A Dynamic Model of State Budget Outcomes Under Divided Partisan Government
James E. Alt
Harvard University
Robert C. Lowry
Michigan State University
Prepared for delivery at the Annual Meeting of the Midwest Political Science Association,
Chicago, April 23-25, 1998. We are grateful to Karen Ferree and Eric Dickson for research
assistance. The research was supported by the National Science Foundation under grant SBR 9223638. This is a preliminary draft; please do not quote without the authors’ permission.
Abstract
This paper describes a model of partisan fiscal adjustment where policy is made in a bicameral
legislature subject to a veto by an independent executive. We show how changes in fiscal policy
depend not just on the configuration of parties, but also on veto institutions and on which party or
parties was responsible for the previous budget. In our model, the legislative party is unable to
shift fiscal policy all the way to its preferred point in one step, but can take advantage of repeated
shocks to shift fiscal policy toward its preferred target slowly, probably over several budget
cycles. We describe how legislatures can use shocks to shift expenditures and revenues and how
this means that over time one can observe persistence of past partisan targets or slower and faster
shifts toward new targets. We show also how the amount of change in policy should vary
according to how control is configured and how patterns of control interact with institutions.
Empirical sections of this paper lay out a specification for estimating a model which is capable of
incorporating these features, and report estimates based on taxes and spending in 35 nonsouthern
states from 1954-1989. While the results are neither strong nor robust, estimated speeds of
adjustment to targets line up as expected, with unified governments showing faster adjustment
than divided, and the party targets go in the right direction, with Democrats apparently targeting a
larger share of state incomes for the public budget than Republicans. Interestingly, Republicans
react much more strongly to budget surpluses by reducing revenues than do Democrats.
5RNKV DTCPEJ The 1995 budget crisis in Congress produced a situation in which no formal budget passed
and the government operated under a series of continuing resolutions which carried over the
provisions of the previous year’s budget. The result, long periods of uncertainty and conflict,
focused attention on the political problem of fiscal adjustment in a bicameral system when parties
are polarized in terms of having different partisan targets for the scale of taxes and public
spending.
In this paper we describe a model of partisan fiscal adjustment where policy is made in a
bicameral legislature subject to a veto by an independent executive. We focus particularly on
“split branch” government, where the legislature and the executive are controlled by different
parties. We show how changes in fiscal policy depend not just on the configuration of parties, but
also on veto institutions and on which party or parties was responsible for the previous budget. In
our model, the legislative party is unable to shift fiscal policy all the way to its preferred point in
one step. However, it is able to take advantage of repeated shocks to shift fiscal policy toward its
preferences slowly, probably over several budget cycles.
Our basic model and a few extensions show that budgets made under divided government
can depend on a lot of features that we often think of as procedural details, like the identity of the
“veto player,” the last legislative vote needed to override an executive veto, or the exact level of
spending that would occur if no new agreement is reached. This is not because control of the
legislature is unimportant. In fact, in our model, the legislature’s proposal power makes it the
only branch capable of unilaterally imposing adjustments in fiscal policy. However, the model
shows that there are circumstances when partisan change in policy instruments can be slow
5RNKV DTCPEJ because of institutional features. Parties, the past, and institutions all matter, and this paper
makes the interconnections between them systematic.
The paper describes what has come to be called a “partisan” model of policy, one in which
politicians of different parties, reflecting differing preferences in the electorate, promise and
produce different policies. A complete partisan model has two sides, a “policy function” which
describes policy outcomes and a “vote function” which describes the electoral reaction. This
paper extends important aspects of the policy function that were neglected in our earlier work
(see Alt and Lowry, 1994). It extends the partisan model to a context in which its usual intuition,
the predictable big swings in policy of the “Westminster” model of party government, don’t
emerge because transitions of power are rarely between party configurations conferring full
control of separated institutions.
Nevertheless, the underpinnings of partisan models shine
through, namely that systematic, persistent differences in the preferences of separate groups of
politicians come to be reflected in predictably different policy actions.
We start by describing the simplest cases in detail, to describe how legislatures can use
shocks to shift expenditures and revenues and how this means that over time one can observe
persistence of past partisan targets or slower and faster shifts toward new targets. We show how
the amount of change in policy should vary according to how control is configured and how
patterns of control interact with institutions. We then discuss the importance of both past party
control and the partisanship of the legislator whose vote is critical to overriding a gubernatorial
veto, as well as the nature of the governor's veto powers as prescribed by state laws.
Furthermore, while our earlier work looked at the long term behavior of a statistical model
in order to estimate the values of partisan targets, here we look at the short term transitions of
power, in a formal model to see how policy should move, assuming that parties have different
5RNKV DTCPEJ targets. If voters wish to change fiscal policy by changing party control of government, how
much change can they expect to see? The answer depends on whether elections change the
partisan identity of the governor and one or both branches of the legislature, as well as veto rules
and the level of spending that would occur even without new appropriations.
The empirical sections of this paper lay out a specification for estimating a dynamic
partisan model of budgets in the states. We report estimates based on taxes and spending in 35
nonsouthern states from 1954-1989. While at this writing we are only able to make a start, we
derive an estimating model which is capable of incorporating the features discussed above. In
estimates from this model, the speeds of adjustment to targets line up as expected, with unified
governments showing faster adjustment than divided, and the party targets go in the right
direction, with Democrats apparently targeting a larger share of state incomes for the public
budget than Republicans. Interestingly, Republicans react much more strongly to budget surpluses
by reducing revenues than do Democrats. We give some preliminary comments on the robustness
of these results and point out how other conjectures can be tested within the same model.
Predicting Budgets Under Divided Government
We focus in this section on divided government, which has been a feature of state politics
for a long time. Key describes it as widespread (due to malapportionment) in the 1930s. Fiorina
(1992) documents how the growth of Democratic legislative control in the 1950s led the
incidence of divided control to grow, outside the South. Over the forty or so years from 19501990, divided government was more common in the 35 nonsouthern states than was unified
control by either of the major parties: the figures for the period are 32 per cent split branch (one
party controls the executive and the other party controls the legislature), 28 per cent unified
5RNKV DTCPEJ Republican control, 22 per cent unified Democratic control, and 20 per cent split control of the
legislature.
Context
Our earlier work distinguished unified (one party controls all branches) from divided
government at the state level, and showed that on the whole unified governments react quicker to
recessionary shocks than do divided governments (and particularly divided legislatures). Stylizing
Alt and Lowry (1994) freely, we also showed that certain balanced budget laws were effective in
compelling adjustments in fiscal policy, at least under unified control of government. We also
showed that the parties differed in the long run target shares of personal income that they wished
to take in revenues for public purposes.1 While both parties seemed equally willing to spend
federal contributions to the states, the Democrats acted as though they had an underlying goal of
taking about 10 per cent of state personal income in taxes, while for Republicans the
corresponding target was 5 per cent.
The other two findings, laws work and split legislatures involve delays regardless of laws,
suggested problems of accountability which we take up in two related papers. In Lowry and Alt
(1997), we inquire why the balanced budget laws are effective, especially since we never observe
courts involved in their enforcement. We show that the laws are effective because the operation of
bond markets gives politicians an incentive to maintain orderly fiscal policies, namely, lower operating
costs of government (and thus more funds to spend on other things) in those cases where unforeseen
economic circumstances compel running a deficit in the short term. Because the laws make it easier for
market participants to discern the nature and intentions of a state's fiscal agents, they are willing to give
5RNKV DTCPEJ a discount to the debt of a state with stringent laws precisely when it is most wanted, in a fiscal crisis.
