[ Main page of Report | Contents of Report ]
Organized into two subsections, section III, Findings and Results, lays out the report's findings. Subsection A provides an overview of trends and patterns in social welfare spending observed over the period from 1977 to 2000, based primarily on Census data. Subsection B presents the results from the 50-state econometric model.
[ Go to Contents ]
The connections between state fiscal capacity and spending on social welfare have undergone vast changes in recent years. Between 1977 and 2000, major shifts occurred in how much was spent on social welfare programs, how states allocated funding across different social welfare functions, and how state fiscal capacity was related to these developments. To compare changes in spending patterns for rich and poor states over time, we classified all of the states plus the District of Columbia into four quartiles with respect to their fiscal capacity, as measured by real per capita income averaged over the 24-year period. Exhibit III-1 shows a map indicating the states in each of the quartiles, from the richest states in Quartile 1 (shown as the lightest colors) to the poorest in Quartile 4 (shown as the darkest colors). States with the lowest per capita income are generally found in the South and the West, while the wealthier states are located in the Northeast, around the Great Lakes, and on the Pacific Coast.
Exhibit III-1
States by Fiscal Capacity Quartile
| Selected Characteristics | Quartile 1 | Quartile 2 | Quartile 3 | Quartile 4 |
|---|---|---|---|---|
| Mean per capita personal income, 1977-2000 (2000 dollars) | $24,794 | $21,387 | $19,242 | $17,058 |
| Mean percentage of people under federal poverty level, 1977-2000 | 11.0 | 11.0 | 13.2 | 17.1 |
| Mean percentage of population unemployed, 1977-2000 | 3.1 | 3.0 | 2.7 | 3.2 |
| Median population density (persons per square mile), 1977-2000 | 338 | 89 | 57 | 52 |
Exhibit III-1 also displays information about states in these different quartiles, including average real per capita income, poverty rates, unemployment (per capita), and population density-all characteristics included in the econometric models as independent variables. As well as differing in income levels, these poor states also had higher poverty rates and lower population densities (they were generally more rural) though, surprisingly, they did not have much higher per capita unemployment rates.(16)
This classification is simplistic. If one calculated fiscal capacity quartiles every year, some states would shift from one to the other over time. However, these simple four-level rankings showed unexpected stability. Most of the states would not shift quartiles even if allowed, and where changes did occur, the changes were nearly always to an adjacent quartile, not a jump of two or three.
When averaged over the entire period from 1977 to 2000, per capita spending on social welfare was positively correlated with state fiscal capacity, as shown in the chart at the top of Exhibit III-2.(17) A similar pattern exists for spending per poor person, shown at the bottom of Exhibit III-2. When public hospital payments were included, the wealthiest 13 states (Quartile 1) spent an average of $825 per capita (in 2000 dollars) over this time period, while the poorest 12 states (Quartile 4) spent $630 per capita. When payments to public hospitals were excluded, mean per capita spending by states in the wealthiest quartile was $639, while average spending by states in the poorest quartile was $407.
Average spending per capita
This difference in spending between rich and poor states resulted largely from differences in states' spending of their own tax revenues, as shown in the chart at the top of Exhibit III-3. Federal grants exerted a complex effect on inequalities in state spending. In dollar terms, federal funding actually increased state differences with respect to fiscal capacity because the Quartile 1 received higher grants per capita than the other quartiles. The richest quartile of states, for example, spent an average $371 per capita from federal sources, while the poorest quartile spent $339 (when public hospital payments are included). However, because poor states spent less money overall on social welfare, that $339 constituted a large proportion (83 percent) of their total spending on such programs. By contrast, the $371 per capita from federal sources spent by the richest quartile of states made up a much smaller share (58 percent) of their total social welfare budgets. That is, more federal money went to rich states than to poor states, but poor states relied more heavily on the federal government to support their social programs.
The chart at the bottom of Exhibit III-3 also shows that state fiscal capacity bore a similar relationship to state spending on non-social welfare functions. Again, the differences were due to how much of their own revenues states spent. However, federal spending played a smaller role in this component of state budgets. Although federal spending averaged over two-thirds (69%) of all spending on social welfare functions, federal grants typically made up only about one-eighth (13%) of total state spending on non-social welfare functions.
When we disaggregated social welfare spending into more specific categories, the relationships between state fiscal capacity and state spending became more complex. As the chart at the top of Exhibit III-4 demonstrates, the poorest states (Quartile 4) showed much lower levels of spending on cash assistance and non-health social services than did wealthier states.(18) Levels of spending on Medicaid were also highest for the rich states in Quartile 1; spending levels were lower, albeit similar, across Quartiles 2 through 4. By contrast, the poorest states spent the most per capita on public hospitals. State fiscal capacity was, in sum, strongly related to spending on cash assistance, moderately correlated with spending on non-health social services, and least correlated (even negative for hospital payments) with spending on health or medical assistance.
Average spending per capita
Average spending per poor person
Because social programs were intended mostly to help low-income people, per capita spending levels might fail to capture differences in the degree to which states met social needs. To understand spending from this perspective, we compared spending levels per poor person in each state. Using this measure, differences among rich and poor states were greater and more consistent, as shown in the chart at the bottom of Exhibit III-4. Disparities between the top and bottom quartiles were even larger for spending on cash assistance and other non-health social services; and Medicaid expenditures and public hospital payments showed stronger and positive relationships to fiscal capacity.
Averages over time cannot show changes in the relationships between state fiscal capacity and spending on social welfare programs. Yet those relationships changed enormously between 1977 and 2000. To see these developments, we traced changes in average spending levels in each of these quartiles and for each category of social welfare spending.
