parkdeejing

Alm and Rogers                                                                                                      509

(continued)

Variable: rvTXTOTAL

Per Capita                                    Percent of Income

Corporate income taxation (rvTXINCcor) is represented as a per capita amount, as a percent of income, or as a percent of total tax revenue. It might be expected that greater reliance on the corporate income tax would have a negative effect on economic growth. However, the coefficient on rvTXINCcor is never significantly negative and is frequently significantly positive at conventional levels, especially in regressions E, F, and G.

Variable: rvTXINCcor

Year: t0

Per Capita             Percent of Income            Percent of Total Tax

B      C       D     EFG    B          C         D       EFG       B         C          D        EFG

asl 1977                            þ3                                  þ3                                        þ1 psl 1977             þ1           þ2           þ0

1959         þ0   þ0   þ0   þ3    þ0       þ1      þ0     þ3       þ0      þ1      þ1       þ2 s 1959                                þ1                                           þ2                                                                       þ0

1947                                 þ0                                  þ0                                        0

Similar results are found for the individual income tax variable (rvTXINCind). The estimated coefficient is never significantly negative at conventional levels, but its coefficient is often significantly positive.

Variable: rvTXINCind

Per Capita             Percent of Income           Percent of Total Tax

(continued)

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

510                                                                                     Public Finance Review 39(4)

(continued)

Variable: rvTXINCind

Year: t0

Per Capita             Percent of Income            Percent of Total Tax

B      C       D     EFG    B          C         D       EFG       B         C          D        EFG

Not all states impose a general sales tax (rvTXSALgen). Even so, the coef- ficients are generally positive, though not always statistically significant.

Variable: rvTXSALgen

Per Capita             Percent of Income           Percent of Total Tax

Perhaps, surprisingly, property taxes (rvTXPROP) are generally found to have a positive impact on state economic growth, a result that may be due to the improved local infrastructure that can be financed with higher property taxes.

Variable: rvTXPROP

Year: t0

Per Capita             Percent of Income            Percent of Total Tax

B      C       D     EFG    B          C         D       EFG       B         C          D        EFG

asl 1977                            þ3                                   þ3                                        þ3 psl 1977             þ3           þ3           þ3

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

Alm and Rogers                                                                                                      511

The   coefficient   on  total   transfers  from  the   federal   government

(rvTRFtot) is always positive and generally significantly so.

Variable: rvTRFtot

Year: t0

Per Capita             Percent of Income            Percent of Total Tax

B      C       D     EFG    B          C         D       EFG       B         C          D        EFG

asl 1977                           þ3                                   þ3                                        þ3 psl 1977              þ3           þ3           þ3

1959        þ2   þ3   þ3   þ3    þ0       þ2     þ1     þ3      þ1      þ3       þ2      þ0 s 1959                 þ3                                           þ3                                                                   þ3

1947                                 þ3                                  þ2                                        þ3

Similarly, federal transfers for education (rvTRFedu) are significantly and positively correlated with income growth in all instances. The magni- tude of the coefficient indicates that each additional one dollar in per capita transfers is associated with an increase in per capita income growth rates by one-hundredth of a percentage point.

Variable: rvTRFedu

Year: t0

Per Capita             Percent of Income            Percent of Total Tax

B      C       D     EFG    B          C         D       EFG       B         C          D        EFG

asl 1977                            þ3                                  þ3                                        þ3 psl 1977              þ3           þ3           þ3

1959        þ3   þ3   þ3   þ3    þ3       þ3      þ3     þ3      þ3       þ3      þ3      þ3

s 1959                              þ3                                  þ3                                        þ3

1947                                 þ3                                  þ3                                        þ3

In contrast, federal transfers for highways (rvTRFhwy) are not con- sistently related to economic growth. Depending on the specification, the estimated coefficient is sometimes positive and significant, some- times negative and significant, and sometimes insignificant. The nega- tive relationship between highway transfers and growth is most pronounced  after  1977,  perhaps,  due  to  the  need  for  state  matching funds. In addition, there are likely to be long lags associated with any benefits from highway construction.

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

512                                                                                     Public Finance Review 39(4)

Variable: rvTRFhwy

Per Capita              Percent of Income            Percent of Total Tax

On   the   expenditure   side,   education   expenditures   (spEDUtot)   are measured by spending on primary and secondary education. This variable is always negatively and significantly correlated with income growth. It is possible that greater expenditures on education reflect a higher propor- tion of the population under the age of eighteen, and this larger population group may not contribute in a positive way to economic growth.

Variable: spEDUtot

Per Capita              Percent of Income            Percent of Total Tax

Similarly, the estimated coefficient for expenditures on highways (including capital construction) always has a negative correlation with per capita income growth, and the coefficient is typically (though not always) significant. This result suggests that highway infrastructure does not con- tribute positively to sustained economic growth.

Variable: spHWYtot

Per Capita              Percent of Income            Percent of Total Tax

(continued)

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

Alm and Rogers                                                                                                      513

(continued)

Variable: spHWYtot

Year: t0

Per Capita              Percent of Income            Percent of Total Tax

B      C       D     EFG    B          C         D       EFG       B          C          D        EFG

1959         1     3     3     3      0     1     2     3      0      2      3      3

Welfare expenditures (spWELtot) include intergovernmental expendi- tures for locally administered welfare programs as well as expenditures to offset federal payments for supplemental programs; cash assistance is included but health and hospital services are not. This variable is always negatively correlated with growth, although its coefficient is not always sig- nificant at conventional levels.

Variable: spWELtot

Year: t0

Per Capita             Percent of Income            Percent of Total Tax

B      C        D     EFG    B           C          D       EFG       B          C          D        EFG

asl 1977                            3                                   3                                       3 psl 1977            3           3           3

1959        1     3     3     3      0     1     2     3      0      2      3      3

s 1959                              3                                   3                                       3

1947                                 3                                   3                                       3

Finally, spCAPhwy denotes direct capital outlays for the construction of roads and for the purchase of equipment, land, and other structures neces- sary for their use; it includes amounts for additions, for replacements, and for major alterations, but it excludes expenditures for repairs. One would expect a positive correlation between spCAPhwy and growth; however, the correlation is always negative and is often statistically significant.

Variable: spCAPhwy

Per Capita              Percent of Income            Percent of Total Tax

(continued)

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

514                                                                                     Public Finance Review 39(4)

(continued)

Variable: spCAPhwy

Year: t0

Per Capita              Percent of Income            Percent of Total Tax

B      C        D     EFG    B           C         D       EFG       B          C          D        EFG

Alm and Rogers                                                                                                      515

F and G, which cover the period from 1977 to 1996 and which exclude the five high volatility states, indicate that passage of a TEL reduces per capita income growth by about three-tenths of a percentage point.

Variable: dmTXref

Per Capita             Percent of Income           Percent of Total Tax

Another political variable measures the frequency of party change (gePOLCgov). One can argue that a state that changes its governing party more frequently is somewhat unstable, which would inhibit growth. One can also argue that a higher value of gePOLCgov indicates a state with a greater willingness to undertake risks or a state with a balanced political orientation, both of which might be reflected  in higher growth (Crain

2003). The  sign of gePOLCgov is always positive  and, at  least  since

1977, always significant.