To keep the discount intact, fiscal agents adopt sound practices. In Lowry, Alt and Ferree (1998), we
present evidence that voter reactions to fiscal policy outcomes are contingent on government
institutions and partisan control: voters punish Republican gubernatorial candidates following
unexpected increases in spending and revenues; they punish the incumbent governor’s party in
legislative elections for fiscal deficits; and the electoral-policy connection is stronger when one
party controls both branches of the government.
So we have evidence for long-run party differences in targets; evidence that laws requiring
budget balance matter and the likelihood that they are underwritten and enforced by market
inducements; and evidence that voters hold parties accountable for fiscal policy. What we need to
provide next is a model of how changes in party control produce changes in fiscal policy through
the budget process, in a way which is sensitive to the fact that changes in the American states are
rarely between unified control of government by different parties. Rather, they are more likely to
involve changes to or from various forms of divided government.
Unified Party Government
In our previous work, we argued that the executive can either approve or veto changes
proposed by the legislature, but must exercise the veto power within a certain (usually limited)
time period after the legislature has acted, or the legislative proposal takes effect. The budget
process that we envisage is as follows. It begins (often around January 1) with the Governor
filing a budget request. In our model, this is a partisan document, which forecasts revenues
(usually by forecasting income growth and setting taxes to produce the desired revenues) and sets
This is an old idea in partisan models, finding implementation in various forms in the work of
Frey (1978), Hibbs (1987), and Alt & Chrystal (1983). All these works abstract away from
5RNKV DTCPEJ out spending, so that the aggregate budget consumes the share of state personal income which is
the desired target level of the party of the Governor. Clearly legislatures bargain over and modify
budgets, but in our simple model of the aggregate size of the budget, if a majority of each
chamber of the legislature is of the same party as the Governor, each chamber passes the
Governor’s budget and he signs it.2
If the same configuration governed in the last period, the income share reflected in the
budget will not change, as revenues can always be adjusted to offset past shocks.
If the
configuration is new, and the previous budget reflected different preferences, nothing stops a
unified government from moving directly to its preferred outcome, though in practice an empirical
model might allow this to take more than one budget cycle. Given the evidence that Democrats
are the high demand party, unified Democrats are likely to inherent budgets that are “too small”
given their preferences, whereas unified Republicans are likely to inherit budgets that are “too
large.” This implies
Hypothesis 1. Short run changes in revenues and spending under unified Democratic
government should be positive or zero on average; those under unified Republican
government should be negative or zero.
Split Branch Government
If parties differ in their preferred levels of spending and one party controls the executive
and the other controls both chambers of the legislature, then the expected spending level for any
partisan differences in the resulting composition of spending which no doubt also exist.
Several states require supermajorities in each chamber to pass budgets. In most of these, party
control has been so one-sided in the last four decades that these requirements do not change our
results. However, in three states, California, Ohio, and Illinois, there are a large number of years
in which neither party “controls” one or both legislative chambers in terms of having enough
partisans to pass a budget without bipartisan cooperation.
5RNKV DTCPEJ given year cannot be optimal for both parties, and will probably not be the same as the level which
would be chosen by either party in unified government. In this section we assume that each party
is internally homogeneous, party-specific preferences are common knowledge, the governor
possesses an “all-or-nothing” veto, and the legislative majority is not veto proof. We show that
adjustments to expected spending or revenues will depend on (i) which party controls which
branch, (ii) whether the previous year’s budget resulted in a surplus or a deficit, and (iii) the
location of the reversion point, i.e., the levels of spending and revenues that would occur in the
absence of a budget agreement. We address the consequences of heterogeneous parties and
alternative veto rules in the next section.
We analyze this situation using a spatial model that differs somewhat from the model in
Alt and Lowry (1994). Figures 1a-1d show four scenarios that are discussed below.
The
horizontal axis represents general spending as a share of state income; the vertical axis represents
general revenues as a share of income; and the 45 degree line is the set of points such that
spending equals revenues. We assume that all fiscal year budgets must lie on the balanced budget
line.3 Both parties have a most-preferred budget, with Democrats preferring to spend and tax
more than Republicans. Party-specific preferences for points away from their most-preferred
budgets are represented by elliptical indifference curves. The ellipses tilt up and to the right,
indicating that spending and revenues are positive complements (Hinich and Munger, 1997).
Intuitively, both parties are more willing to tolerate spending levels above (below) their most
We abstract away from whether or not the budget is planned to balance exactly or includes a
surplus. We treat any planned surplus as fixed, and not adjusted according to current
circumstances or spending desires.
5RNKV DTCPEJ preferred level if revenues are also above (below) their most preferred level, so that the budget is
at least balanced.
[Figure 1 about here]
In each diagram, the point labeled “t1” represents the budget passed for the previous fiscal
year, while “r1” is the previous year’s actual outcome, net of any purely transitory shocks. Point
r1 might be above the balanced budget line if, for example, the economy did better than expected
last year, generating excess revenues, and the new level of income is expected to continue for the
coming fiscal year. As in our earlier paper, we assume that any lasting shocks (the difference
between r1 and t1) affect revenues only, but we no longer assume that r1 is also the reversion
point in the event of a budget stalemate. Rather, we assume that some portion of spending will
not occur unless new appropriations are passed and signed by the governor. It seems reasonable
that federal transfers, spending financed by dedicated taxes and fees, and entitlements will be the
subject of a continuing resolution, so that the maximum difference between r1 and the reversion
point “r2” consists of non-entitlement spending financed out of state general fund revenues. We
have no precise data on how large this figure is, although the percentage of general spending
(including entitlements) financed out of general fund revenues varied from 24 percent in Wyoming
to 70 percent in Massachusetts in 1996. (Statistical Abstract of the United States, 1997). For
expositional purposes, we simply assume that r2 lies somewhere between r1 and the vertical axis,
indicating that some, but not all, spending will be discontinued absent a budget agreement. Note
that revenues at r2 are the same as at r1, because taxes and fees continue to be collected with or
without new appropriations.
Figure 1a shows the case of a Democratic governor and Republican legislature in the year
following a surplus, so that r1 lies above the balanced budget line. The reversion point, r2,
5RNKV DTCPEJ represents an even larger surplus, since some spending would be discontinued absent an
agreement but revenues would continue to be collected. The Republican legislature realizes that
the Democratic governor will cast a sustainable veto against any budget that leaves the Governor
worse off than the reversion point. The legislature thus sets the new budget at the point labeled
“target” where the balanced budget line intersects the Governor’s indifference curve through r2.
The legislature prefers this point to r2, while the Governor is no worse off. The result is a
planned decrease in both spending and revenues relative to the previous year’s budget target (t1),
and the actual outcome (r1).
Figure 1b shows the same governor and legislature faced with a deficit due to a negative
revenue shock. As we have drawn the figure, the shift from r1 to r2 eliminates almost all of the
projected deficit, so the new target is very near the reversion point.