Exhibit III-5 compares trends in spending for cash assistance, Medicaid, and non-health social services, with spending levels adjusted for inflation using the GDP price deflator. Per capita spending on cash assistance was lowest in the poorest quartile of states throughout the 24-year period, as shown in the graph at the top of Exhibit III-5. However, states with different fiscal capacities began to converge in their spending on cash assistance in the mid-1990s. This convergence came about as cash assistance spending in richer states (i.e., states in Quartiles 1, 2, and 3) declined, while states in Quartile 4 saw virtually no change in their already low spending levels. State fiscal capacity thus became less correlated with spending on cash assistance programs at the end of the 1990s when compared to the early 1990s and especially the late 1970s.
The 1990s also produced major changes in state spending on Medicaid, as depicted in the graph in the middle of Exhibit III-5. At the beginning of the decade, Medicaid payments grew rapidly for states in all quartiles, though the greatest growth occurred among poor states. By 2000, per capita spending in Quartile 4 was about 10 percent higher than per capita spending in Quartiles 2 and 3. The relationship between state fiscal capacity and spending on Medicaid thus declined in strength. Like the trends for cash assistance, spending levels of rich and poor states converged. But unlike cash assistance, spending on medical assistance programs saw an upward convergence as previous low spenders joined high spenders, not a downward convergence as high spending states came down to the level of low spending states.
Per capita spending on non-health social services showed no such convergence, as illustrated in the graph at the bottom of Exhibit III-5. Instead, it revealed growing differences between states of different fiscal capacities. The poorest states showed the lowest per capita spending throughout, since 1980. Yet the real separation occurred in the 1990s. Although spending on other social welfare grew in all states during the 1990s, the wealthier states in the top three quartiles showed rapid growth in such spending from 1997 through 2000 at the same time states in the poorest quartile increased their spending much more slowly. Non-health social services include expenditures for child welfare, child care, energy assistance, and many other social services, as well as the costs to public agencies of administering such programs, cash assistance, and Medicaid.
The strong growth in Medicaid spending might result in part from higher levels of inflation for health services. To gauge the importance of health-specific inflation rates, Exhibit III-6 shows trends in spending for medical assistance using the CPI for health care. With this inflationary adjustment, average per capita spending on Medicaid still rose, though less strongly, as illustrated in the graph at the top of Exhibit III-6. Spending on public hospitals actually declined over this period for the three wealthier quartiles, as depicted in the graph at the bottom of Exhibit III-6. However, the poorest quartile showed a slight increase in per capita expenditures. Thus, as was the case for Medicaid, the poorest states increased their spending on health-related functions more than wealthier states did in the 1990s.
Average Per Capita Spending on Cash Assistance Adjusted with GDP Price
Deflator over Time by Income Quartile
Average Per Capita Spending on Cash Assistance Adjusted with GDP Price
Deflator over Time by Income Quartile
Average Per Capita Spending on Non-health Social Services Adjusted with
GDP Price Deflator over Time by Income Quartile
Average Per Capita Spending on Public Hospitals Adjusted with CPI Medical
Index over Time by Income Quartile
Average Per Capita Spending on Public Hospitals Adjusted with CPI Medical
Index over Time by Income Quartile
These increasingly complex relationships between state fiscal capacity and various forms of social welfare spending were not found in trends and patterns for non-social welfare spending, as indicated in Exhibit III-7. Throughout the 24-year period, spending per capita outside the social welfare area was greatest in the richest states and lowest in the poorest states, and growth in spending after adjusting for inflation showed none of the dramatic short-run changes found in cash assistance and Medicaid.
Average Per Capita Spending on Non Social Welfare Adjusted with GDP Price
Deflator Over Time by Income Quartile
Finally, states have changed how they spend their social welfare dollars as well as the way they fund those functions. Exhibit III-8 indicates the percentage of social welfare spending supported from state and local governments' own-source revenues rather than federal grants. Over the entire period, states in all quartiles showed a long-run decline in their reliance on own-source revenues. But the declines were greatest among the wealthier states, which, at any point, relied less on federal grants and more on their own-source revenues than did poor states. Thus, some convergence occurred, especially in the 1990s, as poor states in Quartile 4 slightly increased their proportionate use of own-source revenues, while the richer states decreased their reliance on such sources and increased their dependence on federal dollars.
Percent of Spending From State/Local Sources over Total Social Welfare
Spending By Income Quartile
These many changes combined to produce major realignments in a relatively brief period in the spending profiles of rich and poor states. Exhibit III-9 shows evidence of these shifts by tracking the percentage of total public welfare spending in different program functions in the wealthiest and the poorest quartiles (Quartiles 1 and 4). Medicaid absorbed a much larger share of the budgets of both rich and poor states. Cash assistance spending fell in all states, though most precipitously in rich states, thereby reducing differences between rich and poor states between 1980 and 2000.
Non-health social services declined slightly as a component of overall social welfare spending in all states. But the biggest change with respect to this category of mostly non-health social services was the growing disparity between rich and poor states. Although poor states spent slightly more on such services as a percentage of their total social welfare budgets in 1990, by 2000, they had spent a much smaller percentage of their budget on these non-health social services. In just 2 decades, the components of state social welfare budgets changed in fundamental ways. They moved away from cash assistance and toward health services, and poor and rich states became increasingly different in the role of non-health social services in their total social service budgets.
These findings reinforce the need to estimate different econometric models for different types of social welfare spending, because each type shows distinct dynamics. They also pose a challenge: the relationship between state fiscal capacity and spending on different public goods seems to be mutable, suggesting a need to dig deeper after the econometric analysis and determine what dynamics are accounted for by the models and what changes are not. Finally, the trends show that poor states in particular have seen a radical transformation in the package of social welfare functions they support, a development that argues for special attention to their social welfare budgets and the factors affecting them.
