Nevada

---

178
Hoffman and Hogan
McAllen-Edinburg-Mission, TX was the second fastest
growth and also managed 60+ percent increases in per
growing Sunbelt MSA in the 1990s, but income grew
capita GDP. Others like Nevada and South Carolina
slower than the national average so that its per capita
grew rapidly but had below average increases in per
personal income figure fell from 48 percent of the na-
capita GDP. And at the other end of the scale, Louisi-
tional level in 1990 to 45 percent in 2000. And not all
ana lagged behind the Non-Sunbelt region in all four
Sunbelt states have shared in the rapid growth expe-
measures.
rienced by the rest of the region. Four states (Alabama,
Turning to comparisons of the 15-state Sunbelt re-
Arkansas, Mississippi, and Oklahoma) actually trailed
gion vis-à-vis the rest of the nation in terms of the
the national growth trend over the 1950–2000 period.
proportion of college graduates – the human capital
Focusing on the last decade of the 20th century, the
metric often used in studies of the knowledge econo-
aggregate statistics presented in Table 1 show that the
my – the figures demonstrate that the share of those
population of the 15-state Sunbelt region grew more
25+ with at least bachelor’s degree in 2000 was lower
than twice as fast as the Non-Sunbelt states between
for the Sunbelt region than in the Non-Sunbelt states.
1990 and 2000. The Sunbelt states also outpaced the
Looking specifically at younger adults (aged 25 to 39),
rest of the nation in terms of aggregate economic
the relative ranking remains the same, and the gap
growth – for example, the GDP of the Sunbelt region
between the Non-Sunbelt and Sunbelt regions is even
increased 78 percent over the decade compared with
larger than for all adults.
60 percent for the Non-Sunbelt states. But the Sunbelt
The economies of some areas in the Sunbelt have
region was not as successful in terms of personal eco-
clearly benefited from knowledge economy-based
nomic measures. The average per capita GDP in 2000
growth, but the statistics in Table 1 show that much of
for the 15 state Sunbelt region was $33,104–92 percent
the region still lags far behind in developing know-
that of the Non-Sunbelt region, and per capita GDP
ledge-based resources. For example, a recent Milken
growth for the Sunbelt states also lagged behind the
Institute study of Arkansas’ position in the know-
rest of the nation during the decade.
ledge-based economy ranked the state next-to-last in
While the region as a whole surpassed the rest of
knowledge-economy resources, and the analysis cau-
the U.S. in terms of aggregate growth, the pattern of
tioned that several other Sunbelt states were in similar,
growth during the 1990s was not uniform among the
if slightly better situations (Milken Institute 2004).
individual Sunbelt states. Some states like Arizona,

California, and North Carolina had rapid aggregate

Table 1. Growth in Population, Gross Domestic Product, Per Capita GDP & Proportions of College-Educated Adults









1990-2000
1990-2000
Per Capita
1990-2000
Population
Population
Difference

Population
GDP
GDP
Per Capita
25+
25-39
25-39 vs

Growth
Growth
2000
GDP Growth
2000
2000
25+ 2000

(percent)
(percent)
($)
(percent)
(percent)
(percent)
(percent)