This does not mean,
however, that the legislature cuts all non-entitlement spending financed from general fund
revenues. Points on the diagram represent aggregate general spending and revenues, and the new
budget may also include cuts in entitlement programs or programs financed by dedicated taxes
and fees. The result is a cut in both spending and revenues compared to the previous budget
target, although planned revenues are slightly higher than the previous year’s actual outcome.
Figure 1c shows the case of a Republican governor, a Democratic legislature and a
surplus. Planned spending and revenues both increase relative to t1, but planned revenues may not
increase relative to r1. Note, however, that whereas revenues clearly decrease in Figure 1a under
a Republican legislature, they either increase or decrease by a smaller amount with a Democratic
legislature.
Figure 1d shows a Republican governor and a Democratic legislature faced with a
prospective deficit. Note that there are two different reversion points (r2 and r2’) that produce
5RNKV DTCPEJ the same budget target. Despite the fact that the Democratic legislature passes a budget that it
prefers to the reversion point, the new spending target is actually lower than the previous year’s
target and outcome. The revenue target is lower than the previous year’s target, but higher than
the previous year’s actual outcome. Finally, if the reversion point lies far enough to the left (not
shown), the Democratic legislature may actually be able increase both spending and revenues.
6JKU TGUWNV ECP QEEWT DGECWUG VJG 4GRWDNKECP IQXGTPQT KU KPFKHHGTGPV DGVYGGP C DWFIGV VJCV JCU
OWEJ NQYGT URGPFKPI VJCP JGT OQUVRTGHGTTGF DWFIGV CPF C NCTIG UWTRNWU QT C DWFIGV VJCV JCU
JKIJGT URGPFKPI VJCP JGT OQUVRTGHGTTGF DWFIGV DWV KU QP VJG DCNCPEGF DWFIGV NKPG UGG
Shepsle and Bonchek, 1997, pp. 123-125).
The exact outcomes in the scenarios depicted by Figures 1a-1d can be manipulated by
shifting the location of t1, r1 and r2 relative to the most-preferred budgets, but the general pattern
of the predictions is clear. In the case of split branch government, it is the legislature that controls
the policy target. Republican legislatures are able to ratchet down spending targets following
either a surplus or a deficit. The new revenue target under a Republican legislature will also be
less than the previous year’s revenue target, but may be slightly more than the previous year’s
actual revenues following a deficit. Democratic legislatures can ratchet up spending following a
surplus, but can only hold the line against further cuts following a deficit. The revenue target
under a Democratic legislature should also exceed the previous year’s target (though not
necessarily the outcome) following a surplus, but the prediction following a deficit depends on the
exact location of the reversion point. The asymmetry between Democrats and Republicans results
from the fact that the reversion point always involves a decrease in spending relative to the
previous year’s budget.
Finally, although the legislative party sets the new budget, it will
generally be forced to accept some compromise with the governor in order to avoid a veto. This
5RNKV DTCPEJ implies that the speed of transition following a change in partisan control will be slower under
split branch government than under unified government.
Split Branch Extensions
Past Party control. Budget targets depend not only on the preferences of the current group in
control, but also on which party most recently controlled fiscal policy. This may be represented in
our model by the location of t1 and r1, the previous year’s budget target and outcome. Suppose
that we currently have a Democratic governor, a Republican legislature, and the previous year
ended in a deficit, as in Figure 1b. If last year’s budget reflected (was passed by) a unified
Republican government, t1 should be equal to the Republicans’ most-preferred budget. The
reversion point following a deficit will include lower revenues and lower spending than the
previous budget target (r2 will be below and to the left of t1), so the Republican legislature will
simply adopt the previous year’s budget, and the Democratic governor will sign it. If the previous
budget was set by a unified Democratic government, however, then t1 will be equal to the
Democrats’ most-preferred budget. Now if there is a deficit, the Republican legislature should be
able to impose a budget that lies in between the two most-preferred budgets. If another deficit
occurs in the next year, then the Republican legislature may be able to ratchet revenues and
spending down further, but probably by a lesser amount.
Heterogeneous parties and veto override points. In general, when the legislature under split
branch government inherits a budget different from its most preferred budget, it can take
advantage of shocks that knock the budget out of balance to move policy toward its ideal by
trading a little balance for a little size. Where the legislative party is homogeneous and veto
proof, however, the legislative party immediately proposes its own ideal point. Empirically, this
should look just like unified government.
5RNKV DTCPEJ Where parties are heterogeneous, the budget target depends on the preferences of the
median legislator and either the governor or the last vote needed in the legislature to override an
executive veto, whichever is closest to the median. If the governor’s most-preferred budget is
closest to that of the median legislator, then the results are as in Figures 1a-1d, except that now
we substitute the median legislator for the unitary legislative party. Otherwise, a comparable
game occurs between the last legislator needed to override a veto and the median legislator.
Line-Item Veto. In many states, the governor possesses some version of line-item veto authority
(Advisory Commission on Intergovernmental Relations, 1991). Carter and Schap (1990) argue
that in many circumstances the effect of a line-item veto on aggregate spending is indeterminate,
because the line-item veto alters the composition rather than the aggregate level of spending. That
is, the reduction in expenditure on one line item may be more or less than offset by changes in
expenditure on another line item. In our terms, each line item has its own marginal override
legislator, but there is no guarantee that the budget comprised of all just-overridden line-item
vetoes is itself approved by either the governor or the legislator possessing the veto override vote
on aggregate revenues and spending.
However, if we define an item veto as the governor’s power to counterpropose
proportionate budget cuts, but not increases, an interesting further case arises. If we have split
branch government and the legislature is controlled by Republicans, the case will work out exactly
the same as the all-or-nothing veto, since the Democratic governor lacks the power to propose
increases. If the legislature is the high demand party, it can propose increases and the executive
can fine tune proportionate proposed cuts, so their powers become more symmetric. Empirically,
this could mean that the reaction to a shock will be slowest where there is a line-item veto,
5RNKV DTCPEJ polarized parties, and a low-demand executive, that is, a standoff between a Democratic
legislature and a strong Republican executive.
Empirical Predictions for Split Branch Government
In order to test all of the facets of this model empirically, we would need data on budget
targets, actual budget outcomes less purely transient shocks, reversion points, and most-preferred
points for the executive, median legislator in each chamber, and marginal override legislator in
each chamber. Data limitations require that we test the following, greatly simplified, hypothesis:
Hypothesis 2. Under split branch government, the legislative party sets the new budget,
subject to an executive veto. Changes in revenues and spending should be greater (more
positive) when the Democrats control the legislature than when the Republicans control
the legislature, but the speed of transition will be less than under unified government.
Split Legislature Government
In the case of split legislature government, different parties control each chamber, so it
may be a while before a proposal is made. However, any budget which satisfies the Governor’s
legislative partisans should satisfy the Governor, so once a proposal is made there is no reason to
expect a veto. Clarke (1998) contends that the governor’s party will have more leverage in
conference committee, and thus control policy. We have been unable to generate this result from
any sort of formal model, however. In general, the outcome of this sort of bargaining depends on
the relative costs and benefits to each party of reaching agreement, the two parties’ discount rates
(how desperate are they for a deal?), and their beliefs about each other’s costs, benefits and
discount rates (Alesina and Drazen, 1991). While we are unable to generate a clear prediction for
split legislature government, we take as our null hypothesis Clarke’s assertion, and add that the
5RNKV DTCPEJ rate at which budget targets change following a change in government will be slower if different
parties control different chambers of the legislature than if there is unified party government:
Hypothesis 3: In the case of split legislature government, fiscal policy targets are set by
the governor’s party. Changes in revenues and spending will be greater (more positive)
with a Democratic governor than a Republican governor, but the speed of adjustment is
slower than under unified party government.