We include various geographic variables, reflecting the size of the state’s land area (geSIZ), the ratio of federal land to total land area (geSIZPf), and adjacency to the East Coast (geREGatl) or the West Coast (geREGpac). The coefficient on land area is seldom significant, and the coefficient on geSIZPf is generally negative and significant, indicating that federal occu- pation of state lands discourages economic growth. As for the adjacency variables, being on the Atlantic Ocean or the Gulf of Mexico tends to have a positive impact on growth, while being in a state that adjoins the Pacific Ocean has a consistent negative impact.

Demographic variables—the state’s population in millions (dmPOP) or the ratio of state population to state land area (dmDEN)—both have erratic and inconsistent impacts on growth. Several other variables that measure the  coefficient  of  variation  of  wages in  six  employment  sectors (dm- WAGEcv) and the coefficient of variation of payrolls in these same sectors (dmPRNFPcv) also have inconsistent, though largely negative, effects on growth. Because larger values for these variables indicate greater disparity in either the level of wages (dmWAGEcv) or the level of employment

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

516                                                                                     Public Finance Review 39(4)

(dmPRNFPcv) in these sectors, the negative coefficients on these variables suggest that the concentration of a state’s employment base in fewer sectors has a positive effect on growth.

Overall, these different results tend to be somewhat more robust than those for the fiscal variables (especially the tax variables).

Conclusions

This article reports the results of an empirical analysis of economic growth in the United States for the years 1947 through 1997, presenting empirical results against which theoretical models of economic growth can be com- pared. The analysis uses annual data to examine the effects of government policy variables at the state and local levels, as well as the effects of a wide range  of  other  socioeconomic,  demographic, geographic, and  political variables.

The empirical literature on economic growth includes hundreds of arti- cles examining the growth effects of a multitude of variables. Our article differs from these studies in several important ways: it examines annual data over a longer period than most other studies, it includes a much more comprehensive collection of explanatory variables, and it addresses the measurement errors inherent in per capita income data.

Several main conclusions emerge.

First, our estimation results indicate that a state’s fiscal policies have a measurable relationship with per capita income growth, although not always in the expected direction and seldom in a way that is robust to alter- native specifications. Tax impacts on state economic growth are quite vari- able; expenditure               impacts  are  more       consistent             across    different specifications. The statistically significant correlation between state (and state plus local) total tax revenues and economic growth is very sensitive to the regressor set and the time period examined. Often, there are highly significant correlations measured between these variables and per capita income growth, but further work needs to be done before it can be deter- mined what these results mean.

Second, there is strong evidence that a state’s political orientation, as

indicated by whether the governor is Republican or Democrat, whether the state has enacted TEL legislation, and whether the state frequently elects a governor of the same party as the incumbent, have consistent, measurable, and significant effects on economic growth. Perhaps, surprisingly, having a Republican governor is associated with lower rates of growth.

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

Alm and Rogers                                                                                                      517

Third, the methods commonly used for growth regression analyses could be inadequate and could adversely affect the results because most previ- ously reported results have not taken measurement errors into account. Again, we do not discuss these results in detail here, but we have some evi- dence that it is very likely that measurement errors have had a significant impact on previously reported growth regression results, especially with regard to convergence. Indeed, although ordinary LLS estimates suggest that  there  is  conditional  convergence  in  per  capita  income  across the forty-eight contiguous states, our ODR estimates indicate strong evidence of divergence.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interests with respect to the authorship and/or publication of this article.

Funding

The authors received no financial support for the research and/or authorship of this article.

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

518                                                                                     Public Finance Review 39(4)

Appendix

Table A1. Regression A (1959–1997)a

Variable          Method             Estimate

Standard

Deviation

t-Value

H0: yi  ¼ 0

H0: yi  ¼ 0

One-Sided Marginal Significance Level

usGRW          ODR                    0.9302812     0.0289600      32.12           0.00%*** LLS        0.8378600               0.0531617               15.76         0.00%***

usINF           ODR                    0.0000929     0.0001454        0.64        26.14%

LLS                      0.0001690     0.0001344        1.26        10.44%

usFUELpp   ODR                  0.0019124     0.0003440     5.56           0.00%*** LLS        0.0019699            0.0003242               6.08       0.00%***

geREGcon     ODR                    0.0160398     0.0020163        7.96         0.00%***

LLS                                            0.0203461   0.0019015            10.70  0.00%*** gePOLstate   ODR    0.0036382               0.0006269                              5.80                0.00%*** LLS 0.0034087   0.0001286     26.51  0.00%***

y0                             ODR                  0.0012481     0.0001683     7.42           0.00%*** LLS        0.0015189            0.0000344     44.21 0.00%***

Rho               ODR                    0.0099271     0.0015337        6.47           0.00%*** LLS        0.0046268               0.0004614               10.03         0.00%***

R2                            ODR                  99.9

LLS                    96.7

‘(.)                 ODR           9,781.8

LLS          12,325.3

se                            ODR                163.1

NLS             1,032.7

LLS                      0.05554

sey                                                       ODR                165.2

LLS                 Not applicable

seyt

se

ODR                  39.2

LLS                 Not applicable

ODR                    0.00914

LLS                 Not applicable

* H0    is rejected at a ¼ 20% significance level.

** H0    is rejected at a ¼ 10% significance level.

*** H0    is rejected at a ¼ 5% significance level.

a  ‘(.) denotes the value of the likelihood function. [se, se , se

yt0

, se  ] denotes the standard devia-

tions of the estimated residuals, the measurement error of income, the measurement error for

initial income y0, and the model, respectively; the standard deviation se is a weighted average of the measurement error  standard deviations and the model standard deviation.

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

Alm and Rogers                                                                                                      519

Table A2. Regression B (Per Capita; 1959–1997)a

Variable              Method          Estimate

Standard

Deviation

t-Value

H0: yi  ¼ 0

H0: yi  ¼ 0

One-Sided Marginal Significance Level

usGRW             ODR              0.9164151      0.0305155       30.03            0.00%*** LLS        0.8094562               0.0486160               16.65       0.00%***

usINF               ODR            0.0005217      0.0001752      2.98            0.15%*** LLS                        0.0000344               0.0001224                                 0.28                   38.94%

usFUELpp       ODR            0.0022860      0.0004129      5.54            0.00%*** LLS        0.0018108               0.0003250               5.57      0.00%***

geREGcon        ODR              0.0159364      0.0028225         5.65            0.00%*** LLS        0.0251525               0.0017342               14.50       0.00%***

gePOLstate       ODR              0.0026335      0.0007548         3.49            0.02%*** LLS        0.0024147               0.0001521               15.88       0.00%***

y0                                  ODR           0.0005696      0.0002808      2.03            2.13%*** LLS        0.0019923               0.0000757     26.31   0.00%***