Sunbelt States
18.9
77.9
33,104
49.7
23.0
24.1
1.1

Alabama
10.1
61.2
25,764
46.4
19.0
21.6
2.6

Arizona
40.0
128.7
30,899
63.4
23.5
23.5
0.0

Arkansas
13.7
75.3
24,987
54.1
16.7
18.3
1.6

California
13.8
63.3
38,001
43.5
26.6
26.5
-0.1

Florida
23.5
83.2
29,490
48.3
22.3
23.4
1.1

Georgia
26.4
108.5
35,533
65.0
24.3
27.2
2.9

Louisiana
5.9
40.5
29,430
32.7
18.7
20.3
1.6

Mississippi
10.5
65.8
22,592
50.0
16.9
18.1
1.2

Nevada
66.3
131.7
36,892
39.3
18.2
17.6
-0.6

New Mexico
20.1
88.8
27,885
57.2
23.5
20.7
-2.8

North Carolina
21.4
95.1
34,003
60.7
22.5
25.5
3.0

Oklahoma
9.7
55.5
26,012
41.8
20.3
20.9
0.6

South Carolina
15.1
71.2
28,044
48.8
20.4
22.0
1.6

Tennessee
16.7
84.9
30,733
58.5
19.6
22.6
3.0

Texas
22.8
89.3
34,876
54.2
23.2
23.6
0.4
Non-Sunbelt States
9.0
67.7
35,845
53.7
25.5
29.3
3.9
United States
13.2
71.8
34,642
51.8
24.4
27.0
2.6

Source: Computed by authors based on data from U.S. Census Bureau & U.S. Bureau of Economic Analysis websites; 2000 Census IPUMS 5% files.

---

Sunbelt Growth

179

4. U.S. domestic net migration
states for elderly interstate migrants were Sunbelt

states. Nevada, Arizona and Florida received most of
Recent decades have witnessed steady north-to-
these elderly migrants. In many Sunbelt states this
south net domestic migration in the U.S. A Census
population flow of older people serves to boost local
report, Domestic Net Migration in the United States: 2000
economies. In Nevada, as reported in the Economist
to 2004 (Perry 2006) reveals the progression of north to
(2006), the elderly population is serving as an impor-
south migration continued in a pattern that prevailed
tant source of service industry labor. And across the
throughout the 1990s. Eight of the top ten states for
Sunbelt, the migrating elderly are generally more edu-
domestic net migration were Sunbelt states while Cali-
cated and wealthier in comparison with national aver-
fornia (partially Sunbelt) experienced significant net
ages at the same age cohort – resulting in a positive
domestic out-migration.
tug on average educational attainment and income in
Behind the aggregate net migration numbers, ques-
the 25+ population for many Sunbelt states. The chal-
tions remain. What are the knowledge and skill cha-
lenge of course is that the elderly provide little boost
racteristics of the people that dominate these popula-
to the quality of skills in the working age labor force
tion flows? What role does age play in the observed
and may indeed result in increasing pressures for
domestic net migration patterns? Are the knowledge
more service industry jobs. Further, as the Baby Boom
and skills of the people migrating from north to south
generation ages, Sunbelt states will no doubt find
different from the average state-to-state migrant? And
themselves with increasing proportions of very elderly
is this north-to-south migration consistent with the
people – especially in the most attractive states for the
arguments that knowledge and skill development is
elderly migrants, Arizona, Florida and Nevada.
an important catalyst for economic growth and pros-
A special Census 2000 report highlighted the mi-
perity?
gration patterns of the young (25–39), single and col-
Comparing attributes of people who migrated to
lege-educated population for the period 1995 to 2000
the Sunbelt at some point between 1995 and 2000 with
(Franklin 2003b). The analysis revealed that seven of
people who moved from the Sunbelt in the same time
the ten states with the highest rates of net domestic
period (based on analysis of the 2000 Census migra-
migration among this group were Sunbelt states.
tion DVD dataset), it turns out that their characteristics
These findings suggest that Sunbelt states added sig-
are quite similar. The data suggest the educational at-
nificantly to their numbers of young, college-educated
tainment of people migrating north to south are essen-
people. However, the Sunbelt states also received
tially identical to those migrating south to north. For
most of the total net domestic migration, so it may not
example, 34.9 percent of the north-to-south migrants
be surprising that they saw significant increases in the
had college degrees, while 35.5 percent of the south-to-
number of young, single, educated people as well.
north migrants were college graduates. Similarly, 47.9
Seventeen states, plus the District of Columbia, had
percent of the north-to-south migrants reported in-
net positive domestic in-migration for the young, sin-
comes in excess of $50,000, while 49.5 percent of the
gle and college educated population during the 1995–
south-to-north migrants reported incomes in excess of
2000 period. Eight of the 15 Sunbelt states were in-
$50,000. So, in the aggregate, it appears education and
cluded in the group. The other 33 states, including
income-earning skill characteristics are essentially
seven Sunbelt states and 26 non-Sunbelt states, saw
randomly distributed across domestic migrants re-
outflows of the young, single and college educated in
gardless of their (north vs. south) direction of move-
the late 1990s.
ment.
The figures in the Census report reveal the Sunbelt
But this aggregate analysis ignores age differences
generally performed well in terms of net migration for
that may prevail among the domestic migrants, and it
this subpopulation, but they also show the absolute
assumes the Sunbelt is comprised of essentially a ho-
numbers of this group are small relative to total net
mogeneous set of states. We consider each in turn.
migrants. For example, only 6,788 of the 233,934 net