Empirical Specification
To turn these conjectures into estimates, we first describe the observations we will use,
then how we turn our model of the budget process into an estimating equation, and finally derive
a specification which allows us to test the hypotheses listed above. We take each of these steps in
turn. First, we describe the variety of changes of party control we observe, and how we simplify
them to make estimation tractable. Then we describe the budget process and the key parameters
we will estimate and relationships we expect. Finally, we derive a specification we can estimate.
The next part contains some preliminary estimation results.
Making party control changes tractable: data
We base our estimates on data covering forty years of party government in the 35
nonsouthern American states, from 1950-1989. Spotty data availability in the early years means
that an effectively complete data set (including lagged variables) is only available from 1954 on.
There is, however, a much more serious problem to confront. The intuition of the party controltarget model is one of changes of control between effectively unified parties (see Hibbs’ (1987)
model of fiscal policy, for example) as though divided government did not exist. The problem is
that despite numerous changes of party control, there are exactly ten episodes of switching
between unified control by opposite parties. This is elaborated in Figure 2 and Table 1. Figure 2
5RNKV DTCPEJ gives examples of the variety of partisan histories of the northeastern states, separating out
observations of unified Democrat (DD) and Republican (RR) control, split legislatures under
Democrat (DS) and Republican (RS) governors, and all the varieties of split branch government
with Democratic governors and Republican legislatures (DR) in which the key veto location is
held by Democrats in both chambers (V=D), Republicans in both (V=R), or is split between the
chambers (V=S), and similarly for all the other split branch cases where party control of the
branches is reversed.
[Table 1 and Figure 2 about here]
States outside the South have varied partisan histories: there are large numbers of changes
of party control, but that few of them go the whole way across the party spectrum. Examples
exist in which the “same” configuration in a state has on different occasions inherited budgets
formed under unified control by different parties.
transitions in Figure 2.
Table 1 summarizes all the year-to-year
Clearly, most often configurations are succeeded by identical
configurations, but most of the possible transitions have occurred in practice somewhere at some
time. Most important, as we remarked above, switches of unified control are nearly absent: five
times RR government has turned into DD, and five times DD has turned into RR. This is in 35
states, over 40 years. Even if we add a second conjecture consistent with our model, that
vetoproof legislatures should be statistically indistinguishable from unified government, we don’t
add that many cases: there are two occasions on which RR government became RD (V=D) (that
is, the Republicans lost both houses to vetoproof Democratic majorities), 3 when DR (V=R)
became DD (the same sort of change though the Democrats don’t need to be vetoproof to impose
their target since they hold the governorship in this case), and 2 when DD became DR (V=R)
5RNKV DTCPEJ (Democrats lost both chambers to vetoproof Republican majorities). So we get 17 instances of
all-or-nothing transitions, in forty years.
Clearly, there are many more cases where the partisan configuration moves from RR to
DD, or vice-versa, over longer periods. The very first state in Figure 2, Maine, moves from RR
to DD government, over two decades from the 1960s to the 1980s. Connecticut does the same,
in about six years in the 1950s.
It then moves back, but only briefly, and back again.
Massachusetts’ early transition is much like Connecticut’s, but it stays more Democrat-controlled.
The effects we model should leave their traces in the histories of fiscal policy in these states, but
probably only if we think of the effective partisan target for the budget as itself a time series
variable that moves back and forth in response to shocks in proportion to the distance between it
and the "normal" target of a party finding itself in control. In that way, the rate at which the
underlying current target moves will vary with the actual configurations (and laws, and other
variables) in ways we describe.
For our first cut at estimation in this paper, however, we ignore veto issues4 and simply
focus on three patterns of party control: unified, split branch, and split legislature. Once we have
an empirical model that can distinguish adjustment speeds in unified and divided government, we
will then try to distinguish among different cases of divided government. We will then also
consider the effects of legal constraints, most likely as interactive shifts in the speed and extent of
response to unanticipated surpluses and deficits.
The issue raised by vetoes is this. If control shifts from unified Democrat to a veto-proof
Republican legislature with a Democratic Governor, then we should be able to track downward
movement in the income share going to revenues (since the legislature can achieve its bliss point,
quickly) and that movement should be faster than in the case where the Republican legislature
(following a similar transition) isn't vetoproof.
5RNKV DTCPEJ Setting up the budget process for testing
We want to estimate such a dynamic model based on the revenue and expenditure
equations that formed the basis of the long-run estimates in Alt and Lowry (1994). 5 To
summarize, income is forecast, revenues are targeted in a way which captures how far the current
administration is from its bliss point and how fast its type of configuration can shift policy, and
then expenditures spend target revenues fully, allowing for federal contributions, the business
cycle, the interaction between institutions and past deficits and surpluses, and some degree of
stickiness.
Such a model requires that income be forecast from its own past values state-by-state,
then the income forecast is multiplied by the “partisan target parameter” to get expected revenues.
Formally, the partisan target parameter is an impact coefficient which reflects two things: how far
the previous period’s party target was from the preferred point of the current governing
configuration, and a vector of dynamic coefficients capturing the different expected speeds of
adjustment of different party configurations.
This paper simplifies the above discussion somewhat. We preserve the core conjecture of
the model, that Democrats (Republicans) prefer a public budget which taxes and spends a higher
(lower) share of state incomes. We defer considering expenditures for the moment. To make a
start at showing that revenues within states over time follow predictable patterns that are
consistent with the hypotheses above, in our estimates below the “budget” contains revenues
There we set revenues equal to lagged revenue and state income (embodying the assumption that
the current year’s revenue forecast and party targets are based on those variables), plus
contemporaneous federal contributions, allowing for an adjustment to the previous surplus/deficit.
A second simultaneous equation set expenditures equal to revenues plus unemployment
(unanticipated cyclical expenditures and a further adjustment to the previous surplus/deficit. Party
5RNKV DTCPEJ which, random shocks apart, change in each period as a function of expected state income, federal
contributions, and last year’s surplus or deficit in a way that reflects three political variables:
which party sets the target share of income, whether government is unified or divided, and
whether divided government results from control of a bicameral legislature resting with different
parties in each chamber.
Two core empirical predictions are retained. The first is that the “target share” of income
should be higher under Democrats. Initially, for tractability in estimating, we will assume that
Democrats have the same target in all states (and all Republicans have a common, lower target),
but it would probably be desirable to relax this assumption so that only the difference between or
the ratio of party targets was constant. Furthermore, nothing in our model assumes that the target
shares vary according to whether control is unified or divided. In our model, patterns of control
affect the speed of adjustment to targets, not the levels of the targets themselves.