Rho                   ODR              0.0100623      0.0015213         6.61            0.00%*** LLS        0.0049265               0.0004665               10.56       0.00%***

rvTXTOTAL    ODR            0.0031377      0.0027695         1.13        12.87% LLS           0.0082254               0.0008457               9.73                               0.00%***

rvTXINCcor     ODR              0.0089821      0.0097420         0.92          17.83% LLS              0.0034075               0.0028961               1.18          11.98%

rvTXINCind    ODR              0.0032783      0.0030738         1.07        14.32% LLS           0.0022551            0.0008762               2.57                         0.51%***

rvTXSALgen    ODR              0.0028672      0.0032352         0.89          18.78% LLS              0.0002181            0.0008425               0.26       39.79%

rvTXPROP      ODR              0.0060951      0.0024430         2.49            0.63%*** LLS        0.0022668               0.0005968  3.80         0.01%***

rvTRFtot          ODR              0.0090632      0.0046069         1.97            2.47%*** LLS        0.0018527               0.0017677  1.05      14.74%

rvTRFedu         ODR              0.1209462      0.0185788         6.51            0.00%*** LLS        0.0326072               0.0062439  5.22         0.00%***

rvTRFhwy       ODR            0.0036942      0.0127112      0.29         38.57%

LLS            0.0250445      0.0039599         6.32            0.00%*** spEDUtot                               ODR         0.0210545               0.0033184                         6.34        0.00%*** LLS      0.0119434 0.0010889      10.97     0.00%***

spHWYtot        ODR            0.0017725      0.0098695      0.18         42.87%

LLS            0.0061415         0.0024377         2.52            0.59%***

(continued)

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

520                                                                                     Public Finance Review 39(4)

Table A2 (continued)

spWELtot         ODR           0.0095676      0.0046721      2.05            2.04%*** LLS       0.0080643               0.0017390               4.64      0.00%***

spCAPhwy      ODR           0.0129220      0.0136243      0.95         17.15%

LLS               0.0182959      0.0041463         4.41           0.00%***

R2                                  ODR           99.9

LLS             97.5

seyt

se

ODR           40.7

LLS          Not applicable

ODR             0.00881

LLS          Not applicable

* H0    is rejected at a ¼ 20% significance level.

** H0    is rejected at a ¼ 10% significance level.

*** H0    is rejected at a ¼ 5% significance level.

a  ‘(.) denotes the value of the likelihood function. [se, se , se

yt0

, se  ] denotes the standard devia-

tions of the estimated residuals, the measurement error of income, the measurement error for

initial income y0, and the model, respectively; the standard deviation se is a weighted average of the measurement error  standard deviations and the model standard deviation.

Table A3. Regression C (Per Capita; 1959–1997)a

(continued)

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

Alm and Rogers                                                                                                      521

Table A3 (continued)

(continued)

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

522                                                                                     Public Finance Review 39(4)

Table A3 (continued)

LLS            0.0041743      0.0004759     8.77            0.00%***

gePOLallR          ODR             0.0002229      0.0000990      2.25              1.23%*** LLS       0.0000209            0.0000221     0.95 17.15%

(continued)

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

Alm and Rogers                                                                                                      523

Table A3 (continued)

LLS            0.0046978      0.0008205     5.73            0.00%***

dmISNFPcv       ODR           0.0144514      0.0072304     2.00             2.29%*** LLS       0.0067683               0.0018157     3.73    0.01%***

dmISFEDpc       ODR           0.0019222      0.0013498     1.42            7.73%* LLS        0.0024400            0.0004793     5.09  0.00%***

LLS            0.0000456      0.0000328     1.39            8.22%*

dmPRNFPtpu    ODR           0.0012225      0.0007173     1.70            4.43%** LLS         0.0010796            0.0001606     6.72  0.00%***

dmPRNFPser     ODR              0.0000614      0.0002306      0.27          39.51%

LLS            0.0001070      0.0000518     2.07            1.94%***

R2                                    ODR           99.9

LLS             98.3

‘(.)                      ODR          9,695.6

LLS          11,641.3

se                                    ODR         161.0

NLS          702.6

LLS               0.03953

sey                                                                       ODR         162.9

LLS          Not applicable

seyt

ODR           50.8

LLS          Not applicable

(continued)

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

524                                                                                     Public Finance Review 39(4)

Table A3 (continued)

se                                    ODR             0.00843

LLS          Not applicable

* H0    is rejected at a ¼ 20% significance level.

** H0    is rejected at a ¼ 10% significance level.

*** H0    is rejected at a ¼ 5% significance level.

a    ‘(.) denotes the value of the likelihood function. [se, se , se

yt0

, se  ] denotes  the standard

deviations of the estimated residuals, the measurement error of income, the measurement

error  for initial income y0,  and the  model, respectively; the  standard  deviation se     is a weighted average of the measurement  error  standard  deviations  and the model  standard deviation.

References

Akai, Nobuo, and Masayo Sakata. 2002. Fiscal decentralization contributes to eco- nomic growth: Evidence from state-level cross-section data for the United States. Journal of Urban Economics 52:93–108.

Barro, Robert J., and Xavier Sala-i-Martin. 1991. Convergence across  states and regions. Brookings Papers on Economic Activity 1:107–82.

———. 1992. Public finance in models of economic growth. Review of Economic

Studies 59:645–61.

Berry, Dan L. and David L. Kaserman. 1987. A diffusion model of long-run state economic development. Atlantic Economic Journal 21:39–54.

Berry, William D., Richard C. Fording, and Russell L. Hanson. 2000. An annual cost of living index for the American states, 1960-1995. The Journal of Politics 62:550–67. Boggs, Paul T., Richard H. Byrd, and Robert B. Schnabel. 1987. A stable and effi- cient algorithm for nonlinear orthogonal distance regression. Society for Indus- trial  and  Applied Mathematics  (SIAM) Journal  of Scientific and  Statistical

Computing 8:1052–78.

Boggs, Paul T., Janet Rogers Donaldson, Robert B. Schnabel, and  Clifford H.

Spiegelman. 1988. A computational examination of orthogonal distance regres- sion. Journal of Econometrics 38:169–201.

Canto, Victor, and Robert I. Webb. 1987. The effect of state fiscal policy on state relative economic performance. Southern Economic Journal 54:186–202.

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

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

Alm and Rogers                                                                                                      525

Coughlin, Cletus C., and Thomas B. Mandelbaum. 1989. Have federal spending and taxation contributed to the divergence of state per capita incomes in the 1980s? Federal Reserve Bank of St. Louis Review 71:29–42.

Crain, W. Mark. 2003. Volatile states: Institutions, policy, and the performance of

American state economies. Ann Arbor: The University of Michigan Press. Crain, W. Mark, and Katherine J. Lee. 1999. Economic growth regressions for the

American states: A sensitivity analysis. Economic Inquiry 37:242–57.

De Long, J. Bradford. 1988. Productivity growth, convergence, and welfare: Com- ment. The American Economic Review 78:1138–54.