in-migrants to Nevada were in this young, single, and
4.1 Age issues
college-educated demographic group in Arizona 9,264

out of 316,148, and in Florida only 10,454 out of
As is widely known, significant migration of
607,023. In contrast, California, Alaska, Maryland,
people aged 55+ from north to south has been a con-
Illinois, and the District of Columbia, increased their
tinuing phenomenon. The Census 2000 report, Internal
shares of young, single and college educated while
Migration of the Older population: 1995 to 2000 (He and
actually having a net outflow of domestic migrants
Schachter 2003) shows seven of the top 10 destination
from 1995 to 2000.

---

180
Hoffman and Hogan
While this special Census report chose to focus on
educated individuals from the rest of the nation over
young, unmarried college graduates, it makes more
the 1995–2000 period. And young, college-educated
sense to take a somewhat broader look at all young
persons made up 13.9 percent of total net migration
college graduates as the subgroup most important for
into the region – substantially higher than the 5.8 per-
what is happening with an area’s human capital re-
cent share this subpopulation made up of the 5+ popu-
sources. The numbers presented in Table 1 already
lation of the region in 2000. Still, even after adding
demonstrated this desirable subpopulation is a smaller
gaining these “economically desirable” new residents
share of all young adults in the 15-state Sunbelt region
at the expense of the Non-Sunbelt states, the propor-
than in the Non-Sunbelt states. The net migration fig-
tion of college-educated young adults in the Sunbelt
ures in Table 2 show that the Sunbelt states as a region
region remained substantially below the rest of the
gained a total of more than 200,000 young college-
nation in 2000 (see Table 1).


Table 2. Domestic Migration of Young, College-Educated (YCE) Persons 1995 - 2000










YCE share




Percent
Total 5+



Net
Total 5+
Population

In-migrants
Out-migrants
Migration
Net Migration
in 2000

Sunbelt States

Alabama
35,512
47,552
-12,040
-
4.9

Arizona
84,306
59,235
25,071
7.9
5.6

Arkansas
19,485
21,380
-1,895
-
4.0

California
320,594
239,188
81,406
-
6.6

Florida
170,187
145,864
24,323
4.1
5.1

Georgia
151,572
101,960
49,612
14.3
7.1

Louisiana
30,400
51,950
-21,550
-
4.5

Mississippi
19,781
27,911
-8,130
-
4.1

Nevada
31,255
16,307
14,948
6.4
4.4

New Mexico
22,348
28,661
-6,313
21.9
4.5

North Carolina
127,276
97,213
30,063
8.7
6.3

Oklahoma
24,993
39,479
-14,486
-
4.4

South Carolina
49,855
48,229
1,626
1.3
5.1

Tennessee
69,399
63,410
5,989
4.2
5.3

Texas
204,228
164,438
39,790
29.9
5.9
All Sunbelt States
1,361,189
1,152,779
208,410
13.9
5.8

All Non-Sunbelt States
2,017,952
2,226,362
-208,410
-
6.8
United States
3,379,141
3,379,141
0
-
6.3

Sources: U.S. Census Bureau, 2000 Census IPUMS 5 percent files and Census 2000 Special Tabulations PHC-T-22, Gross and Net Migration Tables.


