AN ECONOMIC THEORY OF THE BUDGETARY PROCESS*
Soumaya Tohamy
Peter H. Aranson
Hashem Dezhbakhsh
February1999
* The authors are affiliated with Berry College, Emory University, and Emory University,
respectively. Send inquiries to Peter H. Aranson, Department of Economics, Emory University, Atlanta,
GA 30322-2240, Fax (404) 727-4639, paranso@emory.edu.
An Economic Theory of the Budgetary Process
Abstract
This paper challenges the view that budgetary decisions are incremental because they are
complex, extensive, and conflicted. Our model interprets incrementalism (or nonincrementalism) as a
legislative political strategy in response to interest group politics and economic conditions. In a given
year, a legislator chooses between single-period budgeting (SPB), where he promises funds for one
year, or multi-period budgeting (MPB), where he promises perpetual funding. SBP is associated with
a greater chance of nonincremental budgeting outcomes than is MPB. We offer a new inferential
procedure for identifying nonincremental outcomes. Statistical results based on post WWII data
series of agency outlays support the model's predictions. Specifically, the Democrats’control over the
political process, a switch in the party controlling the White House or Congress, and presidential
election year promises and political vulnerabilities all appear to cause departures from incremental
budgeting. A persistently large deficit and a higher discount rate have a similar effect. Results are
robust across various specifications.
I. Introduction
This paper challenges the view that legislative budgetary decisions are incremental1 because they
are complex, extensive, and conflicted. We show instead that incrementalism is a distinct legislative
strategy, rationally chosen, which reflects legislators' responses to interest-group politics and economic
conditions. We construct a model that predicts when legislators' budgetary decisions will be incremental
and when nonincremental. Our test of the model shows that we can reject the null hypotheses that nearly
all of the relevant economic variables do not affect the incremental versus nonincremental (meta) strategy
choice in ways that the model predicts.2
Since the publication of Simon's original papers four decades ago (1955, 1959), social scientists
have believed that decision makers approach complex problems by using simplifying rules, sometimes
called heuristics (see Heiner 1983). Because legislative budgetary decisions exhibit complexity,
extensiveness, and conflict, writers since Lindblom (1959) have argued that the use of incremental
strategies would prevail in regularized political contexts. Virtually all of the subsequent research of
which we are aware takes the (atheoretical) case for incrementalism as given and tests for its presence.
We have found no research that approaches the meta-decision of deciding how to decide -- incrementally
or nonincrementally--using a rational choice model.
The extensive empirical literature on incrementalism dates to the work of Davis, Dempster, and
1
The term “incremental” refers to several different phenomena. Here it denotes a marginal outcome of a process that may
or may not involve reconsideration of the base--the budget of the previous year. The most common understanding of
incremental budgeting is that the allocation for year t is the allocation for year t− 1 plus a small (usually positive)
increment. The legislature ordinarily accepts the year t budgetary “base” without reconsideration.
2
Our model predicts that the extent of incremental versus nonincremental budgetary outcomes varies systematically with
certain economic variables (e.g., interest and discount rates, inflation, and the size of federal revenues and expenditures)
embedded in legislators’payoffs. Hence, we are comfortable in asserting that budgetary politics contain a significant
Wildavsky (DDW 1966a), who try to explain budgetary decisions with a set of simple linear equations
relating agency outlays in one year to either agency requests in the same year or agency outlays in the
previous year, and relating agency requests in one year to either agency requests or outlays in the
previous year. They conclude that most budgetary outcomes are incremental.3 Natchez and Bupp
(1973), by contrast, examine different programs within the Atomic Energy Commission and find that
there are considerable variations in program budgets over time. They conclude that by concentrating on
agency budgets and not those of programs within agencies, DDW’s research masks considerable
nonincremental budgetary choices. Gist (1974) shows that after disaggregating the total budget into its
major functional components, defense sector budgets give evidence of nonincremental decision making
more than fifty percent of the time. We can divide most of this literature into studies that show (and
verbally but not theoretically explain) incremental budgetary outcomes and those that show its absence.4
The literature appears wholly devoid of a deductive theory of choice that would explain or predict these
outcomes.