Second, combining hypotheses 1-3 from earlier sections, changes of unified control should
display the biggest effects in terms of the budget changing in response to new partisan targets. In
theory this could happen more or less immediately. However, the presence of large numbers of
entitlements and entrenched constituencies supporting existing programs (and a long history of
budget studies) suggest that there will be some, perhaps considerable friction in the process.
Hence a more reasonable conjecture is that changes of party control under unified government
should reveal the fastest adjustment. That is, if unified Republican control gives way to unified
Democratic control, the income share of revenues should rise more quickly to a level given by the
“Democratic target share” (and then stay there, random shocks apart) than would be the case if
targets were derived from the long-run behavior of the estimated model, with separate sets of
observations corresponding to different partisan configurations.
5RNKV DTCPEJ the new government were divided, but with Democrats setting the target. We lack a theoretical
reason to say whether split branch government adjusts faster or slower than split legislature
government, but an empirical implication of Alt and Lowry (1994) was the split legislature
government was most likely not to adjust at all.
Model specification6
To sum up, which party sets the target determines whether the budget will be intended to
push revenues towards (or to stay at) a higher or lower share of state income. Whether and how
government is divided determines how fast the budget will move toward the target share in a
given year, presuming it is not already at the target. Under unified government the governing
party sets the target share of expected income in planning the budget. Where government is
divided but control of the legislature is not, the party controlling the legislature sets the target
share. Where control of the legislature is divided, we assume the governor’s party sets the target.
Formally, we can represent this by defining three dummy variables (we choose this way for
empirical convenience later):
Ψ = {1 if Divided Government; 0 if Unified Government}
Λ = {1 if Divided Legislature; 0 if Unified legislature}
Γ = {1 if Democratic Governor; 0 if Republican Governor }
Then the Governor’s party sets the target if Ψ = 0 or if Ψ = 1 and Λ = 1; otherwise (Ψ = 1 and Λ
= 0) the party controlling the legislature sets the target.
Now call τR the Republican and τD the Democratic target share of income. A function
which returns the correct value (τR or τD) depending on the three dummy variables above is
9G ITCVGHWNN[ CEMPQYNGFIG VJG VJG OCLQT EQPVTKDWVKQP QH 'TKE &KEMUQP KP FGTKXKPI VJG URGEKHKECVKQP FGUETKDGF
KP VJKU UGEVKQP
5RNKV DTCPEJ f = τR + (τD - τR){(1 - Ψ)Γ + Ψ[(1-Λ)(1 - Γ) + ΛΓ]}
(1a)
where f is the target share of income. It is useful to defines the term in braces as
Ω(Ψ,Γ,Λ) = (1 - Ψ)Γ + Ψ[(1-Λ)(1 - Γ) + ΛΓ]
(1b)
so that Ω, a known combination of exogenous variables, is a dummy variable taking the value 1
when the Democrats set the target and zero otherwise. For simplicity,
f = τR + (τD - τR) Ω
(1c)
Now, consider a state government in which, in the present year t, the target-setting party
changes. Over the next transition period of j years, j > 0, the budget’s actual behavior moves
towards the governing party’s target. When the state has had the same target-setter for at least j
years, the transition is over, and the budget can be considered to be at the target level. (Of
course, the target is not reached if the target-setting party changes again before j years have
passed.) Hibbs (1987) describes a related setup for fiscal policy.
Define the speed of the adjustment process in the transition as the fraction of the distance
to the target which is achieved in a given year. Formally, define
Rt - R t-1 ≡ vt(RtT - Rt-1)
(2)
where vt is the speed (which varies from year to year), Rt is the actual current revenue level, and
RtT is the target level of revenue.
The speed variable in any year must account for two factors, the current configuration of
party control, and whether the budget is still in transition.7 (In the specification we derive,
If we could use a nonlinear transfer function model, estimating impact and dynamic parameters
for several explanatory variables, as Hibbs (1987) does, we would not need to be concerned about
j. But transfer function methods are generally for single series rather than for a pooled crosssection time series like our data, so while we have done some single-state experiments with
5RNKV DTCPEJ revenues outside the transition are an autoregressive process, but the specification can easily be
augmented to allow for the impact of federal contributions and lagged surpluses in all periods.)
Given that a transition is underway, the speed function can be written
v0(Ψ,Λ) = (1 - Λ)[vUG(1 - Ψ) + vSB(Ψ)] + ΛvSL
= vUG + (1 - Λ)Ψ(vSB - vUG) + Λ(vSL - vUG)
(3)
which returns the correct v depending on party configuration. To get the desired function, we
need only add an appropriate indicator function for whether we are still in transition (given a
choice of j). Since Ω indicates the identity of the target-setter
vt(Ψ,Λ,Ω t,Ω t-1,…) = H((Σnj|Ω t - Ω t-n|) - .5) vt
(4)
where H(x) = {1|x ≥ 0; 0|x<0}. If the target has not changed in j periods, Ω t = Ω t-1 = … = Ω t-n
and H(.) = 0, so vt = 0. Otherwise the the function returns the configuration-dependent speed v0.
It will be convenient to define ξtj = H((Σ|Ω t - Ω t-n|) - .5) so that, for any value of j, the length of
the transition, ξtj = {1 in transition; 0 otherwise}.
Now, to get the estimating model set up, define the government’s (retrospective) income
forecast YtF to be autoregressive of order n: YtF = Σkn ϕkYt-k . The forecast will in general be
incorrect: call the forecast error Yt - YtF = εt. Since εt = Yt - Σkn ϕkYk , εt is ma(n).
Now, rewrite (2) as
Rt = R t-1(1 - vt) + vt RtT
(5)
RtT = ft YtF + gtFt + htS t-1
(6)
where
and, in (6), gt and ht are configuration-specific parameters defined analogously to ft in (1c).
transfer function methods, we believe we need to derive a model which can be estimated by more
5RNKV DTCPEJ Combining (5) and (6), multiplying out the parentheses in (5), and subtracting R t-1 from
both sides gives the estimating equation
Rt - R t-1 = -vt R t-1 + ft vt YtF + gt vt Ft + ht vt S t-1 + νt
(7)
Though revenues grow throughout the period, the revenue series is stationary in its first
differences form, and so ordinary regression methods can be used to estimate equation (7). There
could certainly be cross-sectional heteroskedasticity and timewise correlation in the errors. Recall
that since vt = ξtjv0 multiplying each term by vt implicitly separates the estimates in and out of
transition. Out of transition all vt = 0 and so the model is just a first-order integrated process.
Hence it could also contain terms for (1-ξtj)Ft and (1-ξtj)St-1, the impact of federal contributions
and lagged surplus/deficit outside transition periods. Equation (7) could also contain a term for εt1
, the lagged income forecast error, both in and out of transition periods as well, though
presumably some of that variable’s effect is reflected in the lagged surplus/deficit.
We will
examine the effects of those terms in the empirical results section below. We also put in a full set
of state fixed effects, to allow revenues to have a different mean level in each state.
To understand the estimates below better, think of equation (7) as containing three
substantive explanatory variables: the income forecast YtF, exogenous federal contributions Ft, and
the lagged surplus/deficit St-1. Each of these appears six times in the actual estimating equation, in
six interactions representing the product of v t (three adjustment speeds by configuration of
control, in transition periods) and a variable like ft, the two party-specific income shares
determined by τ and Ω. The lagged revenue variable is multiplied by -vt, so it too has effects
confined to transitions (by assumption, the transition is when the adjustment to targets is taking
place) in three interactions, according to the configuration of party control.
conventional regression methods.