Rich states (Q1) are those in the 1st or highest quartile on fiscal capacity;
poor states (Q4) are those in the 4th or lowest quartile
[ Go to Contents ]
Using conventional ordinary least squares, we estimated the 50-state econometric model on pooled time series and cross-section data on state and local spending for the 24 years from 1977 to 2000. We conducted standard tests for auto-correlation of residuals, which sometimes constitutes an issue for time series analysis.(19)
The model and its estimates had two general purposes. First, the model estimates allowed us to assess the magnitude and statistical significance of variables of interest, such as state fiscal capacity as measured by per capita personal income, federal grants, and other determinants of state and local spending. Second, the model estimated state effects used in the comparisons among rich and poor states in general and among the six poor states selected for further analysis. We report an overview of the results in the text of this report. More detailed estimation results appear in Appendix A.
We estimated a number of models,(20) but for purposes of this discussion we will focus on a single preferred model in which we allow for flypaper effects by including both state personal income and federal grants as separate variables, and we attempt to measure state and local needs through the poverty, unemployment, and population density variables. Both state and year dummies are included.
Below, we present the results, first for the regressions estimated over all states and then for the regressions estimated separately for each quartile defined by the historical average of real PCPI.
Exhibit III-10 displays the regression results for the five regressions with dependent variables defined as respective categories of per capita state and local spending (CA - cash assistance, M - Medicaid, NSS - non-health social services, PH - public hospitals, and non-social welfare - NW) for all states.(21) Below the estimated coefficients, t-statistics appear in parentheses. A t-value greater than approximately 1.96 indicates statistical significance at the .05 level, and a t-value greater than approximately 2.44 indicates statistical significance at the .01 level.
| Variable | CA | M | NSS | PH | NSWS |
|---|---|---|---|---|---|
| Adjusted R-Squared | 0.87 | 0.91 | 0.87 | 0.89 | 0.96 |
| Constant | 61.09** (3.12) |
-79.00** (3.57) |
-52.70* (2.13) |
155.96** (8.05) |
2356.25** (8.20) |
| Per capita personal income | -0.0039** (6.89) |
0.0033** (5.24) |
0.0060** (8.47) |
0.0003 (0.50) |
0.1236** (15.07) |
| Federal grants for non-social welfare | 0.0049 (0.64) |
-0.0017 (0.20) |
0.0543** (5.71) |
0.0396** (5.32) |
-0.1222 (1.11) |
| Federal grants for social welfare | 0.0943** (8.82) |
0.3278** (27.18) |
0.0052 0.3800 |
0.0843** (7.99) |
0.0280 0.1800 |
| Population Density | 0.06** (5.89) |
-0.05** (4.89) |
-0.09** (7.21) |
-0.06** (6.59) |
0.65** (4.48) |
| Unemployment per capita | 836.09** (7.13) |
702.63** (5.31) |
-84.84 (0.57) |
193.19 (1.67) |
8295.98** (4.83) |
| Poverty per capita (moving average) | -30.10 (1.05) |
-95.68** (2.95) |
-45.79 (1.26) |
-76.00** (2.67) |
320.19 (0.76) |
| CA means Cash Assistance, M means
Medicaid, NSS means Non-health Social Services, PH means Public Hospital
Spending, and NSWS means Non-Social Welfare Spending. T-statistics are in parentheses. ** Significant at the 1% level. * Significant at the 5% level. |
|||||
As shown in Exhibit III-10, the linear effects of per capita personal income on per capita social welfare spending after we control for federal grants and need are positive, and statistically significant for Medicaid and non-health social services. However, the effects are negative and statistically significant for cash assistance and statistically insignificant for public hospital spending.(22) The impact of per capita personal income on per capita non-social welfare spending is larger, but we expected this outcome because non-social welfare spending is much larger than the individual components of social welfare spending.(23)
When we examine the effects of federal grants on the components of social welfare spending for all states in Exhibit III-10, we find, unsurprisingly, that grants for non-social welfare exert weak and statistically insignificant effects on cash assistance and Medicaid spending, but such grants exert much stronger and statistically significant effects on non-health social services and public hospital spending. This result might constitute evidence of a positive income effect of non-social welfare grants(24) on the latter two categories of non-health social services and hospital spending. The federal grants for social welfare have strong, positive, and generally statistically significant effects on spending for cash assistance, Medicaid, and public hospitals. The effect of federal grants on Medicaid is particularly strong, indicating the attractiveness to the states in matching federal Medicaid dollars.
Exhibit III-10 also shows the results for the three main indicators of need for social welfare spending (i.e., poverty per capita, unemployment per capita, and population density) for all states. The negative signs on the poverty variable seem surprising and difficult to explain. Possibly the measures of fiscal capacity and federal grants are insufficient to capture the state's perceived resources and poverty proxies for available resources (in a negative direction). Also, high poverty states might resist spending because of an omitted unobserved variable correlated with poverty. The poverty variable was statistically significant and negative only for Medicaid and public hospital spending.
Unemployment per capita had the expected positive sign for all categories of social welfare spending except non-health social services, and the effect was statistically significant for cash assistance, Medicaid, and public hospitals. Population density was statistically significant for all categories of social welfare spending, including public hospitals, but was positive for cash assistance and negative for Medicaid, non-health social services, and public hospitals.
Exhibit III-11 shows results for the same regression model estimated separately for each of the 4 quartiles defined by mean real per capita personal income. Below, we summarize the general findings from this analysis by quartile and spending category for the explanatory variables of greatest interest: (1) fiscal capacity, (2) federal grants, (3) need variables, (4) state unemployment rates, and (5) state effects.