We treat the budgetary process as the outcome of the legislator’s5 choice between incremental and
nonincremental budgeting strategies, which in turn leads to five main theoretical results:
(1) the budgetary process tends to display a more incremental outcome, the larger the inflation
element of rational choice.
3
For a discussion and critique of DDW's empirical methodology (and that of others in this literature), see Tohamy,
Dezhbakhsh, and Aranson (1999).
4
Studies that show the presence of incrementalism include: Davis, Dempster, and Wildavsky (1966a, 1966b, 1971, 1974);
Jackson (1972); Moreland (1975); Ripley, Holmes, Franklin, and Moreland (1975); Ripley and Franklin (1975); Bozeman
(1977); Kamlet and Mowery (1985); and Kamlet, Mowery, and Su (1988). Studies that show its absence include: Kanter
(1972); Natchez and Bupp (1973); LeLoup (1978); Lane, Westlund, and Stenlund (1981); and Gist (1982).
5
It is not central to our results whether the legislator is a president who has some measure of control over the budget or a
participant whose preferences represent the median preferences of budget-making participants in all dimensions, or on a
single issue when voting is issue-by-issue in the absence of logrolling.
2
rate, the smaller the discount rate, the smaller the total budget of the previous year, the larger the
total revenues of the previous year, and the larger the real rate of revenue growth;
(2) the budgetary process tends to display a more incremental outcome, the less shirking there
is on the part of legislators, that is, the less strongly they pursue ideologically satisfying projects;
(3) the budgetary process tends to display a more incremental outcome the higher the positive
change in the legislator’s credibility with interest groups from one year to the next;
(4) the budgetary process displays a more incremental outcome if legislators expect to remain
in office for longer than one period;6 thus, the model sheds some light on the issue of term
limitations and how they might affect the marginality of budgetary outcomes;
(5) the total budget tends to grow over time; this result supports Wagner’s law of an
expanding public sector;7 the budget grows at a faster rate under incremental than under
nonincremental legislative strategies.
We test the theory using an empirical procedure that we develop to overcome the statistical
shortcomings of earlier studies. The procedure exploits both the cross sectional and time series
characteristics of the budgetary data, to identify statistically the occurrence of incremental decisions and
to estimate incremental cycles for each agency. We apply a Poisson regression model to the resulting
series of counts to examine how economic and political variables affect the occurrence of incremental
outcomes.
Section II introduces the basic model. Section III presents the empirical analysis and specifically
tests the first set (1) of predictions just listed. Section IV discusses the model's implications and those of
6
If the legislator remains in office for a very long time, this result may be reversed. For a discussion of this prediction, see
Section IV.
7
See Wagner (1877; 1890).
3
an extension to it, highlights some issues for future research, and concludes.
II. The Basic Model
To introduce the model we first distinguish between single-period budgeting and multi-period
budgeting. When deciding on how much money (benefits) to allocate to a certain program (interest
group), the legislator chooses not only the amount of funds to be allocated, but also the period over
which the allocation will flow. The legislator can allocate money to groups, or to programs beneficial to
groups, with the understanding that the funds are allocated for a single year and will not be allocated in
future years unless new deals are negotiated--the single-period budgeting strategy. He also can allocate
money with the understanding that the current year's funding will continue permanently to be allocated in
future periods as well--the multi-period budgeting strategy. Which strategy the legislator prefers
depends on the income (and utility) he derives from each.
The legislator’s choice of multi-period budgeting leads to more occurrences of incrementalism.
Multi-period budgeting represents a promise of continued future benefits. This strategy necessarily
implies that to keep past promises of future funds, the amount of funds remaining from the total budget
for the legislator to distribute freely among programs declines significantly. Single-period budgeting, by
contrast, allows the legislator freely to distribute the total budget among existing and new programs (and
groups). This strategy leads to more variations in budget outcomes. Single-period budgeting may lead
to some marginal outcomes, but multi-period budgeting leads to proportionately more marginal
outcomes.