5RNKV DTCPEJ Estimation Results
As of this writing, we have completed only preliminary work with a model based on
equation (7) above, from which we report results from data for 32 states from 1954-89.8 The
dependent variable is the change in annual general revenues as reported in the Census of
Governments. Explanatory variables include the income forecast YtF, measured as the predicted
values from a regression of state aggregate personal income on its own first two lags9, exogenous
federal contributions Ft, and the lagged surplus/deficit St-1 defined as the difference between
general revenues and expenditures, lagged one year. Each variable appears in six interactions as
described above, to isolate the effects of each combination of control configuration and targetsetting party. We set j=2, so adjustment speed and party target estimates are based on the
transitional period of two years after a change in the identity of the target-setting party.
Table 2 provides descriptive statistics for our data,10 while Table 3 contains the estimated
results.
The model also contains terms for (1-ξtj)Ft and (1-ξtj)St-1, the impact of federal
contributions and lagged surplus/deficit outside transition periods, εt-1, the lagged income forecast
error, both in and out of transition periods, and all state fixed effects not shown in the Table).
[Tables 2, 3 about here]
Estimates for Adjustment Speeds and Party Targets
AZ, CA, CO, CT, DE, ID, IL, IN, IA, KS, ME, MA, MI, MN,
MO, MT, NV, NH, NJ, NM, NY, ND, OH, OR, PA, RI, SD, UT, VT, WA, WI, and WY.
6JG UVCVGU YJKEJ CTG KPENWFGF CTG
The income forecast equation included a separate intercept for each state. Results are not
shown.
State fiscal data are from the Statistical Abstract and State Government Finances. Income
data are from the Statistical Abstract, and the Survey of Current Business. Partisan
configurations are from The Book of the States. Fiscal and economic variables are measured in
real (1982-1984), per capita terms. Regional CPI deflators are available only from 1968 on; prior
to 1968, we use the national CPI deflator.
5RNKV DTCPEJ With j, the length of the transition period, set equal to two, the estimates in Table 3 that
relate to the transition period are based on about 230 observations. The fit is not too good,
though the corrected R2 is reduced by the inclusion of many insignificant state-specific intercept
shift parameters. Moreover, there is a lot of multicollinearity, especially within the groups of
explanatory variable within-transition interactions: for instance, the R2 values when one of the six
interactive explanatory variables is regressed on the other five are on the order of .35, .75, and .95
for income forecasts, federal contributions, and lagged surpluses, respectively.
Since
multicollinearity lies behind some of the apparently high coefficient standard errors in Table 3, we
will use simulation methods to judge the robustness of the inferences we draw.
The first parameter of interest is the adjustment speed, which we expect to be greater for
unified than divided governments.
The first coefficient in Table 3 is (minus) the unified
government adjustment speed11, so the adjustment speed, the proportion of the distance between
the party’s ideal share of income and the current actual income share of the public budget, is
estimated to be about 15 per cent under unified government, based on transitions which last two
periods. If (currently unified) Democrats would like to tax and spend at ten per cent of state
income, and the publuc budget currently stands at eight per cent, the prediction is that it moves to
8.3 per cent after a year, other things equal, on average.
Is adjustment slower under divided government? Apparently, though the standard errors
of the estimates do not permit firm inferences. The second coefficient, .0245, is the estimated
difference between the adjustment speeds of unified and split branch government: doing the
The non-obvious parameterization used to derive the specification was chosen to make the
unified government adjustment speed parameter stand alone in the regression, since it is the one
for which data should be clearest, and differences most marked.
5RNKV DTCPEJ subtraction reveals that the estimated adjustment speed of split branch government is about .13.
Of course, all the sources of heterogeneity discussed in the first part of this paper (identity of the
veto player, direction in which adjustment is taking place, whether the last year’s outcome was a
surplus or deficit) are still omitted from the analysis, and could affect these estimates. Analogous
calculations reveal the estimated speed of adjustment under split legislature government to be .10,
about two-thirds the adjustment speed of unified government. So, qualitatively at least, the
estimates line up in an order that corresponds to what was expected, based both on the model and
past research: unified government adjusts fastest, and split legislature government slowest.
The party differences in desired income shares
;V(
(tau parameters) have to be derived
from the estimates in Table 3, contingent on the adjustment speeds. Since we did not employ a
nonlinear regression model, we get separate estimates of the party targets for each configuration,
and of course the errors of estimation in the adjustment speeds also become conflated with the
estimation errors of the targets. To get the first (unified Republicans) target τ , divide the
4
coefficient (.0036) by the unified government adjustment speed
X7)
(.15). The result (.023) and
all the other implied party targets are recorded in Table 4, to spare the reader a lot of tedious
calculations, and the adjustment speeds are included as well. The units of these targets12 are
described in equation (6), and are the fractions of state income (.02 would be 2 per cent, for
instance) which, along with federal contributions and the lagged surplus of deficit, determine the
party’s revenue targets.
[Table 4 about here]
Whether this is the target for discretionary taxing and spending (that is, excluding entitlements)
and how it is affected by whether the party targets the level or change in revenues is an issue we
are still trying to clarify.
5RNKV DTCPEJ The party targets from the income forecasts, for unified government, are very much as
expected: the Republican target is smaller and the Democrat target larger, and both are positive.
Under split branch government, we get a negative estimate for the Republican target. Certainly
the difference in magnitudes is consistent with income changes producing more revenue growth
under Democrats. The pattern τ < 0 and τ > τ is consistent with the spatial split branch model
4
&
4
above (R t - Rt-1 is greater for RD than DR following both deficits and surpluses, although both
changes may be negative following a deficit). Nevertheless we have to state clearly that we are
only just – as of this writing – beginning the task of fitting together the detailed predictions of the
spatial models with the further interactions and restrictions that this estimating model makes
possible! Again, with split legislatures, the results are qualitatively consistent with more revenue
growth from a positive income shock when Democrats are the target-setters. So we take some
comfort from the fact that the inequality τ > τ holds up across configurations, but would not
&
4
want to claim more than a preliminary, tentative result. The standard errors are also quite large in
this case, so we will turn to simulations in a moment.