Garofalo, Gasper A., and Steven Yamarik. 2002. Regional convergence: Evidence from a new state-by-state capital stock series. The Review of Economics and Sta- tistics 84:316–23.

Holcombe, Randall G., and Donald J. Lacombe. 2004. The effect of state income taxation on per capita income growth. Public Finance Review 32:292–312. Leamer, Edward E. 1983. Let’s take the con out of econometrics. The American

Economic Review 73:31–43.

———. 1985. Sensitivity analysis would help. The American Economic  Review

75:308–13.

Levine,  Ross, and David Renelt. 1992. A sensitivity analysis of  cross-country growth regressions. The American Economic Review 82:942–63.

Mofidi, Alaeddin, and Joe A. Stone. 1990. Do state and local taxes affect economic growth? The Review of Economics and Statistics 72:686–91.

Mullen, John K., and Martin Williams. 1994. Marginal tax rates and state economic growth. Regional Science and Urban Economics 24:687–705.

Persson, Torsten, and Guido Tabellini. 1992. Growth, distribution, and  politics.

European Economic Review 36:593–602.

Phillips, Joseph M., and Ernest P. Goss. 1995. The effect of state and local taxes on  economic  development:  A  meta-analysis.  Southern   Economic  Journal

62:320–33.

Reed, W. Robert. 2008a. The robust relationship between taxes and  U.S. state income growth. National Tax Journal 61:57–80.

———. 2008b. The determinants of U.S. state economic growth: A less extreme bounds analysis. Economic Inquiry 47:685–700.

Romer, Paul M. 1987. Growth based on increasing returns due to specialization.

The American Economic Review 77:56–62.

———. 1990. Endogenous technological change. The Journal of Political Economy

98:S71–102.

Sala-i-Martin, Xavier. 1997. I just ran two million regressions. The American Eco- nomic Review, Papers and Proceedings of the American Economic Association

87:178–83.

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

Table 4. State Demographic Variables

Variable Name                        Class                       Subclass                                   Qualifier                                              Units

dmDENnf               Population                      None                     Population density on nonfederal land

dmDENnfsq           Population                      None                     Squared population density on

nonfederal land

land area

Population per land area

Squared population per land area

dmPOLgov             Political orientation           None                     Republican governor                                        Dummy variable dmPOLup                                Political orientation           None                    Republican majority in upper house               Dummy variable dmPOLlow                                Political orientation         None                     Republican majority in lower house                Dummy variable dmPOLboth                                Political orientation           None                    Republican majority in both houses                Dummy variable dmPOLnone                                Political orientation         None                     Republican majority in neither house              Dummy variable

dmPOLallR             Political orientation           None                     Republican governor and republican

majority in both houses dmPOLallD         Political orientation   None        Not republican governor and not

republican majority in either house

Dummy variable

Dummy variable

dmTXref                 Political orientation           None                     Tax and expenditure limit enacted                Dummy variable

dmTXsvl                Political orientation           None                     Percentage of total state plus local tax revenues

collected at the state level dmPRNFPcv  Payroll distribution None        Coefficient of variation of private

sector payrolls

dmISNFPcv            Income distribution         None                     Coefficient of variation of private sector incomes sources

dmWAGEcv           Wage distribution           None                     Coefficient of variation of private

sector wages

Percent

Percent of population Percent of population Percent of population

(continued)

Table 4 (continued)

Variable Name                         Class                       Subclass                                   Qualifier                                              Units

dmISFEDpc            Income sources                Nonfarm                Per capita income from federal income sources

dmPRNFtot            Payroll                            Nonfarm                Total, all private and government

sectors

Percent of population

Percent of population

dmPRNFPtot          Payroll                            Nonfarm                Total, all private sectors                               Percent of population dmPRNFPcon                                Payroll                            Nonfarm                Private: construction sector                          Percent of population dmPRNFPman                                Payroll                              Nonfarm               Private: manufacturing sector                           Percent of population

dmPRNFPtpu         Payroll                            Nonfarm                Private: transportation  and public utilities sector

dmPRNFPtrdt        Payroll                            Nonfarm                Private: wholesale & retail trade sector

dmPRNFPfin          Payroll                            Nonfarm                Private: finance and insurance and real estate sector

Percent of population Percent of population Percent of population

dmPRNFPser          Payroll                            Nonfarm                Private: services sector                                  Percent of population

dmPRNFGtot         Payroll                            Nonfarm                Government: federal, state, and local sectors

Percent of population

dmIStot                   Income sources                None                     Total personal income                                 Real $

dmISPROtot           Income sources                Proprietors’          Total                                                             Percent of total income dmISFRMtot                                Income sources                Farm                     Total                                                             Percent of total income dmISNFtot                                Income sources                Nonfarm                Total                                                             Percent of total income dmISNFPtot                                Income sources                Nonfarm                Private: total, all private sectors                  Percent of total income

dmISNFPag             Income sources                Nonfarm                Private: agricultural services and

forestry and fishing and other sector

Percent of total income

Table 4 (continued)

Variable Name                        Class                        Subclass                                   Qualifier                                              Units dmISNFPman   Income sources                     Nonfarm                          Private: manufacturing sector            Percent of total income

dmISNFPtpu           Income sources                Nonfarm                Private: transportation  & public utilities sector

dmISNFPtrd            Income sources                Nonfarm                Private: wholesale and retail trade sector

dmISNFPfin            Income sources                Nonfarm                Private: finance and insurance

& real estate sector

Percent of total income Percent of total income Percent of total income

dmISNFPser            Income sources                Nonfarm                Private: services sector                                 Percent of total income

dmISNFGtot           Income sources                Nonfarm                Government: federal, state, and local sectors

Percent of total income

dmISNFGfed           Income sources                Nonfarm                Government: federal                                    Percent of total income

dmISNFGmil           Income sources                Nonfarm                Government: federal military                      Percent of total income

dmISNFGsl             Income sources                Nonfarm                Government: state and local sectors

Percent of total income

dmWAGEmin         Wage & salary                Nonfarm                Private: mining sector                                    Percent of total income

dmWAGEcon          Wage & salary                Nonfarm                Private: construction sector                         Percent of total income dmWAGEman                                Wage & salary                Nonfarm                Private: manufacturing sector                            Percent of total income

dmWAGEtpu          Wage & salary                Nonfarm                Private: transportation public utilities sector

dmWAGEtrd           Wage & salary                Nonfarm                Private: wholesale and retail trade sector

dmWAGEfire          Wage & salary                Nonfarm                Private: finance, insurance, real estate sector

Percent of total income Percent of total income Percent of total income

dmWAGEser           Wage & salary                Nonfarm                Private: services sector                                 Percent of total income

dmWAGEgov          Wage & salary                Nonfarm                Government: federal, state, and local sectors