Six of the 15 Sunbelt states actually had net out-
4.2 Knowledge and skills across the Sunbelt
flows of young, college-educated individuals, and for

states like Florida, Nevada, and Arizona these poten-
While considerable north-to-south migration has
tial knowledge economy workers made up a very
occurred, it is clear the migration patterns to and from
small part of their population gains. In a few Sunbelt
individual Sunbelt states vary considerably. Table 3
states, however, the young college-educated in-
depicts the distribution of domestic in-migrants into
migrants were a major positive factor. For Georgia
individual Sunbelt states by income and educational
and Texas in particular they constituted a large share
attainment. The figures reveal considerable hetero-
of the states’ total net-migration. And California had a
geneity across the Sunbelt. The states with the highest
net gain of more than 80,000 young, college-educated
income in-migrants are California, Georgia, and Texas;
individuals at the same time that the state’s overall net
the share of high income migrants exceeds the average
outflow was almost 800,000 over the 1995–2000 period.
state by 21.1 percent for California and Georgia and by

15.3 percent for Texas. All three of these states also

---

Sunbelt Growth

181
reported shares of college educated among in-
$50,000 level that lagged the average Sunbelt state by
migrants over 20 percent above the average Sunbelt
21.7 percent, 13.9 percent and 19.5 percent respective-
state. The states with the lowest income in-migrants
ly. In each of the lowest income states the proportion
are Arkansas, Mississippi and Oklahoma with the
of in-migrants with college degrees was more than 20
proportions of in-migrants with income above the
percent below the average Sunbelt state.


Table 3. Income and Educational Attainment of Domestic In-Migrants to the Sunbelt States: 1995-2000




Income over
Deviation
Income over
Deviation
Migrant
Deviation
Migrant
Deviation

$50,000
from
$50,000
from
reports some
from
has
from

Non- Hispanic
Sunbelt
Total
Sunbelt
college
Sunbelt
college
Sunbelt

Migrants
average
Migrants
average
education
average
degree
average
State
(percent)
(percent)
(percent)
(percent)
(percent)
(percent)
(percent)
(percent)














AL
42.5
-7.8
42.0
-5.8
31.3
-1.4
29.9
-4.9
AZ
49.8
8.0
47.7
7.0
34.2
7.8
31.3
-0.5
AK
5.9
-22.1
34.9
-21.7
30.2
-4.9
22.7
-27.8
CA
55.9
21.3
54.0
21.1
29.8
-6.1
44.8
42.5
FL
46.5
0.9
45.2
1.4
31.1
-2.0
29.6
-5.9
GA
54.8
18.9
54.0
21.1
30.6
-3.6
38.0
20.8
LA
40.2
-12.8
39.8
-10.7
31.3
-1.4
31.0
-1.4
MS
38.6
-16.3
38.4
-13.9
32.9
3.7
24.5
-22.1
NV
52.4
13.7
49.1
10.1
34.1
7.4
21.5
-31.6
NM
42.7
-7.4
38.5
-13.7
32.4
2.1
35.0
11.3
NC
49.8
8.0
48.4
8.5
30.4
-4.2
37.4
18.9
OK
36.8
-20.2
35.9
-19.5
34.0
7.1
24.4
-22.4
SC
46.0
-0.2
45.4
1.8
32.2
1.4
32.3
2.7
TN
44.6
-3.3
44.2
-0.9
30.6
-3.6
31.4
-0.1
TX
55.0
19.3
51.4
15.3
31.0
-2.3
37.9
20.5