The total budget level and the individual agency and program components grow out of interactions
4
between interest group members and legislators.8 The budget for each year is the result of the
intersection of group members’aggregated demands for budget money and the legislator’s supply of
budget money. The demand for money reflects group members’willingness to pay for every dollar that
the legislator allocates in year t to any of the government programs that benefit them. Members are
willing to pay a smaller amount per dollar of budget money if the legislator adopts a single-period
budgeting strategy than if he adopts a multi-period one: under multi-period budgeting, a dollar allocated
to a group in year t continues to be allocated in perpetuity to the same group, so the group will be
willing to pay in year t for the discounted present value of a stream of dollars extending in perpetuity.
The legislator faces political opposition that reflects the supply of budget money for any real
expansion of the total budget. He expands the total budget up to the level where his marginal benefit
from the payment received for the last dollar spent equals the marginal political opposition that the last
dollar in taxes or increased deficit generates. The legislator considers both strategies (single and multiperiod budgeting) separately, determines the total budgets resulting from the utility-maximization
problem for each strategy for each year in office, calculates his total utility under each strategy, and
chooses the strategy that leads to a higher total utility. The model sometimes predicts a departure from
incrementalism, which we explain by the inclusion of the temporal element in the legislator's budgetary
promise.
Because the theory relies on elements of the Landes and Posner theory of the independent judiciary
(1975), we review their argument first and explain how it relates to our model. Landes and Posner
8
Our model incorporates currently politically active interest groups and does not incorporate the analysis of the formation
of new groups or the disintegration of existing ones. (See Olson (1977) for an analysis of the formation of interest groups.)
The flow of benefits to interest groups will change as new interest groups form or existing ones dissolve. A group that
organizes will demand benefits, and budgeters will help satisfy its demands in return for votes, campaign contributions, or
other forms of compensation.
5
believe that an independent judiciary that interprets legislation in accordance with original legislative
intent facilitates rather than limits the practice of interest-group politics. They assert that legislators have
powerful incentives to devise methods that increase the permanence of legislation. By doing so,
members of the present legislature can appropriate more income from an enactment because its benefits
extend into the future. The legislation will be worth the discounted present value of a perpetual stream
of benefits, where each future annual payment equals the benefit provided in the year the legislation is
implemented. The legislature achieves permanence through "establishing [legislative] procedures for the
enactment of legislation that increase the costs of repealing it (p. 882)."9
An independent judiciary enhances this permanence. Being independent, the judges could refuse to
enforce legislation they oppose, and thus the expected value of permanent legislation to the group
seeking it would decline. Yet the reduced probability (of sustaining long-run legislative bargains) from
judicial independence does not outweigh the even greater reduction in this probability that might occur if
the judiciary were not independent, but could be subjected to pressure from existing legislators who
oppose the (earlier) legislation. Such legislators might use a “dependent” judiciary to circumvent the
same legislative procedures that make repeal of permanent legislation difficult.
Our concern with Landes and Posner's theory has little to do with the independence of the
judiciary; instead, we use the theory's incorporation of a temporal dimension, particularly the choice
between legislation whose benefits run for one period and legislation whose benefits run for more than
one period. Suppose a legislature in year t decides to repeal a “permanent” allocation made in an earlier
year. The benefits that the present legislature can reap from enacting its own permanent legislation
9
These procedures include the Senate’s filibuster, agenda control by the leadership, and the structure of committee
jurisdictions. See, for example, Shepsle and Weingast (1987) and Weingast and Marshall (1988).
6
decline, because a present interest group's expected benefits, reflecting the greater probability that some
future legislature will rescind its own program, likewise decline. When choosing to rescind otherwise
permanent legislation passed in a previous year, therefore, the legislature includes, and is constrained by,
the lower expected benefits of a future expected benefit stream in the present interest group’s net benefit
calculation.