Percent of total income

Table 5. State Geographic Variables

Variable Name   Class                            Subclass              Qualifier                                                             Units

geHEtotP           People                        Socioeconomic  1947 higher education enrollment:

total first time students geHEwmP        People     Socioeconomic  1947 higher education enrollment:

women first time students

Percent of 1947 population

Percent of 1947 enrollment

geSIZt                Land                           Area                    Total excluding water areas                                  Hundreds of square miles geSIZPf   Land                           Area                           Federal surface area, 1982                                                             Percent of total state area geSIZPw                           Land                           Area                    Woodlands, 1982                                               Percent of total state area geSIZPr                           Land                           Area                    Rangelands, 1982                                                  Percent of total state area gePOLstate                           Political orientation   None                     Statehood granted prior to 1800                      Dummy variable

gePOLgov          Political orientation   None                     Average of dmpolgovs,t  for all t                          Percent gePOLup                           Political orientation   None                    Average of dmpolups,t  for all t                             Percent gePOLlow                           Political orientation   None                     Average of dmpollows,t  for all t                         Percent gePOLboth                           Political orientation   None                    Average of dmpolboths,t   for all t                          Percent gePOLnone                           Political orientation   None                     Average of dmpolnones,t   for all t                       Percent gePOLallR                           Political orientation   None                    Average of dmpolallrs,t  for all t                             Percent gePOLallD                           Political orientation   None                     Average of dmpolallds,t  for all t                          Percent gePOLCgov                           Political orientation   None                    Standard deviation of dmpolgovs,t  for all t            Percent gePOLCup                           Political orientation   None                     Standard deviation of dmpolups,t  for all t          Percent gePOLClow                           Political orientation   None                    Standard deviation of dmpollows,t  for all t           Percent gePOLCboth                           Political orientation   None                     Standard deviation of dmpolboths,t   for all t      Percent gePOLCnone                           Political orientation   None                    Standard deviation of dmpolnones,t   for all t         Percent gePOLCallR                           Political orientation   None                     Standard deviation of dmpolallrs,t  for all t         Percent gePOLCallD                           Political orientation   None                    Standard deviation of dmpolallds,t  for all t            Percent

geREG1             Land                           Region                 New England(CT,  ME, MA, NH, RI, and VT)   Dummy variable

geREG2             Land                           Region                 Middle Atlantic (DE, MD, NJ, NY, and PA)        Dummy variable

(continued)

Table 5 (continued)

ND, and SD)

NC, SC, TN, VA, and WV)

Table 5 (continued)

deposits

individual state economies are small relative to the U.S. economy as a whole, usGRW is exogenous. Its coefficient is expected to be roughly one.

The regressor usINF is the national inflation rate and its coefficient is expected to be negative because inflation is generally presumed to be harmful to economic growth. The variable usFUELpp, which denotes the average national producer price of fuels, is another exogenous variable intended to capture the effects of external (fuel) shocks to the U.S. economy.

The baseline regression is estimated for three time periods 1947 through

1997, 1959 through 1997, and  1977 through 1997. These correspond, respectively, to the longest period for which annual state data are available, the longest period for which state plus local total tax and property tax rev- enue and state and local expenditure values are available, and the longest period for which all state plus local tax and expenditure values have been recorded at a relatively fine level of detail. The results from the baseline regression are presented in appendix tables and are discussed in the section on results. All regressions correct for first-order autocorrelation.

Beyond the baseline regression. In addition to the baseline regression, several other primary regression specifications are analyzed for the period 1959 through 1997, using each of three representations of the fiscal variables (value per capita, value as a percent of income, and value as a percent of total tax). These other primary specifications are denoted Regression B, Regression C, and Regression D.

Regression B includes the six core variables plus twelve fiscal variables:

     rvTXTOTAL: the sum of all state plus local taxes

     rvTXINCcor: corporate income tax revenues

     rvTXINCind: individual income tax revenues (only available at the local level after 1977)

     rvTXSALgen: state-level general sales tax revenues

     rvTXPROP: the sum of state plus local property taxes

     rvTRFtot: the total amount of revenues transferred from the federal to the state government

     rvTRFedu: the amount of revenues earmarked for education that is transferred from the federal to the state government

     rvTRFhwy: the amount of revenues earmarked for highways that is transferred from the federal to the state government

     spEDUtot: the sum of state plus local expenditures for primary and secondary education, including capital construction

     spHWYtot: the sum of state plus local expenditures for highways, including capital construction

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

Alm and Rogers                                                                                                      503

Table 6. Variables Associated with the United States as a Whole

Variable

Name               Class              Subclass Qualifier                          Units

usFUELpp     Miscellaneous None    Average producer price for fuels

usDEFcw        Miscellaneous None    Chained weight

deflator (1996 ¼ 1.0)

usDEFfw        Miscellaneous None    Fixed weight deflator

(1996 ¼ 1.0)

Real U.S.$ Percent Percent

usPOP            Miscellaneous None    Total population of 48 Thousands of persons contiguous states

(excluding DC)

usINCtot         Miscellaneous None    Total income of 48 contiguous states (excluding DC)

Thousands of persons

usGRW            Miscellaneous None    U.S. growth rate                 Percent

usINF             Miscellaneous None    U.S. inflation rate                 Percent

     spWELtot: the sum of state plus local expenditures for welfare

     spCAPhwy:  the  sum  of  state  plus  local  expenditures  for  capital construction of highways.

(Remember that many of the tax variables are not available at the local level prior to 1977.) This set of variables is assembled to examine the impact of tax, transfer, general expenditure, and capital outlay variables. See tables 2 and 3 for variable definitions.

Regression C includes the six core variables from Regression A plus the twelve fiscal variables included in Regression B, along with another thirty variables from the demographic, geographic, and national variable sets. This set of variables includes every variable in which anyone might reason- ably have any interest plus a few others thrown in for good measure. These variables are defined in tables 4 through 6.

Regression D represents a subset of variables used in Regression C. It  includes  the  twelve  fiscal  variables from  Regression B plus  thirteen variables selected on the basis of their explanatory power. These thirteen vari- ables are dmPOLgov, dmTXref, dmTXsvl, dmPOP, dmDEN, dmWAGEcv, dmPRNFPcv, geHEtotP, geSIZ, geSIZPf, geREGatl, geREGpac, and gePOLCgov. We believe that this variable set is the most representative and it is the one discussed in greatest detail in the section on results.

Finally, we have also estimated a wide variety of additional specifications.

Regression E uses the same regressors and time spans as Regression D but

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

504                                                                                     Public Finance Review 39(4)

excludes the five states with the highest variability of growth rates (Iowa, Montana, Nebraska, North Dakota, and South Dakota). Regression F has the same explanatory variables and uses the same subset of states as Regression E, but the time span is 1977 through 1996 rather than 1959 through 1996. Regression G is the same as Regression F except that total state plus local fiscal values are used for all tax and expenditure variables. We discuss summary results for these specifications later.

Aside from these specifications, it should be noted that we have estimated

many additional specifications, including ones in which we examine alterna- tive time periods, in which we include state plus local measures of all tax and expenditure variables, in which dummy variables for the presence (or absence) of specific tax instruments are used rather than their values, and in which the growth experience of two individual states (Colorado and Georgia) are examined separately. All results are available on request.