| Variable | Q 1 CA |
Q 2 CA |
Q 3 CA |
Q 4 CA |
Q 1 M |
Q 2 M |
Q 3 M |
Q 4 M |
|---|---|---|---|---|---|---|---|---|
| Adjusted R-Squared | 0.90 | 0.74 | 0.87 | 0.37 | 0.93 | 0.88 | 0.89 | 0.89 |
| Constant | 231.38** (6.22) |
-288.79** (3.21) |
-156.30** (5.23) |
77.23 (1.49) |
-38.83 (0.82) |
42.12 (0.53) |
-202.85** (3.42) |
-164.18* (2.22) |
| Per Capita Personal Income | -0.0078** (9.07) |
0.0068** (2.84) |
0.0012 (1.14) |
-0.0042 (1.88) |
0.0032** (2.93) |
0.0001 (0.03) |
0.0053** (2.54) |
0.0084** (2.62) |
| Federal Grants for Non-social Welfare | 0.0284** (2.66) |
0.0119 (0.41) |
0.0222 (1.80) |
-0.0339 (1.53) |
-0.0077 (0.56) |
0.0081 (0.32) |
0.0161 (0.66) |
0.0311 (0.99) |
| Federal Grants for Social Welfare | 0.1547** (7.33) |
0.0527 (1.78) |
0.1526** (10.19) |
-0.0275 (1.70) |
0.3545** (13.21) |
0.2074** (8.01) |
0.3467** (11.65) |
0.3422** (14.77) |
| Population Density | 0.054** (4.10) |
0.40** (3.43) |
1.13** (8.75) |
1.01** (3.04) |
-0.04* (2.38) |
-0.04 (0.36) |
0.35 (1.36) |
-0.04 (0.07) |
| Unemployment Per Capita | 841.88** (3.17) |
1277.17** (4.28) |
386.67** (2.78) |
139.93 (0.59) |
1081.63** (3.20) |
77.50 (0.30) |
451.51 (1.64) |
382.24 (1.14) |
| Poverty Per Capita (moving average) | -42.40 (1.01) |
499.69** (3.50) |
164.87** (3.50) |
94.24 (1.51) |
-130.43* (2.44) |
94.71 (0.76) |
84.89 (0.91) |
-4.42 (0.05) |
| Q means Quartiles CA means Cash Assistance, M means Medicaid, NSS means Non-health Social Services, PH means Public Hospital, NSWS means Non-social Welfare. T-statistics are in parentheses. ** Significant at the 1% level. * Significant at the 5% level. |
||||||||
| Variable | Q 1 NSS |
Q 2 NSS |
Q 3 NSS |
Q 4 NSS |
Q 1 PH |
Q 2 PH |
Q 3 PH |
Q 4 PH |
Q 1 NSWS |
Q 2 NSWS |
Q 3 NSWS |
Q 4 NSWS |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Adjusted R-Squared | 0.86 | 0.87 | 0.86 | 0.80 | 0.87 | 0.90 | 0.89 | 0.95 | 0.95 | 0.94 | 0.96 | 0.95 |
| Constant | 98.43 (1.78) |
-100.50 (1.13) |
25.85 (0.39) |
100.31* (1.95) |
43.61 (1.17) |
367.17** (5.71) |
-91.59 (1.48) |
-221.98** (4.69) |
405.14 (0.47) |
4118.30** (5.39) |
2099.03** (5.48) |
1797.11** (3.68) |
| Per Capita Personal Income | 0.0063** (4.89) |
0.0097** (4.10) |
0.0049* (2.09) |
0.0048* (2.14) |
0.0025** (2.94) |
-0.0056** (3.29) |
-0.0007 (0.33) |
0.0024 (1.18) |
0.1496** (7.57) |
0.0529** (2.62) |
0.0798** (5.86) |
0.0050 (0.23) |
| Federal Grants for Non-social Welfare | 0.0307* (1.93) |
0.0063 (0.22) |
0.1842** (6.75) |
-0.0155 (0.70) |
0.0415** (3.87) |
-0.0017 (0.08) |
0.0308 (1.21) |
-0.0303 (1.50) |
-0.4988* (2.03) |
-0.3084 (1.27) |
0.5041** (3.19) |
0.7086** (3.40) |
| Federal Grants for Social Welfare | 0.0401 (1.28) |
0.0406 (1.39) |
0.1402** (4.22) |
-0.0461** (2.85) |
0.805** (3.80) |
0.0561** (2.65) |
0.1205** (3.88) |
0.0281 (1.89) |
-0.3581 (0.74) |
0.7400** (2.95) |
0.0314 (0.16) |
0.0767 (0.50) |
| Population Density | -0.07** (3.39) |
-0.31** (2.71) |
-1.59** (5.54) |
-0.34 (1.03) |
-0.06** (4.90) |
-0.30** (3.59) |
0.32 (1.18) |
2.87** (9.39) |
0.81** (2.69) |
2.00* (2.05) |
0.88 (0.53) |
17.17** (5.45) |
| Unemployment Per Capita | 310.34 (0.79) |
94.33 (0.32) |
196.59 (0.64) |
-452.51 (1.93) |
-218.42 (0.82) |
594.61** (2.79) |
-420.39 (1.46) |
372.15 (1.73) |
15519.00** (2.54) |
11648.00** (4.60) |
5923.35** (3.32) |
-5225.86* (2.35) |
| Poverty Per Capita (moving average) | -64.62 (1.03) |
40.52 (0.29) |
-258.74** (2.47) |
89.02 (1.43) |
-36.41 (0.86) |
13.34 (0.13) |
-69.47 (0.71) |
56.02 (0.98) |
1461.13 (1.51) |
29.16 (0.02) |
-2192.36** (3.62) |
162.03 (0.28) |
| Q means Quartiles CA means Cash Assistance, M means Medicaid, NSS means Non-health Social Services, PH means Public Hospital, NSWS means Non-social Welfare. T-statistics are in parentheses. ** Significant at the 1% level. * Significant at the 5% level. |
||||||||||||
d. Fiscal Capacity Effects
From Exhibit III-11, we can see that when the states are separated by quartiles based on average per capita personal income, differences across quartiles in the estimated effects emerge. For example, for cash assistance and non-health social services, the effect of personal income for the richer states (Quartiles 1 and 2) is statistically significant(25), while for the poorer states (Quartile 3 and 4), the effect is nearer zero for both categories and statistically insignificant for cash assistance. The reverse occurs with respect to effects on Medicaid, for which the effects of personal income is near zero for the richer states, although statistically significant and positive for Quartile 1, and larger and statistically significant for the poorer states. For public hospitals, the effect of personal income is positive and statistically significant only for the richest states.