The results for partisan responses to surpluses and deficits are negative in nearly every
case (τ > &
KP VJG QPG ECUG KU CEVWCNN[ XGT[ PGCT \GTQ UQ QP VJG YJQNG UWTRNWUGU FGHKEKVU NGCF
VQ FGETGCUGU KPETGCUGU KP TGXGPWGU CU KP #NV CPF .QYT[ 6JG TGURQPUG CRRGCTU VQ DG
UWDUVCPVKCNN[ DKIIGT KP CDUQNWVG XCNWG OQTG PGICVKXG WPFGT 4GRWDNKECPU VJCP &GOQETCVU
YJKEJ KU CNUQ EQPUKUVGPV YKVJ VJG GCTNKGT RCRGT
RGTKQF
6JGTG KU CNUQ C TGURQPUG QWVUKFG VJG VTCPUKVKQP
(GFGTCN EQPVTKDWVKQPU CNUQ JCXG CP GHHGEV QP TGXGPWG EJCPIGU DWV VJGTG YGTG PQ ENGCT
RCTV[ FKHHGTGPEGU KP VJKU ECUG UQ VJG GUVKOCVGU CTG PQV UJQYP KP 6CDNG #U 6CDNG TGXGCNU
6JGTG KU QPG XGT[ NCTIG EQGHHKEKGPV HQT &4 IQXGTPOGPV YJKEJ EQWNF DG C EQORQUKVKQPCN
GHHGEV C FCVC GTTQT QT LWUV CP GHHGEV QWVUKFG QWT OQFGN
5RNKV DTCPEJ VJG NCIIGF KPEQOG HQTGECUV GTTQT KH RQUKVKXG WPGZRGEVGF KPEQOG ITQYVJ VJG RTGXKQWU [GCT
CNUQ RTQFWEGU TGXGPWG ITQYVJ FGURKVG UQOG QH KVU GHHGEVU RTGUWOCDN[ UJQYKPI WR KP VJG
TGXGPWG UWTRNWU VJG RTGXKQWU [GCT DWV PQ FKXKUKQP D[ RCTV[ QT EQPHKIWTCVKQP QH EQPVTQN
UKIPKHKECPVN[ KORTQXGF VJG RTGFKEVKXG RQYGT QH VJKU XCTKCDNG
+P UWO
the adjustment speeds line up as expected, changes in revenues as a share of
income are larger (more positive) when Democrats set policy than when Republicans set policy,
and Republicans take much more of the lagged surplus out of revenues than do Democrats. The
fit is not great, but the dependent variable is in changes, not levels. All in all, we take this result
as an encouraging starting point.
Simulation results
The results in Table 4 are based on combinations of coefficients in Table 3, several of
which are not statistically significant at conventional levels. In order to get a better feel for the
relationships between different parameters in Table 4, we conducted monte carlo simulations by
making random draws of coefficients using the full variance-covariance matrix underlying the
results in Table 3. These simulations revealed that the point estimates for each of the adjustment
speeds are more than twice their standard errors, but that the standard errors for all of the party
difference parameters exceed their point estimates. 14
We also calculated the probability that each of our hypothesized relationships between
parameters occurs. Table 5 shows the estimated probabilities, based on 100,000 random draws.
The simulation results are affected by the nonlinear relationship between τ and v in the
parameter estimates. In particular, a chance very small draw for v, which becomes the
denominator of a ratio, can result in an outlying estimate of τ. However, we could re-present the
analysis using confidence intervals around median simulated values and the qualitative
interpretations we present would not change.
5RNKV DTCPEJ Although each of the hypothesized relationships occur more often than not, the results are not
overwhelming. The probability that the estimated adjustment speed for unified government
exceeds that for split legislature is .772, but the probability that unified government adjusts more
quickly than split branch government, or split branch more quickly than split legislature
government, is less than 0.7. Similarly, the probability that the change in revenues as a share of
the income forecast is greater under Democrats than Republicans ranges from .6 to .65. The
probabilities of party differences in responses to lagged surpluses and deficits are somewhat
greater, ranging from .64 to .853. These probabilities make it clear that there is a good deal
ofparameter estimate instability, not just multicollinearity, underlying these results, though they
are suggestive of some underlying relationships, nevertheless.
[Table 5 about here]
Robustness checks
In this section, we will briefly remark on some of the estimation we have done around the
results summarized in Tables 3-5 above. We have rerun this model using Stata’s cross-section
time series fixed effects model, and get substantially the same results, save that the centering is a
little different, and the number of parameters is lower since the fixed effects are handled
differently. In fact, one can almost reduce the fixed effects to a few regions (the northeast, the
Ohio valley, the upper midwest, and the midwest/west) but a few discrepant cases (NM, DE, NH,
maybe 1or 2 others) do not fit this classification very well, so we have not pursued this further at
this time. In terms of possible heteroskedasticity, the residuals from the model in Table 2 do not
differ significantly by ξ (in/out of transition) or Ω (whether the Democrats or Republicans set the
V
target). This is mildly reassuring.
5RNKV DTCPEJ When we increase j (two years is a short transitional period), the split branch and split
legislature party target results hold up better than the unified government ones, but things really
get messy in a hurry. It is possible that we need to revise the specification so that the transition is
shorter for unified government than for divided government. We will pursue that in future work.
Conclusions
Conclusions are premature, at this time. We started from an interest in the effects of
parties on the magnitude of budgets at the state level. Previous research suggested that such a
long-term association existed and was perceived by voters and used to form expectations about
the performance of partisan politicians in office. There were also suggestive empirical results
hinting at differences between parties in their responses to fiscal imbalance. We have written
down and analyzed a formal model of the dynamics of short-term changes in budgets in a partisan
system, and derived predictions from it. The empirical specification yielded some preliminary
estimates consistent with the broadest qualitative conjectures from the formal model, namely that
Democrats target larger budgets from incomes and that unified government adjusts faster to new
party targets. Nevertheless, this is only a beginning. Some likely next steps include reclassifying
party configurations to take account of supermajority voting rules, whether the legislature is veto
proof, distinguishing between surpluses and deficits, and adding a model of expenditures to our
revenues equation.
We welcome comments and suggestions.
5RNKV DTCPEJ References
Advisory Commission on Intergovernmental Relations. 1991. Significant Features of Fiscal
Federalism. Washington, DC: Author.
Alesina, Alberto and Allan Drazen.
1991.
“Why are Stabilizations Delayed?”
American
Economic Review 81:1170-1188.
Alt, James E. and K. Alec Crystal.
1983.
Political Economics.
Berkeley: University of
California Press.
Alt, James E. and Robert C. Lowry. 1994. “Divided Government, Fiscal Institutions, and Budget
Deficits: Evidence from the States.” American Political Science Review. 88(December):
811-828.
Carter, John and David Schap. 1990. “Line Item Veto: Where is Thy Sting?” Journal of
Economic Perspectives. 4:103-118.
Clarke, Wes. 1998. “Divided Government and Budget Conflict in the U.S. States.” Legislative
Studies Quarterly, 23(February): 5-22.
Council of State Governments. Various. The Book of the States Lexington, KY: Author.
Fiorina, Morris. 1992. Divided Government. New York: Macmillan.
Frey, Bruno. 1978. Modern Political Economy. New York: Wiley.
Hibbs, Douglas. 1987. The American Political Economy. Cambridge: Harvard University Press.
Hinich, Melvin J. and Michael C. Munger. 1997. Analytical Politics. Cambridge; New York:
Cambridge University Press.
Lowry, Robert C. and James E. Alt. 1997. “A Visible Hand? Balanced Budget Laws, Imperfect
Information, and Fiscal Policy in the States.” Manuscript.
5RNKV DTCPEJ Lowry, Robert C., James E. Alt, and Karen E. Ferree. 1998. “Fiscal Policy Outcomes and
Electoral Accountability in American States.” Manuscript.
Shepsle, Kenneth A. and Mark S. Bonchek. 1997. Analyzing Politics: Rationality, Behavior,
and Institutions. New York: W.W. Norton and Company.
United States Department of Commerce, Bureau of the Census. Various. Statistical Abstract of
the United States. Washington, DC.