Average
46.1

44.6

31.7

31.4

Source: US Census, 2000 Census Migration DVD



Table 3 also reveals the income/skill correlation is
of the states – California and Texas – actually had pro-
not monotonic across the Sunbelt states. Arizona, Flor-
portions above the Non-Sunbelt average.
ida, and Nevada, the preferred destinations of many
This examination of domestic migration data re-
elderly in-migrants, all reported above average income
veals that, while the Sunbelt has been the beneficiary
frequencies with below-average college graduation
of significant net domestic migration, these flows
frequencies overall, with Nevada the lowest frequency
(with the exception of California, and to some degree
of college graduates among the Sunbelt states – yet
Georgia and Texas) are not being accompanied by
reporting high-income frequency that was 10 percent
large numbers of people prepared to contribute to
above the average Sunbelt state. Similarly, New Mex-
knowledge economy endeavors.
ico had a high frequency of college educated in-

migrants but a below average frequency of high-
5. The “knowledge economy” explanation
income in-migrants.

Data for the in-migrant streams of young, college-
This section summarizes an exploratory analysis of
educated individuals demonstrate similar patterns
alternative metrics of economic success and of the
(Table 4). For the entire 15-state Sunbelt region, the
knowledge economy for the 116 MSAs in the Sunbelt
proportion of high-income ($50,000+) individuals was
that are included in the Department of Housing and
only slightly below that for the Non-Sunbelt region,
but this aggregate measure is misleading as only two

---

182
Hoffman and Hogan
Urban Development’s State of the Cities Data System.2
ment growth. Population growth is included to mirror
The metro areas included in the dataset are very di-
Glaeser and Saiz. Table 5 lists the simple correlation
verse – in terms of whatever characteristic one might
coefficients between each of the four metrics. These
choose – ranging from huge (Los Angeles-Long Beach,
figures show a very high correlation between popula-
CA and Houston, TX) to small (Pine Bluff, AR and
tion growth and employment growth. The coefficients
Enid, OK); rapidly growing (Las Vegas, NV) to declin-
also indicate positive correlations between per capita
ing (Alexandria, LA); rich (West Palm Beach-Boca Ra-
income and all three growth measures, but a negative
ton, FL) to poor (McAllen-Edinburg-Mission, TX), etc.
relationship between income growth and population
The analysis looked at four alternative measures of
growth; although none are statistically significant at a
economic success:
.05 confidence level. The correlation statistics indicate

that income level, income growth, and employment
1. The growth rate of the MSA population over the 1990-
growth were not highly correlated among the set of
2000 period – The population growth measure was
116 Sunbelt MSAs.
included to be consistent with the Glaeser and Saiz
The three alternative education-related variables
analysis. Glaeser and Saiz could argue either (a)
used in the analysis are (1) the share of college gra-
they focused on population rather than economic
duates in the adult (25+) population in 2000, (2) the
variables because they were looking at “urban
growth rate of that share over the 1990–2000 period,
growth” not economic growth, or (b) their analysis
and (3) the ratio of net in-migrants to the MSA ages 25
was in the context of looking at growing areas ver-
to 39 in 2000 who were single and college graduates to
sus declining areas. But population growth is not
the population of the MSA ages 5+ in 2000 (net migra-
really a good measure of economic growth – par-
tion of young, single, college educated persons hereaf-
ticularly for Sunbelt regions. In the entire sample of
ter referred to as YSCMR).3 This statistic was com-
116, only three MSAs suffered a decline in total
puted from data produced for the special Census 2000
population, and many of the poorest areas have
report (Franklin 2003b) discussed earlier. Table 5 also
grown rapidly.
presents the simple correlation coefficients between
2. The growth rate of total employment for the 1990–2000
the three education metrics, and also between each of
period – Job growth has more validity than popula-
the three and the four economic measures. The corre-
tion growth as a measure of economic growth. His-
lation coefficients imply that the share of college gra-
torically, it has been one of the primary metrics
duates and growth of that share are not closely re-
used to measure regional economic growth by poli-
lated, but somewhat surprisingly (at first glance any-
cy makers, economic development professionals,
way) that there is a statistically significant negative
and economists. However, it does not do an ade-
relationship between share of college graduates and
quate job of monitoring what is happening to the
the YSCMR. Upon closer examination of the data, it
standard of living of area residents.
appears that this is a result of out-migration of young,
3. Per capita personal income in 2000 – For this analysis
single college-educated adults from “college towns”
per capita personal income was chosen as the
like Austin. On the other hand, the figures show a
proxy measure for the material standard of living
modest positive (statistically significant at the .05 per-
of area residents.
cent level of confidence) correlation between the
4. The growth rate of per capita personal income for the
YSCMR and growth in the share of college graduates.
1990–2000 period – The growth rate of per capita