In general, the regression analysis confirmed that personal income was an important factor in causing social welfare spending disparities among rich and poor states. These disparities based on sample means were reported, for example, in Exhibit III-2. When we controlled for the effects of non-income explanatory variables, the differences in per capita spending on social welfare across rich and poor states narrowed but did not disappear. For example, if Quartile 1 states were assigned the same average income as those in Quartiles 2, 3 and 4, the regression model for Quartile 1 predicted that Quartile 1 per capita spending on social welfare would fall by $60, $98, and $137, respectively. This amounted to reductions of 7 percent, 12 percent, and 17 percent.
e. Federal Grant Effects
When we estimated the regression separately for each quartile, the signs of the effects of federal grants reverse or the coefficients are estimated with less statistical significance for the poorest states (Quartile 4), indicating weak grant income effects for those states, as shown in Exhibit III-11. In particular, for the social welfare spending categories, federal grants have positive and statistically significant effects only for Medicaid, suggesting a substitution toward Medicaid spending as federal grant income rises.(26) For the other quartiles, the results are mixed, as also shown in Exhibits III-11; however, the effect of social welfare grants on Medicaid is generally large, positive, and statistically significant.
f. Need Effects
When we repeated the regressions separately by quartile, poverty continued to exert consistently negative signs only for the richest states (Quartile 1) and generally had more positive signs for the other quartiles. However, the only statistically significant and positive effects of the poverty variable occurred for cash assistance in quartiles 2 and 3.
The effect of unemployment per capita on the various categories of social welfare spending was positive and statistically significant generally only for the richer states (Quartiles 1 and 2), although an exception is cash assistance in Quartile 3. For the richer states, the unemployment effects were positive and statistically significant for cash assistance and Medicaid in Quartile 1 and for cash assistance and public hospitals in Quartile 2.
The strongest unemployment effect occurred in the cash assistance category for which we observed statistically significant and positive effects for Quartiles 1, 2, and 3. This result might constitute a kind of "caseload effect," but it does not occur in Quartile 4, perhaps because the spending levels are so low they decline no further with lower unemployment.(27) But the spending for Quartile 4 apparently also declines no further as unemployment increases. Instead, the spending levels seem "stuck" and independent of the state of the economy.
The effects of population density seem mixed with positive and statistically significant signs on the effects for cash assistance for all quartiles and statistically significant and negative signs on the effects for the remaining categories of social welfare spending, including public hospitals in the richer states (Quartiles 1 and 2). The results were more mixed for Quartiles 3 and 4 with no particular pattern discernable.
To confirm that the effects estimated for the unemployment per capita variable were truly state labor market effects and to get a better sense of how spending changed in response to unemployment changes alone, we estimated simple linear relationships between spending per capita in each category and the state unemployment rate with and without per capita personal income as an additional explanatory variable. In the regressions of per capita spending on the state unemployment rate alone (but including state effect dummies) for all states, as shown in Exhibit III-12, spending for cash assistance and public hospitals was positively related to unemployment, in other words an anti-cyclical effect,(28) but spending on Medicaid and other social welfare was negatively related to unemployment, a pro-cyclical effect.
| Variable | Overall | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| CA | M | NSS | PH | NSWS | ||||||
| Adjusted R-Squared | 0.83 | 0.83 | 0.56 | 0.76 | 0.68 | 0.85 | 0.86 | 0.87 | 0.84 | 0.95 |
| Constant | 22.23** (4.46) |
57.91** (7.10) |
158.15** (14.15) |
-178.30** (12.95) |
161.04** (18.85) |
-120.92** (12.54) |
241.19** (49.31) |
305.19** (39.40) |
5577.77** (45.42) |
937.94** (8.45) |
| State Unemployment Rate | 3.07** (8.29) |
1.66** (3.70) |
-11.34** (13.67) |
2.03** (2.69) |
-10.70** (16.87) |
0.51 (0.96) |
3.66** (10.07) |
1.11** (2.62) |
-156.62** (17.17) |
27.86** (4.57) |
| Per Capita Personal Income | -0.00** (5.49) |
0.01** (30.68) |
0.01** (36.72) |
-0.00** (10.37) |
0.17** (52.47) |
|||||
| Q means Quartiles CA means Cash Assistance, M means Medicaid, NSS means Non-health Social Services, PH means Public Hospital, NSWS means Non-social Welfare. T-statistics are in parentheses. ** Significant at the 1% level. * Significant at the 5% level. |
||||||||||
To see how much of the effect of unemployment was operating through the per capita personal income variable, we included per capita personal income in the regression. These results can be seen also in Exhibit III-12. Over all states, the unemployment rate exerted a positive effect on spending for all categories of social welfare spending, and the effect was statistically significant for all categories except non-health social services. Thus, the presence of the per capita personal income variable has eliminated the negative effect for Medicaid and non-health social services spending observed in the regression on the unemployment rate alone. The fact that unemployment and personal income are negatively correlated suggests the negative effects observed in Exhibit III-12 are due to the absence of a control for income. Across all states, then, unemployment seems to do an effective job of picking up positive need effects on spending, except for other non-health social welfare, with personal income held constant. This finding is generally consistent with the results reported in Exhibit III-10 with other explanatory factors (e.g., poverty, population density, federal grants, and year dummy variables) in addition to personal income included in the regression.