Results

Estimation results from some basic specifications are presented in appendix tables A1 through A3; all other results are available on request. The boxes included below summarize the outcomes from the regressions. Column head- ings within the boxes indicate the time period, the fiscal variable parameteri- zation (e.g., per capita, percent of income, and percent of total taxes), as well as the regression identifier (e.g., A, B, C, D, E, F, or G). The row headings indicate the construction of the fiscal variables: ‘‘asl’’ denotes that all fiscal variables are constructed using the sum of state plus local amounts, ‘‘psl’’ denotes that property taxes, total taxes, and expenditures values are constructed using the sum of state plus local amounts (while other revenue variables are composed of only state values), and ‘‘s’’ denotes that fiscal values are constructed using only state values. The individual box entries can be interpreted using the fol- lowing ‘‘legend’’:

Entry                              Sign of Coefficient                               Significance Level (a)

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

Alm and Rogers                                                                                                      505

so that numbers þ3 and þ2 denote positive coefficients with statistical significance at accepted levels (as do    3 and    2 for negative coefficients with some statistical significance), while numbers   1,   0, þ0, and þ1 denote little statistical significance. Note that the appendix tables display the results for both ODR and LLS methods. In general, the coefficient estimates from the two methods are similar, but there are also some striking differences. In partic- ular, nearly one-fourth of the coefficients estimated by the two methods have different signs. Furthermore, for more than 30 percent of these cases, one or the other estimate is significant at conventional levels; for more than 20 percent of these cases, both estimated coefficients are significant at conventional levels, but the correlation of one coefficient is positive while the correlation of the other coefficient is negative. When the two methods produce significantly dif- ferent coefficients, the Monte Carlo results discussed earlier indicate that the ODR coefficients are more likely to be reliable. For this reason, our discussion focuses on the ODR results. Note also that the constant term in the regressions (geREGcon) is generally positive and significant.

Core Variables

The coefficient of the U.S. per capita income growth rate (usGRW) is positive and significantly different than zero (at the 95 percent confidence level) in every instance. The estimated value is roughly 0.9 but it is signif- icantly different than 1.0 in nearly all regressions, indicating that the indi- vidual states follow approximately the same growth pattern as the country as a whole and are responding individually to the same shocks and stimuli in roughly the same manner as the national economy.

Variable: usGRW

Year: t0

Per Capita               Percent of Income         Percent of Total Tax

A     B      C       D    EFG   B          C         D      EFG     B           C         D        EFG

asl 1977  þ3                         þ3                                þ3                                     þ3 psl 1977              þ3           þ3           þ3

1959        þ3  þ3  þ3  þ3  þ3    þ3    þ3    þ3    þ3         þ3      þ3     þ3      þ3 s 1959                 þ3           þ3                                                           þ3

1947        þ3                         þ3                                þ3                                     þ3

For the U.S. inflation rate (usINF), our results indicate that higher infla- tion rates are significantly negatively correlated with per capita income growth in most specifications. The estimated coefficient suggests that a one

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

506                                                                                     Public Finance Review 39(4)

percent increase in inflation is associated with lower per capita income growth of roughly one-quarter of a percentage point.

Variable: usINF

Year: t0

Per Capita               Percent of Income         Percent of Total Tax

A     B      C       D    EFG   B          C         D      EFG     B           C         D        EFG

asl 1977    0                        3                                3                                     3 psl 1977              3           3           3

1959        þ0   3    3    3    3     3      3      3      3      3     3     3      3

s 1959                                   3                                3                                     3

1947        þ3                         3                                3                                     3

Rising energy costs are generally thought to adversely affect economic growth. Our results consistently confirm this, with a negative and statisti- cally significant coefficient on usFUELpp.

Variable: usFUELpp

Year: t0

Per Capita               Percent of Income         Percent of Total Tax

A     B      C       D    EFG   B          C         D      EFG     B           C         D        EFG

asl 1977    3                        3                                3                                     3 psl 1977             3           3           3

1959        3    3    3    3    3     3      3      3      3     3      3     3      3

s 1959                                   3                                3                                     3

1947        3                         3                                3                                     3

The dummy variable gePOLstate provides a simple designation of the

‘‘age’’ of the state, as determined by the year in which statehood was obtained. Values of 1 for gePOLstate identify states that acquired statehood prior to 1800 (e.g., ‘‘old’’ states), while values of 0 identify states that acquired statehood after 1800 (e.g., ‘‘young’’ states). The estimated coefficient for gePOLstate is always positive and statistically significant, indicating that older states have higher per capita income growth than younger states. This is a plausible result and is consistent with the presence of more developed infrastructures in older states. However, this result is not consistent with convergence.

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

Alm and Rogers                                                                                                      507

Variable: gePOLstate

Year: t0

Per Capita               Percent of Income         Percent of Total Tax

A     B      C       D    EFG   B          C         D      EFG     B           C         D        EFG

asl 1977  þ3                         þ3                                þ3                                     þ3 psl 1977             þ3           þ3           þ3

1959        þ3  þ3  þ3  þ3  þ3    þ3    þ3    þ3    þ3         þ3      þ3      þ3      þ3 s 1959                                þ3           þ3                                                           þ3

1947        þ3                          þ3                                þ3                                     þ3

The neoclassical growth model asserts that, ceteris paribus, an economy with a lower initial income will grow faster than an economy with a higher ini- tial income. However, in our results, initial income (ys,t ) has quite variable effects on the various specifications. When the explanatory variables include the full set of socioeconomic regressors, our results provide little support for conditional convergence and strong evidence of divergence after 1977.

When the coefficient on ys,t

is not significant at conventional levels,

it might be argued that multicollinearity among the regressors is the prob-

lem. However, variance decomposition results indicate that this is unlikely. Moreover, our Monte Carlo experiments indicate that the mea- surement errors in per capita income have a significant and adverse effect on LLS results when annual data are used. However, if annual data are not used, then the fiscal and policy variables that are being examined must be aggregated over the period between observations to obtain a single repre- sentative value, even though it is the effect of the variation of these fiscal and policy variables that we are seeking to measure. Hence, previously reported results of convergence are suspect either because they have not taken measurement errors into account or because the fiscal and policy variables have been recorded in such a way that their effect on economic growth cannot be accurately determined. The ODR results reported here suffer from neither of these problems.