personal income serves as a metric for the change
5.1. Correlating education and economic growth
in the material standard of living.


The correlation coefficients between the three edu-
The following discussion focuses on the two in-
cation metrics and the four economic measures pro-
come-related measures and secondarily on employ-
vide mixed signals. All of the coefficients are positive,
but not all are statistically significant. The highest cor-

relation is found between the share of college gra-
2 The 116 MSAs in the dataset include two multi-state MSAs for
which some portion of the area lies outside the formal Census defi-
duates and per capita income. This could be inter-
nition of the Sunbelt region. In some cases, HUD’s State of the Cities
preted simply as the result of a higher proportion of
database included two or more PMSAs that are part of a single
college graduates with higher incomes or more broad-
CMSA as separate observations. The authors have chosen not to
ly in terms of a knowledge economy-based argument –
include a complete list of the 116 MSAs but would be happy to pro-
vide one on request. All of the data in the dataset were compiled
from the HUD database, with the exception of the per capita per-

sonal income data compiled from the BEA REIS CD and the climate
3 Data for this variable was not available in the Census report for
measures from the U.S. Census Bureau’s City-County Databook.
seven MSAs in the dataset.

---

Sunbelt Growth

183

Table 4. Young, College-Educated (YCE) Domestic In-Migrants 1995 - 2000










Percent with
Deviation
Deviation


Percent of
Incomes
from
from

YCE
Total 5+
of $50,000
National
Sunbelt

In-migrants
In-Migration
or more
Average (%)
Average (%)

Sunbelt States







Alabama
35,512
10.9
24.3
-18.8
-17.9

Arizona
84,306
10.6
27.8
-7.2
-6.2

Arkansas
19,485
7.8
26.1
-12.9
-11.9

California
320,594
21.6
36.7
22.4
23.7

Florida
170,187
9.1
25.8
-13.8
-12.9

Georgia
151,572
15.6
30.1
0.5
1.6

Louisiana
30,400
11.8
24.3
-19.0
-18.2

Mississippi
19,781
8.8
23.0
-23.2
-22.4

Nevada
31,255
6.8
23.6
-21.1
-20.3

New Mexico
22,348
11.0
21.1
-29.5
-28.7

North Carolina
127,276
13.7
26.3
-12.3
-11.3

Oklahoma
24,993
8.0
21.9
-27.1
-26.3

South Carolina
49,855
11.3
23.8
-20.6
-19.7

Tennessee
69,399
12.3
25.2
-15.9
-15.0

Texas
204,228
14.9
32.6
8.8
10.0
All Sunbelt States
1,361,189
13.0
29.6
-1.1








All Non-Sunbelt States
2,017,952
17.2
30.2
0.7

United States
3,379,141
15.2
30.0



Sources: U.S. Census Bureau, 2000 Census IPUMS 5 percent files and Census 2000 Special Tabulation PHC-T-22, Gross Migration Table.