One result of the 50-state model was the estimation of unexplained variance in spending across different states. These state effects were estimated intercepts or constant terms for each of the states in the econometric models. They may be interpreted as general dispositions of states-averaged across the entire period, 1977-2000-to support certain types of spending after controlling for the linear effects of fiscal capacity, social needs, federal grants, and population density.(29)
In their original form, the state intercepts, or effects, were difficult to interpret. To make them easier to understand, we standardized them with respect to the mean and standard deviation of the state effects. That is, for each set of estimated state effects-one set of 50 for each dependent variable, such as cash assistance or Medicaid-the mean of the 50 state effects was set at zero and the standard deviation was set at 100. Thus, if a state's effect for Medicaid was 2 standard deviations above the mean of the 50 state effects that particular effect was scored as 200. If the state's effect for cash assistance was 1/4 of a standard deviation below the mean for all states, then that effect was scored -25.
Exhibit III-13 shows the standardized versions of the estimated state effects for all of the states. The larger and more positive the number, the greater the tendency of the state to spend on that particular category of public function over the entire time period, 1977-2000, after controlling for the linear effects of the other independent variables, including per capita personal income, per capita grants, and the various need variables. For example, even after controlling for these factors, Alaska shows a strong additional propensity to spend on cash assistance and non-social welfare and a tendency to spend less on Medicaid and public hospitals relative to other states. New York and Minnesota, however, show additional propensities (again, compared to other states) to spend more than predicted by the econometric model on all social programs.
Despite the good fit of the models to the data, these state effects show variation in their spending on different types of social programs. For example, the difference between the 25th and 75th percentiles in the state effect estimates for cash assistance is $49 per capita, a large amount compared to the mean per capita spending for all states (averaged over all years, 1977-2000) of $82.
| Cash Assistance | Medicaid | Non-heath Social Services | Public Hospitals | Non-social Welfare | ||
|---|---|---|---|---|---|---|
| Quartile 1 | Alaska | 273 | -171 | 19 | -190 | 611 |
| California | 275 | -47 | -2 | 30 | -18 | |
| Connecticut | 123 | 92 | -14 | 52 | -106 | |
| Delaware | -24 | -141 | 75 | -96 | 9 | |
| Hawaii | 207 | -39 | -122 | -57 | 33 | |
| Illinois | 102 | 1 | 14 | -52 | -62 | |
| Maryland | 27 | 85 | -51 | -48 | -63 | |
| Massachusetts | 79 | 292 | 152 | 48 | -64 | |
| Nevada | -59 | -98 | -140 | 64 | -10 | |
| New Hampshire | -10 | 77 | 115 | -134 | -61 | |
| New Jersey | -43 | 190 | 117 | 39 | -95 | |
| New York | 189 | 110 | 309 | 133 | 38 | |
| Quartile 2 | Colorado | -3 | -109 | -16 | -30 | -3 |
| Florida | -64 | 12 | -107 | 55 | -39 | |
| Kansas | -3 | -47 | -93 | 23 | -9 | |
| Michigan | 147 | -79 | 122 | 8 | -27 | |
| Minnesota | 125 | 169 | 129 | 46 | 18 | |
| Ohio | 44 | 48 | 45 | -23 | -61 | |
| Oregon | -10 | -167 | -19 | -63 | 35 | |
| Pennsylvania | 72 | -43 | 232 | -74 | -56 | |
| Rhode Island | 4 | 260 | 208 | 1 | -70 | |
| Virginia | 32 | -25 | -112 | -1 | -51 | |
| Washington | 88 | -58 | -49 | -34 | 54 | |
| Wisconsin | 46 | 107 | 94 | -65 | -1 | |
| Wyoming | -59 | -116 | -173 | 210 | 140 | |
| Quartile 3 | Arizona | -36 | -21 | -77 | -85 | 34 |
| Georgia | 6 | 52 | -103 | 221 | -31 | |
| Indiana | -97 | 48 | -44 | 55 | -61 | |
| Iowa | -3 | -6 | 56 | 90 | -7 | |
| Maine | 32 | 128 | 96 | -156 | -16 | |
| Missouri | -34 | -50 | -93 | -16 | -68 | |
| Nebraska | -34 | -43 | 24 | 66 | 75 | |
| North Caroline | -20 | -61 | -49 | 52 | -27 | |
| North Dakota | -92 | 37 | 61 | -134 | 47 | |
| Oklahoma | -48 | -3 | -37 | 37 | -32 | |
| Tennessee | -101 | 8 | -35 | 72 | -7 | |
| Texas | -52 | -43 | -122 | 35 | -33 | |
| Vermont | 39 | -98 | 52 | -197 | 30 | |
| Quartile 4 | Alabama | -71 | 12 | -117 | 181 | -24 |
| Arkansas | -129 | 37 | -40 | -10 | -53 | |
| Idaho | -85 | -68 | -58 | -3 | -23 | |
| Kentucky | -78 | 85 | 17 | -87 | -32 | |
| Louisiana | -125 | -39 | -33 | 153 | -15 | |
| Mississippi | -169 | 30 | -37 | 172 | -23 | |
| Montana | -92 | -116 | -21 | -145 | 36 | |
| New Mexico | -52 | -98 | -5 | -7 | 43 | |
| South Carolina | -113 | 15 | -33 | 137 | -17 | |
| South Dakota | -59 | -68 | -54 | -125 | 16 | |
| Utah | -31 | -83 | -58 | -81 | 68 | |
| West Virginia | -111 | 41 | -21 | -68 | -17 |
When these estimated state effects are analyzed, they show that state fiscal capacity interacts with program area (i.e., the relationship with fiscal capacity varies with program area). Exhibit III-14 shows these variations by displaying the average state effects, in their standardized versions, for states of different fiscal capacities, using our basic four quartiles. The relationship between state effects and fiscal capacity are compared across four different program areas: cash assistance, Medicaid, non-health social services, and public hospitals. We should note that the differences in state effects are most important, not the absolute values (e.g., whether they are negative or positive, that is, above or below the average state effects across all states).