United States Department of Commerce, Bureau of the Census. Various. State Government
Finances. Washington, DC.
United States Department of Commerce, Bureau of Economic Analysis. Various. Survey of
Current Business. Washington, DC.
Table 1
A.
Transitions in Party Control
Thirty-five Nonsouthern States
Type of
| Party control in previous period
Party
|
Control
|
RR
DR V=R
RS
DR V=D
IND |
Total
-----------+-------------------------------------------------------+---------RR |
291
6
11
16
0 |
344
DR V=R |
3
30
0
5
0 |
45
RS |
17
0
75
2
0 |
114
DR V=S |
7
5
0
57
0 |
75
DR V=D |
8
1
3
73
0 |
98
IND GOV |
0
0
0
1
3 |
4
RD V=R |
6
0
11
0
0 |
104
RD V=S |
1
0
3
0
0 |
48
DS |
12
1
6
11
1 |
153
RD V=D |
2
0
0
0
0 |
41
DD |
5
3
6
7
0 |
288
-----------+-------------------------------------------------------+---------Total |
352
46
115
172
4 |
1314
Type of
| Party control in previous period
Party
|
Control
|
RD V=R
DS
RD V=D
DD |
Total
-----------+--------------------------------------------+---------RR |
7
8
0
5 |
344
DR V=R |
0
5
0
2 |
45
RS |
12
3
2
3 |
114
DR V=S |
1
4
0
1 |
75
DR V=D |
1
10
0
2 |
98
IND GOV |
0
0
0
0 |
4
RD V=R |
73
0
0
14 |
104
RD V=S |
31
2
7
4 |
48
DS |
0
104
0
18 |
153
RD V=D |
5
1
25
8 |
41
DD |
15
13
6
233 |
288
-----------+--------------------------------------------+---------Total |
145
150
40
290 |
1314
B.
Fifteen Southern States
Type of
| Party control in previous period
Party
|
Control
|
RD V=R
RD V=D
DD |
Total
-----------+---------------------------------+---------RD V=R |
4
1
2 |
7
RD V=S |
4
1
2 |
7
RD V=D |
3
77
20 |
100
DD |
1
17
462 |
480
-----------+---------------------------------+---------Total |
12
96
486 |
594
Note – configurations of party control are DD for unified Democrat, DS for Democratic
Governor, split legislature, etc. Split branch (DR and RD) governments are further divided by
whether the veto point V in the legislature is Republican, Democratic, or split.
Table 2
Summary of the Data Used in Estimations
Fiscal Policy and Economic Variables (real (1982-84), per capita dollars)
Variable
Mean S.Dev.
Min.
Max.
General revenues
1120
453
270
3100
37
64
-335
512
10448
2320
3
407
-1505
3868
294
135
29
839
17
92
-294
734
Change in general revenues
Income forecast
Income forecast error
Federal transfers
Lagged surplus or deficit
Party configurations
5265 19099
Cases
Unified Democrat
236
Unified Republican
223
Democratic Governor, Republican Legislature
187
Republican Governor, Democratic Legislature
176
Democratic Governor, Split Legislature
126
Republican Governor, Split Legislature
88
Total
1040
Table 3
Estimation Results
Dependent variable = Revenue first differences 4 4
V
Explanatory Variables
Coef.
Std. Err.
V
t
Lagged revenues
X7) 4V
(X
(X
7)
X5$ 4V
7)
X5. 4V
-.1534336
.02453
.0558181
.0613566
.073205
.0758948
-2.501
0.335
0.735
.0036018
.0086826
.0012099
.0016801
-.0106297
-.0052915
.0059576
.0075978
.0063927
.0053338
.0071857
.0066503
0.605
1.143
0.189
0.315
-1.479
-0.796
.0926839
.4405657
-.3526629
-.2467952
.0746052
.2723348
.1573089
.0214523
.1978807
.2615873
.2254812
.2110163
.2723765
.2540532
4.320
2.226
-1.348
-1.095
0.354
1.000
0.619
.0292301
.2653343
.3048028
.2925938
.3145846
.3708848
.3669119
-6.120
-1.849
1.866
1.859
1.061
-1.593
-0.919
.0048759
2.358
Income forecast
τX ;
τ X X
τ X X
τ τ X
τ τ X
τ τ X
4
7)
(
V
4
5$
7)
;V
4
5.
7)
;V
(
(
;V(
&
4
7)
&
4
5$
X7);V(
&
4
5.
X7);V(
Federal contributions
ξVLV(V
I4X7)(V
I4
X5$ X7)(V
I4
X5. X7)(V
I& I4 X7)(V
I& I4
X5$ X7)(V
I& I4
X5. X7)(V
Lagged surplus/deficit
ξVL5V
J4X7)5V
J4
X5$ X7)5V
J4
X5. X7)5V
J& J4 X7)5V
J& J4
X5$ X7)5V
J& J4
X5. X7)5V
-.178895
-.4904885
.5687467
.5439468
.3337002
-.590944
-.3370648
Lagged income forecast error
ε
.0114952
V
Intercept
Number of obs =
26.04141
17.15927
1040, Ctd. R-squared =
1.518
0.0568, Standard error =
62.547
Note - Estimation by OLS, using Stata v5. State fixed effects omitted from
table.
Transition period j is set to 2.
Table 4
Unified Government
Parameters of Interest
Split Branch
Split Legislature
Adjustment speeds
.15343356
.12890353
.09761544
Party Differences for Income Forecast
Republicans .02347475
Democrats
.0344247
-.35395613
.07937862
-.0216763
.07312273
Party Differences for Lagged Surplus
Republicans -3.1967482
Democrats
-1.0218639
-23.18574
.90490116
Note: Source is coefficient estimates in Table 3.
derivation in text.
-9.7449878
-3.706361
Calculations follow
Table 5
Simulation Results
Adjustment speeds
Prob.(VUG > VSB) = 0.639
Prob.(VUG > VSL) = 0.772
Prob.(VSB > VSL) = 0.694
Party Differences for Income Forecast
Unified government: Prob.(τD > τR) = 0.622
Split branch:
Prob.(τD > τR) = 0.604
Split legislature:
Prob.(τD > τR) = 0.647
Party Differences for Lagged Surplus
Unified government: Prob.(τD > τR) = 0.853
Split branch:
Prob.(τD > τR) = 0.640
Split legislature:
Prob.(τD > τR) = 0.747
Probabilities are based on 100,000 monte carlo simulations using the
coefficients and the full variance-covariance matrix for the results in Table
3.
Figure 1a - Democratic Governor, Republican Legislature, Surplus
Revenues
r2
r1
t
D Gov
target
t1
R Leg
Spending
Figure 1b - Democratic Governor, Republican Legislature, Deficit
Revenues
t
t1
target
R Leg
r2
D Gov
r1
Spending
Figure 1c - Republican Governor, Democratic Legislature, Surplus
Revenues
target
D Leg
r2
r1
t1
R Gov
Spending
Figure 1d - Republican Governor, Democratic Legislature, Deficit
Revenues
D Leg
target
r2’
t1
R Gov
r2
r1
Spending
(KIWTG 2CTVKUCP 6TCPUKVKQPU KP 5GNGEVGF 5VCVGU