Inflation poses a final threat to a scheme of perpetual benefits. If legislators fund programs in
perpetuity at constant nominal levels, a few years of high inflation rates, or several years of more modest
rates, can seriously erode the value of constant nominal funding levels. Hence, to give credence to their
own commitments in the enactment of perpetual program funding today, legislators must commit to
keeping past programs at constant real funding levels. We believe that adjustments for inflation along
with reactions to changing factor costs, give the budgetary process its distinctive incremental character.10
Here we compare two strategies: the single-period budgeting strategy and the “perfect” multiperiod budgeting strategy. The "perfect" multi-period budgeting strategy refers to the case where
groups expect the real value of their program budgets to continue indefinitely into the future with a
probability equal to one.11 Multi-period budgeting enables a legislator to exercise partial control over the
budget, not just during his time in office but also during the years after he leaves the legislature. His
control is only partial, because to maintain interest-group expectations of a constant real budget, a
10
Research on legislatures has concentrated on the function of institutions to solve various problems of disequilibrium
(Shepsle and Weingast 1981) and post-contractual opportunism (Weingast and Marshall 1988). Landes and Posner
assume that rules stabilize expectations. And, the operation of institutions would be consistent with a legislative process
that remains largely incremental. By associating the meta-decision of incremental versus nonincremental choice with
exogenous variables, however, we suppress institutional effects on public policy. But our analysis does not, nor do we
intend it to, settle the question whether institutions “matter.” We believe that they do, and it is a close question whether we
would find the kinds of regularities in the data that we do find, were the process we study not as confined as it is with
longstanding procedural rules.
11
Section IV briefly discusses the case of imperfect multi-period budgeting (where group members expect benefits to
7
constant level of services, he must respect similar multi-period decisions of previous legislatures. So, he
controls only an amount equal to the growth of the current budget over the real value of the previous
year's budget. Single-period budgeting, by contrast, gives the legislator complete control over the entire
budget, but only during a single term in office.
The basic model examines the legislator’s short-run decision. He does not consider the effect of his
current year’s strategy choice on his future income from his enactments in subsequent terms.12 The
legislator's overarching decision problem is which decision-making strategy--single- or multi-period
budgeting--to follow. To make this choice he compares the utility he derives from the optimal choice of
program budgets under each strategy and chooses the strategy (and its implied allocations) that gives him
the higher utility.
To compare the single- and (perfect) multi-period strategies, we begin with a legislator who enters
office at year t . Because we are analyzing the case of "perfect" multi-period budgeting, that method
must have been in existence for some time before the current legislator's tenure began, for group
members to believe that their program budgets will continue with certainty up to infinity. With each
subsequent term in office, the legislator can either continue multi-period budgeting or switch to singleperiod budgeting.
A. Assumptions of the Model: The legislator's decision problem is to determine the total budget for
year t , Bt , given that the budget for year t − 1 is Bt− 1 . We assume that there was only one politically
active interest group in year t − 1 , group i , and that the legislature allocated the total budget of that
year through a multi-period promise to a single program, pi , that benefits group i . That is,
continue with a probability less than one) .
12
Section IV briefly highlights the results of the case where the legislator chooses the strategy that maximizes his utility
8
Bt − 1 = bit − 1 . ( bit − 1 is the budget of program pi at year t − 1 .) At year t a new group, j , organizes
and demands funding for a new program, p j . The legislator receives the same payment from either
interest group for every dollar spent on the program that benefits it.13 The legislator's decision problem
is to determine what proportion of the total level of funding, Bt , to allocate to each program.14 The
legislator could use one of two strategies. The first is single-period budgeting, where the decision about
how to divide the total budget between the two programs occurs every single year. The second is
perfect multi-period budgeting, where the legislator takes the real budgets of the previous year as given,
because they are already promised to members of group i . His decision problem then reduces to
dividing the excess of the year t total budget over the year t − 1 real budget, between the two
programs.
The legislator's income equals the payment he receives from both interest groups, which in turn
equals a percentage, s , equal to one hundred, of the level of funding allocated to the program that
benefits a group.15 The legislator charges both interest groups the same percentage and is thus
indifferent between expanding bi by a certain amount and expanding or initially funding b j by the same
amount. The legislator's utility increases at a decreasing rate as the result of any increase in the income
over his full tenure in office.
13
We assume that the number of interest groups is large enough to preclude any market power or bargaining among them.
The use of one interest group in this version of the model helps to highlight the model's conclusions by suppressing its
complexities. Therefore, we treat the two interest groups in this model as if they were two among a very large number, to
maintain the relevant assumptions.
14
Funding programs for ideological reasons does not exist in this simple version of the model. For a discussion of the
effects of ideological animus in budgeting, see Section IV.