Variable: ys,t

Per Capita               Percent of Income         Percent of Total Tax

(continued)

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

508                                                                                     Public Finance Review 39(4)

(continued)

Variable: ys,t

Per Capita               Percent of Income         Percent of Total Tax

Researchers have used a wide range of explanatory variables in their cross-regional studies. For example, Canto and Webb (1987) present a non-pooled regression of  cross-region annual  U.S. data  for the  period

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

Alm and Rogers                                                                                                      487

1957 through 1977. Their independent variable is the average state growth rate and explanatory variables include the U.S. growth rate, the difference between the state’s government purchases and the average of all states’ gov- ernment purchases, the difference between the state’s transfer payments and the average of all states’ transfer payments, and the difference between the state’s relative tax burden and the average of all states’ relative tax burden. Similarly, Coughlin and Mandelbaum (1989) compare U.S. state per capita incomes as a percent of average state per capita income and overall state income inequality with regional variables that indicate coastal, energy pro- duction, sun belt, and ‘‘farm-crises’’ states. Barro and Sala-i-Martin (1991) examine cross-region data for the U.S. states using various subintervals for the period 1840 through 1985. They regress the average growth rate against initial income, three regional specifications (South, Midwest, and West), and employment composition for nine industrial sectors. For some other cross-region studies, see Berry and Kaserman (1987), Mofidi and Stone (1990), Mullen and Williams (1994), and Phillips and Goss (1995). More recently, Crain and Lee (1999), Caselli and Coleman (2001), Akai and Sakata (2002), Garofalo and Yamarik (2002), Tomljanovich (2004), and Holcombe and Lacombe (2004) conduct similar analyses, with quite mixed results. In perhaps the most comprehensive work to date, Reed (2008a,

2008b) uses five-year data from 1970 to 1999 for the forty-eight continental states and finds a significant negative relationship between taxes and state economic growth across a wide range of specifications and estimation procedures.

These growth regressions have produced a variety of results and only modest consistency. A similar lack of consensus exists in cross-country growth regressions. In a survey of this latter work, Levine and Renelt (1992) quantify whether the conclusions from cross-country studies are robust or fragile when there are small changes to the conditioning informa- tion set. Using the extreme bounds analysis of Leamer (1983, 1985), they find that the estimation results are quite fragile. Sala-i-Martin, Doppelhofer, and Miller (2004) report somewhat more optimistic results in cross-country studies by examining an approximation to the cumulative distribution function of the estimators. Even so, their results find that only eighteen of the sixty-seven explanatory variables (or only 27 percent) are robustly correlated with measures of economic growth. Crain and Lee (1999) report similar results for cross-region analysis of U.S. states.

As for the more specific impact of fiscal policies, the generally held presumption is that higher taxes tend to lower economic growth because of their distortionary effects, because they tend to discourage the creation

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

488                                                                                     Public Finance Review 39(4)

of new firms and jobs, and because they inhibit investment. For example, it is widely held that higher income taxes will lower the rate of growth because they lower the net return to private investment and make invest- ment activities less attractive. Even so, there is at least some recognition that the government expenditures financed by tax revenues might provide superior public services, thereby making a higher tax area more, not less, attractive. For example, high public spending on infrastructure investment (e.g., transportation, communications, and education) is generally believed to increase growth rates. Indeed, Mofidi and Stone (1990) find that state economic performance depends on the interrelationship between state taxes and the programs on which the taxes are spent. They also find that state and local taxes have a negative effect on growth when the revenues are devoted to transfer payments but that expenditures on health, education, and public infrastructure have positive effects on growth.

It should also be noted that public sector ‘‘institutions’’ are also likely to affect economic growth. For example, Persson and Tabellini (1992) outline a theory that relates different political incentives and political institutions to growth. They conclude that income inequality is ‘‘bad’’ for growth in democ- racies, while land concentration is bad for growth everywhere. Relatedly, there is much empirical work that suggests that factors such as the number of local governments, the presence of TELs, and the political composition of the governing party affect (and are in turn affected by) fiscal policies.

In sum, existing results for the effects of fiscal policies on state economic growth are quite variable. The next section presents our approach to esti- mating the impacts of fiscal (and other) factors on economic growth.

Method, Data, and  Specifications

Method

The specification of growth regression models is complicated by the likeli- hood that the observed value of per capita income in state s in year t (ys,t) includes an unknown and unknowable measurement error eys,t; that is, eys,t denotes any random disturbance in the observed value of per capita income, so that observed ys,t   is related to ‘‘true’’ yts,t  by the relationship:

ys;t  ¼ yt    þ y

s;t

ð3Þ

where  the  superscript  t denotes  the  true  but  unobservable value  that excludes all measurement errors. Consequently, the error term in the result- ing growth regression consists of a combination of the error term associated

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

es  ¼ [ey s,t; eg s,t] for any state s as N (0, Oe ), where 0 denotes a conformably dimensioned array of zeros and Oe   denotes the covariance matrix for all model and measurement errors associated with state s. As a result, asymptotically maximum likelihood estimators can be obtained  using ODR. Unlike LLS methods, which minimize the sum of the squared vertical deviations between the dependent variable and the fitted ‘‘line,’’  ODR methods minimize the orthogonal (or perpendicular) deviations from the fitted line. See Boggs, Byrd, and Schnabel (1987) and Boggs et al. (1988) for detailed discussions of ODR methods, including an algorithm that can be used to calculate ODR coefficient estimates. In fact, we use weighted ODR methods, which allow for heterosce- dastic variances within and between and observations, for nonzero covariances within observations (even though covariances between observations are iden- tically zero), and for nonlinearity in the explanatory variables and/or the esti- mated coefficients.

Monte Carlo experiments that we have conducted show that measure- ment errors do in fact significantly affect LLS estimates from growth regressions. These experiments also show that ODR methods noticeably improve bias and mean square error results, even when the assumptions imposed on the solution are wrong. In particular, the results show that LLS estimates designed to test the ‘‘convergence hypothesis’’ have a strong ten- dency to be more negatively biased than the same coefficient estimated using ODR methods. Furthermore, for all but one of the more than eighty pairs of median bias examined in our Monte Carlo study, the bias in the LLS estimator is larger than that in the ODR estimator by more than a factor of two. These experiments demonstrate that the measurement errors inherent in growth regression data are important and should be considered explicitly when attempting to analyze the factors that affect economic growth. All of our Monte Carlo results are available on request.

Data

Response variable. The response variable in our basic specifications is the annual growth rate in per capita personal income for the forty-eight contiguous states over the period 1947 through 1997. Personal income is computed by the U.S. Department of Commerce Bureau of Economic Analysis as the sum of wages and salaries, other labor income, proprietors’ income, dividends, inter- est, rent, and transfer payments, less personal contributions for social insur- ance. The main difference between state personal income and gross state product involves the treatment of capital income. Personal income includes corporate net income only when individuals receive payment as dividends; gross state product includes corporate profits and depreciation. In addition,

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

Alm and Rogers                                                                                                      491

gross state product attributes capital income to the state in which the business activity occurs, while personal income attributes capital income to the state of the asset holder. Neither measure includes capital gains.

The personal income of a state is defined as the income received by the residents of the state. However, the estimates of wages and salaries, other labor income, and personal contributions for social insurance are based mainly on source data that are reported by place of work, not by place of residence. Accordingly, an adjustment for residence, equal to the net inflow of the earnings of wage and salary workers who are interstate commuters, must be estimated so that the place-of-residence measures of earnings and personal income can be derived. Descriptive statistics for the resulting growth rates in per capita income data are provided in table 1.