Exhibit III-14 indicates that states of different fiscal capacities still vary in their long-run spending patterns even after controlling for the linear effects of annual changes in states' per capita personal income, as the 50-state model does. For example, the wealthiest states (Quartile 1) spent on average about $180 more per capita per year on cash assistance than did the poorest quartile (Quartile 4). A consistent and positive, albeit weaker, relationship between fiscal capacity and average state effects is also evident in spending on non-health social services.
Health-related expenditures show a different pattern. With respect to Medicaid, the average state effects for the richest states were higher than for states in the other quartiles, but the differences among the three less wealthy quartiles were small. The relationship between fiscal capacity and spending on public hospitals was actually reversed. Per capita spending was lowest among the richest states and highest among the poorest states. After controlling for the linear effects of annual changes in fiscal capacity and other variables, as the 50-state model does, poor states still spent less on cash assistance and other social welfare, while their spending on health-related programs was not much lower and sometimes higher than the amount wealthier states spent.
Poor states, on average, thus revealed greater support for spending on health-related programs than for spending on non-health programs. One possible consequence of this pattern was a weaker statistical relationship among poor states in their support across different program areas. Among non-poor states (i.e., states in the first three quartiles for fiscal capacity), tendencies to spend on different social welfare functions were, for the most part, either positively correlated with each other or not correlated at all, suggesting that no major tradeoff existed among these states between their financial support for one type of social welfare and their support for another.
We can see these relationships in the first column of Exhibit III-15, which shows the bivariate correlation coefficients between the state effect estimates for cash assistance, Medicaid, non-health social services, and public hospitals. For the 38 states in the first three quartiles of fiscal capacity, the correlations were generally either positive or especially small. The strongest correlation was between Medicaid and non-health social services, though a moderate relationship also existed between cash assistance and non-health social services. Only the state effects for public hospitals showed a slight negative relationship to state effects for other types of spending.
| Types of Spending | Pearson correlations between estimated state effects
for different types of spending, by state fiscal capacity |
|
|---|---|---|
| Non-poor States | Poor States | |
| Cash Assistance vs. Non-health Social Services | .33 | -.15 |
| Cash Assistance vs. Medicaid | -.01 | -.51 |
| Cash Assistance vs. Public Hospitals | -.23 | -.51 |
| Non-health Social Services vs. Medicaid | .50 | .13 |
| Non-health Social Services vs. Public Hospitals | -.22 | -.40 |
| Medicaid vs. Public Hospitals | .17 | .33 |
| Number of cases | 38 | 12 |
By contrast, among the 12 poorest states, the correlations among these spending tendencies were more likely to be negative. Cash assistance was negatively correlated with both types of health-related functions, Medicaid and public hospitals. Non-health social services was also negatively related to spending on public hospitals and, albeit weakly, cash assistance. On the other hand, the poor states showed a slightly stronger relationship between the two types of health program areas. We can see an example of the contrasting structure of these relationships in Exhibit III-16, which shows the scatterplots between the state effects for cash assistance and Medicaid-separately for poor and non-poor states. No correlation existed between the estimated state effects for non-poor states, but a clear negative relationship existed among the poor states.
Scatterplot between state effects estimated for Medicaid and cash assistance,
only states in Quartiles 1, 2, and 3 in fiscal capacity (i.e., wealthier
75%)
Scatterplot between state effects estimated for Medicaid and cash assistance,
only states in Quartile 4 in fiscal capacity (plus Arizona, because it is
one of the study states and is near the cutoff point between Quartiles 3
and 4)
More generally, low fiscal capacity states divided between those that put money into health programs and little else and those that put money into other programs, especially cash assistance. As Exhibit III-16 shows, the former were southern and border states, including Mississippi, Arkansas, West Virginia, and South Carolina. Poor western states, including Utah and New Mexico, showed greater levels of support for cash assistance. Whatever the reasons for these differences among poor states, such states clearly showed divisions in their spending patterns across different functions. Poor states, unlike wealthy states, seemed to choose or specialize in one or another type of social program area. Their packages of social programs were, thus, more particularized as well as smaller. Although knowing how much a wealthy state spent in one social program area often helped us know how much it spent in another area, the same was untrue for poor states.
i) Effects of Fiscal Capacity
Although per capita income generally had the expected positive effect on spending, notable differences occurred between rich and poor states. When we analyzed the sample separately by quartile, we found the income effects on cash assistance, non-health social services, and public hospitals much more consistently larger and statistically significant for the rich states than for the poorer states. On the other hand, the income effects on Medicaid were larger and more positive for the poorer states than for the richer states. This finding suggests that when personal income rises in the richer states, the states are more likely to increase social welfare spending across the board, and when income rises in the poorer states, spending is likely to occur largely on Medicaid.
ii) Effects of Federal Grants
Although federal grants largely increased state and local spending on social welfare, the effects on federal grants were hardly noticeable for the poorest states (Quartile 4), except for a positive effect on Medicaid. The grant effects were most apparent on payments to Medicaid, suggesting the importance of the Medicaid matching funds.
iii) Effects of Need Variables, Including Unemployment
Estimating a stable needs function that would predict well state and local spending proved impossible. That poverty seemed negatively correlated with spending in a number of spending categories was puzzling, particularly for the richest states. Although the sign on the per capita unemployment was much more likely to be positive than the sign on the poverty variable, the statistically significant positive unemployment effects on spending seemed generally confined to the richer states. The strongest positive effects of unemployment occurred on cash assistance spending. This result might constitute a kind of caseload effect, but it fails to occur in Quartile 4 for the poorest states. The effects of population density on social welfare spending were generally mixed, but we estimated a number of coefficients to be statistically significant.