15
s equals one because the legislator maximizes his income, and hence charges the highest possible s . Group members
agree to pay a value of s = 1 , because they keep the surplus, resulting from their downward sloping demand schedules for
budget money. These assumptions grow out of the legislator’s monopoly power and the perfectly competitive environment
of interest groups. In legislatures with more than one member, which require a majority vote of a quorum to allocate funds,
each majority member's share would be smaller than one, and the sum of individual payoffs also might be less than one.
9
he receives from both groups.
The legislator's utility varies inversely with the level of political opposition he faces, which, in turn,
is a linear function of real increases in the deficit level.16 The legislator's budgetary strategy does not
affect political opposition, because we assume that for any total spending level, political opposition
remains the same whether the legislator adopts single or multi-period budgeting.17 We also assume that
the marginal cost of increasing the deficit (and political opposition) increases with the deficit level. (The
legislator derives negative, increasing (in absolute terms) utility from the additional political opposition
that he faces.) We define the deficit level at year t as the difference between total spending, Bt , and
total revenues, Rt . Revenues are exogenous to the budget determination process, and we assume that
they grow in real terms.18 Rt = Rt − 1 (1 + i )(1 + g ) , where
i
is the inflation rate, g is the real rate of
The model also assumes the absence of price discrimination.
16
Opposition results from an increase in the deficit or in taxes to finance greater spending. As long as either one increases
opposition, our results are the same. Hence, we make taxes grow at a positive real rate and let additional spending be
reflected in a larger deficit, which in turn results in higher opposition.
17
Political opposition is the same under single- or multi-period budgeting, because we assume that the budget never
shrinks: a dollar once spent to benefit an interest group may be taken from it to finance other programs, but it will not be
returned to taxpayers. In real life one would expect multi-period budgeting to crowd out spending for other programs in
the future. One also would expect more opposition from interest group i (under single-period budgeting, because the
legislator broke a multi-period budgeting promise) or from interest group j (under multi-period budgeting, because the
legislator denied it budget funds). In the simple model we can assume that either group will exercise the same amount of
opposition per dollar of budget funds, and hence the total opposition (from both groups) will be the same regardless of how
the funds are split. The extended model (Section IV) analyzes dissatisfaction with breaking a multi-period budgeting
promise, through constituents’lower expectations that the legislator will keep future promises.
18
The extended model discussed in Section IV explores the case where revenues decrease in real terms. Even though the
model here assumes that revenues grow or remain constant in real terms, it could still explain a situation where revenues
decline. Political opposition would increase only if the deficit increases at a level that outweighs the reduction in revenues,
even though revenues decrease. The decrease in revenues will not affect opposition from group i or j , per dollar of
budget funds; opposition is affected by the budget level, which will be the same whether financed by increased revenues or
deficits. If the deficit were to increase enough to allow for what would have been constant real revenues (expenditures are
equal to revenues plus the deficit), no additional political opposition will result. Hence, as far as political opposition is
concerned, we treat the part of the deficit that compensates for lower real revenues as if it were actual revenues, so it does
not affect political opposition. Therefore, all the results of the model will continue to hold in the case of decreasing
revenues.
10
revenue growth, and Rt− 1 is total revenues in year t − 1 . The legislator faces no additional opposition at
year t as long as the deficit level for year t is less than or equal to the real value of the deficit in the
previous year. We call this level the "acceptable" deficit level. The acceptable deficit level,
d t = d t − 1 (1 + i ) , is the maximum level that imposes no additional political cost on the legislator.19 For
any deficit level greater than d t , the legislator faces additional political opposition that reduces his
utility.
B. Results of the Basic Model: The legislator's problem at year t is first to determine the total level of
spending that maximizes his utility. Then, he decides how to divide this total between groups i and j .