Explanatory variables. We have assembled more than 130 explanatory vari- ables for the analysis. These variables can be grouped into five categories: revenues, expenditures, demographics, geographics, and national. The first three categories include values that vary by state and by year; the fourth includes values that vary by state but not by year; and the fifth includes values that vary by year but not by state. All variables in each category are identified in tables 2 through 6. Note that the first two letters of each variable name denote the category (e.g., rv for revenues, sp for spending or expenditures, dm for demographics, ge for geographics, and us for national).

The revenue and expenditure variables are of obvious interest to policy makers. The various tax sources (e.g., individual or corporate income, sales, and property taxes) have implications for the returns to individuals and firms from their activities. Similarly, how a state chooses to spend these revenues is also important. For example, expenditures on education can have a direct impact on growth by producing a more capable work force and are also likely to have an indirect effect related to the perception of the importance that the state places on education.

The revenue, expenditure, demographic, and geographic variables have been recorded at a relatively fine level of detail. Composite variables have then been constructed from these values. For example, the data include values for general and select sales taxes (variables rvTXSALgen and rvTXSALsel, respectively), as well as total sales taxes (variable rvTXSALtot) computed from their sum. Similarly, the geographic category includes dummy variables to indicate natural resources in the state (such as variables geMNau, geMNfe, and geMNcoal), as well as a dummy variable to indicate the occurrence of one or more of these resources (variable geMN). The revenue and expenditure variables are included either as per capita values, as a percent of per capita income, or as a percent of total tax revenue. All explanatory variables are lagged one year.

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

492                                                                                     Public Finance Review 39(4)

Table 1. State Per Capita Income Growth Rates

Variable                           Minimum      Medium    Maximum      Mean

Standard

Deviation

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

Annual values for state revenues and expenditures, as well as for all demographic, geographic, and  national  variables,  are  available  for the period 1947 through 1997. Annual estimates for total state plus local reven- ues and expenditures are not recorded prior to 1959 (and not all variables are available until 1977); as a result, our combined state and local analysis is for the shorter periods 1959 through 1997 (and 1977–1997). The data are obtained from various issues of the Book of the States, the Statistical Abstract of the United States, Current Population Reports (Series P60), State Government Finances reports, and the World Almanac; some vari- ables are obtained from personal communication with staff at the U.S. Bureau of the Census. The primary source for the estimates of total earnings and employment by place of work is the ES-202 series from the U.S. Bureau of Labor Statistics.

Specifications

Baseline regression. Levine and Renelt (1992), Sala-i-Martin (1997), Crain and Lee (1999), and others have explored the sensitivity of regression

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

494                                                                                     Public Finance Review 39(4)

Table 2. State Revenue Variables

Variable Name           Class            Subclass                    Qualifier                     Units

rvTXTOTAL     Tax             None                  Total taxes from all sources

Real U.S.$

rvTXCON          Tax             Consumption    Total (sales þ income)        Real U.S.$

rvTXINCtot        Tax             Consumption    Income: total

(individual þ corporate)

Real U.S.$

rvTXINCcor       Tax             Consumption    Income: corporate              Real U.S.$

rvTXINCind       Tax             Consumption    Income: individual              Real U.S.$

rvTXSALtot       Tax             Consumption    Sales: total

(general þ selective)

Real U.S.$

rvTXSALgen      Tax             Consumption    Sales: general                       Real U.S.$ rvTXSALsel     Tax          Consumption    Sales: selective               Real U.S.$ rvTXPROP                    Tax                    Wealth                                 Property          Real U.S.$ rvTXSEV                Tax          Other       Severance   Real U.S.$ rvTXNEC   Tax             Other                 Not elsewhere classified  Real U.S.$ rvTRFtot          Transfers    Federal Total transfer          Real U.S.$ rvTRFedu           Transfers    Federal              Transfers for education      Real U.S.$ rvTRFhwy       Transfers    Federal Transfers for highways          Real U.S.$ rvTRFnec   Transfers    Federal              Not elsewhere classified  Real U.S.$

results by comparing outcomes against the results from a set of ‘‘core’’ vari- ables. However, there is little agreement on which variables should be included in the core set of regressors. For example, Levine and Renelt use the investment share of gross domestic product (GDP), the initial level of real GDP per capita, the initial secondary school enrollment rate, and the average annual rate of population growth as the core variables in their cross-country analysis. Sala-i-Martin uses the level of income, life expec- tancy, and primary school enrollment rate as the core variables for his cross-country sensitivity analysis.

In our work, we choose a set of six core regressors, and our baseline regression (denoted Regression A) includes only these variables. These core regressors are

     usGRW: the U.S. per capita income growth rate

     usINF: the U.S. inflation rate

     usFUELpp: the average U.S. producer price of fuels

     gePOLstate: a dummy variable equal to 1 if statehood was attained before 1800, and 0 otherwise

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

Abstract

What  factors influence state  economic growth? This article uses annual state (and local) data for the years 1947 through 1997 for the forty-eight contiguous states  to  estimate  the  effects of a large number  of factors, including taxation and expenditure policies, on state economic growth. A special feature of the  empirical work  is the  use of orthogonal  distance regression (ODR) to deal with the likely presence of measurement error in many of the variables. The results indicate that the correlation between state (and state and local) taxation policies is often statistically significant but also quite sensitive to the specific regressor set and time period; in contrast, the  effects of expenditure  policies are much more  consistent. Of some interest, there is moderately strong evidence that a state’s political orienta- tion has consistent and measurable effects on economic growth; perhaps, surprisingly, a more ‘‘conservative’’ political orientation is associated with lower rates of economic growth. Finally, correction for measurement error is essential in estimating the growth impacts of policies. Indeed, when mea- surement error is considered via ODR estimation, the estimation results do not support conditional convergence in state per capita income.

1    Department of Economics, Andrew Young School of Policy Studies, Georgia State University

2    Chief  State  Economist for  the  Department  of Administration, Division  of Budget and

Planning, for the State of Nevada

Corresponding Author:

Email: jrogers@budget.state.nv.us

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011

484                                                                                     Public Finance Review 39(4)

0.5

0.4

0.3

0.2

0.1

0

–0.1

–0.2

–0.3

–0.4

–0.5

Figure 1. Difference between individual state and U.S. average growth rates,

1947–1997

Keywords

fiscal policies, regional economic growth, orthogonal distance regression

Introduction

The average annual growth rates of per capita income for the individual forty-eight contiguous U.S. states over the last half of the twentieth century range from 1.73 percent to 3.15 percent. Six states have annual growth rates that exceed the national growth rate by more one-half of a percentage point at least half the time. Another four states have annual growth rates that are more than one-half of a percentage point less than the national growth rate at least half the time. Figure 1 identifies the states with the highest and the lowest average growth rates.

Why is this issue important? In 1947, the median real value of per capita income for the forty-eight contiguous states was just under $7,500 (in 1997 dollars). If, over the fifty-year period from 1947 to 1997 the annual growth

Downloaded from pfr.sagepub.com at THAMMASAT UNIVERSITY on June 17, 2011