The poorer states seem to have less protection against adverse unemployment effects, and their needs are more likely to go unmet in a downturn. When we more closely evaluated the effect of state unemployment on spending using the state unemployment rate, we found that cash assistance and Medicaid spending were positively related to the unemployment rate with no income control, particularly for the richer states. However, for non-health social services, the coefficient on unemployment was consistently negative and statistically significant across quartiles and largely statistically insignificant for public hospital spending. We conclude that the total effect of a rise in unemployment is likely to be a cutback in spending for non-health social services across all states with increases in spending for cash assistance and Medicaid in richer states.
iv) State Effects
Stable differences among states in their spending patterns persisted even after controlling for the linear effects of fiscal capacity, need, federal grant, and other independent variables. These propensities to spend (i.e., estimated state effects) suggested that state fiscal capacity was more strongly related to non-health expenditures than to health-related expenditures. They also suggested that the basic structure of expenditures was different in rich and poor states. In wealthier states, spending on each social welfare function was more likely to be positively related or largely independent of spending on other social welfare functions. In the poorest states, however, spending on each social welfare function, such as Medicaid, was more likely to be negatively related to spending on other functions, such as cash assistance. These negative relationships between expenditures by poor states result in some interesting differences among the states with respect to their spending patterns, one difference being the regional split between western and southern states in their relative emphasis on cash assistance and Medicaid. This is discussed further in the next section.
In sum, the multivariate econometric analyses suggested the following:
[ Go to Contents ]
(16) Unemployment rates per capita are lower than unemployment rates usually reported. The former rate is the number of unemployed divided by the total population in the state, while the latter is the number of unemployed divided by the number of persons seeking jobs.
(17) Social welfare spending in this analysis includes no spending on public hospitals.
(18) The Non-health Social Services category includes support activities, such as operational payments for administrative workers; payments for child care, foster care, low-income energy assistance, and social services for the physically disabled programs; Social Services Block Grant-funded programs; and temporary shelters and services for the homeless. The category also includes direct payments to private vendors for non-health services and commodities and for the provision, construction, and maintenance of governmentally owned and operated welfare institutions and nursing homes.
(19) The Durbin Watson statistics indicated some autocorrelation but insufficient, in our opinion, to adjust the analysis.
(20) The model structures are described in Appendix A. Generally, we estimated three generic models with sub-models within each model. In Model 1D, we introduced need variables into the structure for Model 1C. Model 1D is the model described in the body of this report. Models 2 and 3 experimented with specifications that explicitly introduced price effects of federal grants using the method McGuire (1978) recommended. . For reasons related to both the lack of appropriate data to estimate price effects, econometric difficulties related to the same variable appearing on both sides of the equations, and the model statistical results being weak and inconsistent (including many "wrong" signs), we focus primarily on Model 1. Although no explicit grant price effects occur in Model 1, we expect the combination of state and year dummies to pick up much of the variation of grant price effects across states and over time.
(21) For definitions of these variables in terms of Census categories, see Exhibit II-1 and the accompanying discussion.
(22) The relationship between fiscal capacity and spending on cash assistance seems to be more complex, however. When the equations were estimated for each fiscal capacity quartile, only the wealthiest quartile (Quartile 1) showed a significant and negative relationship. Also, when we examined the estimated state effects from the cash assistance equation, we found the state effects were positively correlated with our four quartiles of states, quartiles that reflected long-run differences in state fiscal capacity. One possible interpretation of these inconsistencies was that our indicator of fiscal capacity-per capita personal income-had two counteracting effects on cash assistance: one long-run effect and one short-run effect. Higher state income in the long run might have encouraged states to adopt more generous cash assistance policies (in terms of higher maximum benefits and need standards, greater earnings disregards, and less stringent asset limits). However, in the short run, increases in per capita income during economic downturns would have decreased welfare rolls as recipients left voluntarily for jobs or became disqualified because their incomes were too high. This latter dynamic (the wage effect) might have been particularly strong among richer states because they tended to have more generous benefit policies, and their cash assistance rolls were thus more likely to include large numbers of working families, whose income fluctuations exerted a greater, countercyclical effect on state spending. (Maximum benefit levels, one indicator of the relative generosity of cash assistance policies, averaged $575 for a three-person family in 2000 for the states in Quartile 1 (the wealthiest). The median maximum benefit levels for Quartiles 2, 3, and 4 were $429, $292, and $277 respectively.)
(23) For example, the means of the variables are cash assistance ($82), Medicaid ($151), non-health social services ($125), public hospitals ($115), and other non-social welfare ($3,422). The small relative value for cash assistance spending leads to the odd result that the coefficient on PCPI for cash assistance spending is close to zero but nonetheless statistically significant because its standard error is also extremely small.
(24) We should note, however, that the non-social welfare federal grants (i.e., intergovernmental revenue) include federal grants for the public hospital category of spending because the Census classifies public hospital spending as spending for non-social welfare. Census data do not permit disaggregating federal grants in more detail than the broad categories of federal grants for social welfare spending and other federal grants.
(25) It should be noted, however, that the effect is negative in quartile 1 for cash assistance. This mirrors the negative impact of personal income on cash assistance found in the overall regression, as reported in Exhibit III-10.
(26) The large positive effect might also reflect the increase in federal spending for Medicaid in response to increases in state matching funds. McGuire (1978) reviews the arguments why federal grants can be viewed as exogenous in a model such as this one. However, his major assumption is that the federal government acts through its grants to induce a target level of state spending, implying that if state matching funds increase, no resultant change in federal spending would occur as a result because the state behavior was anticipated. This assumption might prove untrue in practice, and reverse causality effects might occur.
(27) One possibility is that in the poorest states, less movement on and off cash assistance occurs as the state of the economy improves or worsens.
(28) A pro-cyclical effect means spending moves in the general direction of the economy. When the economy improves, so does spending and vice versa. Conversely, an anti-cyclical effect occurs when spending moves contrary to the direction of the economy.
(29) For more on state effects, see footnote 7 in subsection II.B.1.
Top of Page | Contents
Main Page of Report | Contents
of Report
Home Pages:
Human Services Policy
(HSP)
Assistant Secretary for Planning and Evaluation
(ASPE)
U.S. Department of Health and Human Services
(HHS)