We first show the legislator’s income and present separately the utility maximizing budget level for each
strategy, SBP and MBP. Then, we discuss how the legislator divides the total budget between both
groups. Finally, we report the conditions that lead the legislator to prefer one strategy -- single- or
multi-period budgeting -- to the other.20
1. The legislator’s income and the resulting total budget under each strategy.
Benefits to group members in year t under single-period budgeting equal or exceed (through
surplus) the total budget received by (the programs) both groups (support). Hence, they will pay the
legislator an amount equal to s , the price he charges for every dollar of budget money, times the total
allocation received. Therefore Yt = Bt . The utility-maximizing budget level under the single-period
budgeting strategy is given by:
19
The legislator gains no additional votes for any deficit level below the acceptable one, because a lower deficit, for the
same amount of revenues, implies lower spending. Lower spending brings forth more opposition from the affected groups.
The additional opposition offsets the reduction in opposition resulting from reducing the deficit below the acceptable level.
20
Derivations using more general utility functions yield similar results, which are available from the authors upon request.
11
Bt =
Bt − 1 (1 + i ) + Rt − 1 (1 + i ) g +
[Bt − 1 (1 + i ) +
2
Rt − 1 (1 + i ) g ] +
2
2α
β
,
where α and β are the weights of income and political opposition in the legislator's utility
function, respectively.
The total budget level, Bt , is greater than the real value of the total budget of the previous year,
Bt− 1 (1 +
i).
The greater is α β , the legislator’s valuation of additional income relative to the higher
political opposition resulting from expanding the deficit, the larger is the increase in year t ’s total
budget. This result shows that if α and
i
are equal to zero, the total budget reduces to the budget of
the previous year plus the increase in revenues -- the deficit will not increase. Also, if the real value of
total revenues is constant ( g = 0) , the total budget will continue to expand at a rate greater than the
inflation rate, leading to a constantly increasing total deficit.21 This result is consistent with Wagner's
law, which predicts public sector spending that expands faster than the real growth rate of the economy.
Furthermore, the larger is g , the real rate of revenue growth, the larger is the increase in the total budget
for year t .
Under multi-period budgeting, group i , which received bit − 1 = Bt − 1 in year t − 1 , will expect the
same real amount at year t and will not give the legislator any income for funding program pi at a level
less than or equal to bit− 1 (1 + i ) . Group j , which received nothing in year t − 1 , will pay the legislator
for any positive level of funding. Because the legislator operates under "perfect" multi-period budgeting,
21
If revenues do not increase in real terms, the budget will still expand because the legislator is always trying to balance
the marginal utility from additional income with the marginal cost of increased political opposition. Because marginal
utility of income is always positive, the legislator’s utility maximizing deficit level will lead to a positive level of political
opposition. That is, it always will exceed the ideal deficit level (the deficit of the previous year adjusted for inflation), and
hence the real budget will grow even if revenues do not.
12
any funding level allocated to an interest group in year t will continue with certainty to be allocated in
real terms to the same group in perpetuity.
Group i 's payment to the legislator equals one hundred percent of the discounted present value of
the increase in its stream of funds. Group i was already expecting a constant level of services:
bit− 1 (1 + i ) in year t , bit− 1 (1 + i ) 2 in year t + 1 , bit− 1 (1 + i ) 3 in year t + 2 , and so forth. (As we note
earlier, the inflation adjustment represents at least part of the “increment” in incrementalism.) The
discounted value of this stream of expected benefits equals bit− 1 (1 + i )
1+ r
, where r is the discount
r− i
rate -- the marginal rate of time preference.
If the legislator decides to give group i a budget for year t equal to bit > bit − 1 (1 + i ) , the income
stream group i expects to receive will equal bit in year t , bit (1 + i ) in year t + 1 , bit (1 + i ) 2 in year
t + 2 , bit (1 + i ) 3 in year t + 3 , and so forth. The discounted value of this stream of expected benefits
equals bit
1+ r
. Therefore, the increase in the discounted present value of group i 's income, and the
r− i
legislator's income received from group i , equals [
bit − bit − 1 (1 + i ) ]
1+ r
.
r− i
Group j 's budget equals b jt . Similar to group i , group j expects its program to receive a
budget of b jt in year t , b jt (1 + i ) in year t + 1 , b jt (1 + i ) 2 in year t + 2 , b jt (1 + i ) 3 in year t + 3 , and
so forth. The payment group j is willing to make to the legislator equals b jt
[
legislator's total income equals bit + b jt − bit − 1 (1 + i )
1+ r
. Therefore, the
r− i
] 1r +− ri , which equals [Bt − Bt − 1 (1 + i )] 1r +− ri .