Table 5. Simple Correlation Matrix, 116 Sunbelt MSAs


PG
EG
PCI
PCIG
SC
SCG

EG
0.920






PCI
0.192
0.096





PCIG
-0.079
0.047
0.104




SC
0.281
0.241
0.647
0.108



SCG
0.284
0.345
0.227
0.144
0.058


YSCMR
0.180
0.102
0.257
0.140
-0.270
0.277

PG: Population growth rate, 1990-2000






EG: Employment growth rate, 1990-2000






PCI: Per capita personal income, 2000






PCIG: Per capita personal income growth rate, 1990-2000






SC: Share of college graduates in the 25+ population, 2000





SCG: Growth rate of the share of college graduates in the 25+ population




YSCMR: Share of single, college educated net migrants in the 5+ population





Note: Coefficients with statistical significance at the .05 level are in bold font.





Source: computed by the authors.








a more productive/innovative workforce produces a
argument. Similar positive and significant coefficients
higher standard of living. The correlations are more
are found between both the share of college graduates
modest between the other two education variables and
and growth in that measure and job growth – support-
per capita income, but still statistically significant –
ing the idea that a more educated workforce promotes
providing more support for the knowledge economy
aggregate economic growth. However, these simple

---

184
Hoffman and Hogan
statistical tests do not indicate the presence of strong
level implies convergence over the period with faster
links between the education measures and per capita
income growth in lower-income MSAs. It should also
income growth – a result not supporting the know-
be noted that the value of adjusted R2 shows that the
ledge economy hypothesis.
percent of college educated was able to explain little of

the pattern of income change among Sunbelt MSAs –
5.2. Glaeser and Saiz-like regression models
at least in the simpler regression model.

For the more complex version (Equation B), posi-
Our regression analysis using the Sunbelt MSA
tive and statistically significant coefficients were found
database mirrors some of the regression models esti-
in the population growth and both income equations,
mated by Glaeser and Saiz. As in their analysis, our
but not for employment growth. In the population
models take a form in which the change in the eco-
change equation, no significant link was indicated be-
nomic measure over the 1990–2000 period is a function
tween initial population size and the growth rate, but
of values of the explanatory variables at the beginning
for the income change equation, the negative and sig-
of the period. While regression models only identify
nificant coefficient for initial income level implies con-
correlation, not causality, this formulation is much less
vergence with faster growth in the lower-income
subject to the added confusion with respect to the di-
MSAs. For employment growth, the results do not in-
rection of causality found in models based on contem-
dicate a significant link with either the size of the labor
poraneous dependent and explanatory variables.
market or the unemployment rate at the beginning of
We estimated alternative regression equations for
the period.
each of our four measures (Tables 6a through d). In
Including the initial population as a scale variable
each of the four sets, equation A includes only the per-
in the employment and income models (Equation C)
cent college educated in 1990 and the log of the 1990
produced very different results. While the R2 for the
level of the respective dependent variable. Equation B
employment growth equation remained small it did
includes initial percent college educated, the log of the
improve substantially, and the estimated coefficient
initial level of the dependent variable, the log of heat-
for the education was positive and significant. The
ing degree days, the log of average precipitation, the
estimated coefficients for the initial population and the
unemployment rate, and the shares of the labor force
initial employment level were also indicated to be sta-
in (a) manufacturing, (b) trade, and (c) professional
tistically significant but with opposite signs. The posi-
services. For employment growth and the income va-
tive coefficient for the initial population variable
riables, a third model, equation C was also estimated
would indicate faster employment growth in the larg-
that included the log of 1990 population as a scale va-
er MSAs, but the negative sign for employment seems
riable.
to contradict that implication. One possible explana-
Note that our models retained the two climate va-
tion might be convergence with faster growth in those
riables included in the Glaeser and Saiz models, even
MSAs with relatively low labor force participation. For
though our analysis is based on a sample of Sunbelt
both income models, the inclusion of the initial popu-
MSAs rather than a national sample. This approach
lation variable caused the estimated coefficient of the
was chosen for two reasons: first, we wanted to follow
education variable to become statistically insignificant.
their formulation to investigate how the results
As in the Glaeser and Saiz analysis, all the explana-
changed looking only at the Sunbelt region. More im-
tory variables except education were included in the
portantly, however, there is substantial heterogeneity
equations as controls, with the major focus of the exer-
in climate across the Sunbelt, and we wanted to inves-
cise to look at the impact of the stock of human capital
tigate whether climate differences also had effects
(as measured by percent college educated) on econom-
within the region.
ic growth. However, it is interesting to note in passing
Looking first at the regression results for equation
the differences in the results with respect to the cli-
A, the coefficient for the education variable was posi-
mate variables. Glaeser and Saiz used a national sam-
tive and statistically significant only in the equations
ple for their analysis and found that “warm, dry plac-
for the two income variables. Not surprisingly the pos-
es grew much more quickly than cold, wet places.”
itive and significant coefficient for initial income level
(2003, p. 10) Since our sample included only Sunbelt
in Table 6c implies a strong connection between the
MSAs, we were not sure what to expect, and the re-
1990 income level and at its level in 2000. The high
sults varied among the four sets of equations. In the
value of adjusted R2 for the equation also emphasizes
population change equation – equivalent to the Glaes-
the strength of that relationship. For the income
er and Saiz models – no significant link was found
change equation (Table 6d) on the other hand, the
with the temperature variable but the estimated coeffi-
negative and significant coefficient for initial income
cient for average precipitation was negative and sig-