( Bt − 1 = bit − 1 because at year t there was only one interest group, i ).
The budget that maximizes the legislator’s utility under multi-period budgeting is given by:
13
[Rt − 1(1 + i )g ]2 +
R (1 + i ) g
Bt = Bt − 1 (1 + i ) + t − 1
+
2
2α
2
β
.
Therefore, in the multi-period budget determination problem, the legislator chooses a total budget
i ) , the real value of the total budget of the previous year.
Bt ≥ Bt − 1 (1 +
As before, the greater is α β ,
the legislator's valuation of additional income relative to the higher political opposition resulting from
deficit expansion, the larger is the increase in the year t total budget over the budget in year t − 1 . If α
and
i
equal zero, the total budget again reduces to the budget of the previous year plus the increase in
revenues--the deficit will not increase. Also, if the real value of total revenues is constant ( g = 0) , the
total budget will continue to expand at a rate greater than the inflation rate, leading to a constantly
increasing total deficit. The greater is g , the real rate of revenue growth, furthermore, the larger is the
increase in the budget.
These are the same qualitative conclusions reached under the case of single-period budgeting. Yet,
other things equal, the total budget tends to expand more under multi-period than under single-period
budgeting.22 This result seems intuitive, because the legislator’s marginal benefit from increasing the
22
To find out which strategy will yield a higher total budget, we subtract the total budget under the single-period budgeting
strategy, BtS , from the total budget under the multi-period budgeting strategy, BtM , and we get:
BtM − BtS = Bt − 1 (1 + i ) +
−
=
Rt − 1 (1 + i ) g
2
1
B (1 +
2  t− 1
−
Rt − 1 (1 + i ) g
[Bt − 1 (1 + i ) +
2
[Rt − 1 (1 + i ) g ]2 + 2α β
+
2
Rt − 1 (1 + i ) g
]2 +
2α
−
Bt − 1 (1 + i )
2
β
2
i ) + [Rt − 1 (1 + i )g ]2 +
2α
β
−
[Bt − 1 (1 + i ) +
Rt − 1 (1 +
i )g ]2 +
2α

β .

To find the sign of the term in the square brackets, because all three terms are positive, we square the sum of the first two
terms and square the third term, subtract the squares from each other, and find the sign of the resulting term. The resulting
term equals:
14
budget by one dollar is greater under the multi-period than under the single-period strategy for every
dollar greater than the real value of the previous year’s budget. Hence, when equating marginal benefit
to marginal cost, the legislator will support a higher total budget in the case of the multi-period
budgeting strategy.
2. The allocation of the total budget between interest groups i and j under each strategy.
The legislator is indifferent, under single-period budgeting, between giving a dollar to group i and
the same amount to group j . There are no issues of credibility involved, and each group will pay him
the same amount. Therefore, we assume that the legislator will split the total budget between i and j
according to a uniform distribution over [0,1]. The share of i is θ , and the expected value of θ = 1 2 .
[
In the case of multi-period budgeting, the legislator's income equals bit + b jt − bit − 1 (1 + i )
] r1 +− ri ,
where we constrain bit to be greater than or equal to bit− 1 (1 + i ) . That is, we constrain the budget of
program pi in year t to equal at least its budget at year t − 1 adjusted for inflation. Any choice of
program pi 's budget less than bit− 1 (1 + i ) lowers the probability with which group i expects its
program to receive, at any year x greater than t , the same real amount of benefits it received in year
x − 1 . This reduction would lower the group's expected flow of benefits and hence the legislator's
income. Funding pi at a level less than bit− 1 (1 + i ) harms the legislator without any extra benefit,
because we assume that he receives the same payment for every budget dollar spent, regardless of which


2 2α
2 Bt − 1 (1 + i )  Rt − 1 (1 + i ) g +
+ Rt − 1 (1 + i ) g ,
β




[
]
which is greater than zero. Therefore, BtM is greater than BtS .