---

Sunbelt Growth

185

Table 6a. Regressions for Population Growth: log (2000 Population) – log (1990 Population)




(A)
(B)

Coeff.
S.E.
Coeff.
S.E.


Percent College Graduates-1990
0.2067
0.1819
0.7473
0.3150

Log(1990 Population)
0.0313
0.0094
0.0036
0.0122

Log(Ave. Heating Degree Days)


-0.0154
0.0136

Log(Ave. Precipitation)


-0.0409
0.0165
Unemployment Rate - 1990


-0.1118
0.6273

Percent Employment by Industry



Manufacturing


-0.1556
0.2078

Trade


0.4828
0.5452

Professional Services


-0.8639
0.3560

Observations
116

112


Adjusted R-squared
0.114

0.233









Note: Coefficients with statistical significance at the .05 level are indicated with bold font.


Source: Computed by the authors.









Table 6b. Regressions for Employment Growth: log (2000 Employment) – log (1990 Employment)


(A)
(B)
(C)

Coeff.
S.E.
Coeff.
S.E.
Coeff.
S.E.








Percent College Graduates-1990
0.2527
0.1986
0.6140
0.3594
0.9033
0.3575

Log(1990 Population)




0.6228
0.2001

Log(1990 Employment)
0.0120
0.0100
0.0006
0.0135
-0.6126
0.1974

Log(Ave. Heating Degree Days)


0.0015
0.0153
0.0168
0.0155

Log(Ave. Precipitation)


-0.0092
0.0186
-0.0148
0.0180

Unemployment Rate - 1990


0.3941
0.7000
-1.8610
0.9883

Percent Employment by Industry






Manufacturing


-0.2905
0.2344
-0.2814
0.2252

Trade


0.5077
0.6147
0.3974
0.5914

Professional Services


-0.5863
0.4035
-0.7362
0.3905



Observations
116

112

112

Adjusted R-squared
0.025

0.059

0.097






Note: Coefficients with statistical significance at the .05 level are indicated with bold font (source: computed by the authors).




nificant. In the employment growth equation, no sta-
6. Commentary
tistically significant link with either climate measure

was found. In the two income equations, however, the
A review of domestic migration data reveals that
coefficients of both climate variables were positive and
while the Sunbelt has been the beneficiary of signifi-
significant – implying higher levels of per capita in-
cant net domestic migration flows, these flows (with
come and faster income growth in relatively cooler,
the exception of California, and to some degree Geor-
wetter places versus warmer, drier places.
gia and Texas) are not necessarily being accompanied

by large numbers of people prepared to contribute to

knowledge economy endeavors.

---