15
group benefits from the payment.23 To maximize his utility, therefore, the legislator commits to funding
pi at a level equal to bit− 1 (1 + i ) , and he divides the excess budget between both programs, pi and
p j .24 Because the legislator is indifferent between funding both programs, we assume that his decision
to divide the excess budget is again according to a uniform distribution on [ 0,1] . We denote the
percentage of the extra budget funds that group i receives as θ ( 1 − θ is the percentage that group j
receives). The expected value of the budget that groups i and j receive equals
bit − 1 (1 + i ) +
1
2
[Bt − bit − 1 (1 + i )]and 12 [Bt − bit − 1 (1 + i )], respectively.
of funding between bit− 1 (1 + i ) and bit − 1 (1 +
i ) + [Bt −
of funding between zero and [Bt − bit − 1 (1 +
i )].
bit − 1 (1 +
i )]= Bt .
Group i receives some level
Group j receives some level
Because the increase in the budget is small relative to
the total budget, variations in the budget that either group receives are smaller under multi-period
budgeting than under single-period budgeting.25 We would observe more incremental (nonincremental)
outcomes, therefore, if the legislator chooses multi- (single-) period budgeting.
3. Comparison of single- and multi-period budgeting.
To predict the legislator's choice between both strategies, we solve for the real rate of revenue
growth, g , that makes equal the legislator's total utility derived from the optimal choice of variables
under both strategies. We choose g to illustrate how the growth rate of total revenues affects the
23
It harms the legislator by reducing his income from both groups, as they observe his breaking the multi-period promise to
group i .
24
i ) only if he chooses multi-period budgeting; he always can choose
with a budget lower than bit − 1 (1 + i )
The legislator commits to the funding level bit − 1 (1 +
single-period budgeting and fund pi
25
We use this characteristic in Section III to identify years when the legislature is likely to adopt more instances of singleperiod budgeting for a set of programs and fewer instances of multi-period budgeting.
16
legislator's choice between strategies.26 We call the value of g that makes the legislator indifferent
between both strategies the critical g , g * . The legislator chooses multi-period over single-period
budgeting for any g > g * , because his total utility is higher. He remains indifferent between both
strategies if the real rate of revenue growth equals g * . So, we assume that he chooses multi-period
budgeting then also. Any growth rate less than g * leads the legislator to choose single-period over
multi-period budgeting. The critical g is given by:
g* =
Bt − 1 ( r − i )
α
−
.
Rt − 1 (1 + i ) 2 βBt − 1 Rt − 1 (1 + i )(1 + r )
There is no restriction on the sign or the value that g * may take. A relatively high α makes g *
negative. A negative g * means that any g ≥ 0 will lead the legislator to choose multi-period budgeting.
And, because we assume that g ≥ 0 , the legislator always chooses multi-period budgeting. We could
expect this result because a very high α means that the legislator values present income very highly
relative to political support. Hence, he would pay little attention to the increase in opposition associated
with the higher (multi-period) budget. A high β , by contrast, makes the value of g * positive, yet less
than one. Depending on the actual value of g * , which is a function of the discount rate and the inflation
rate, the legislator may end up choosing single-period budgeting, because the cost (disutility) that he
bears if opposition increases is too high. Given the very high marginal cost in terms of opposition, β , no
expansion of the budget to yield greater income may be worthwhile, and the legislator will choose singleperiod budgeting.
Table 1 summarizes the results of the basic model. First, the legislator is more likely to choose
26
We use the same analysis for the other exogenous variables. The results are similar, and therefore we do not repeat the
17
multi-period budgeting if the previous year's revenues are large. For a not very large α β , the larger is
Rt− 1 , the smaller is the value that g * takes. A large value for the revenue of the previous year,
resulting in a lower g * , other things equal, makes the additional revenue of the current year bigger
(because revenue in one year is equal to the real value of revenue of the previous year times 1+ g ),
hence making it more attractive for the legislator to choose multi-period budgeting. That is, the larger is
the increase in revenues, the more attractive multi-period budgeting becomes.
Second, for a not extremely large α , the larger is Bt− 1 , the larger is g * . A very large Bt− 1 , other
things equal, makes commensurately large the amount of funds group members already expect to receive.
Thus, they are less willing to pay the legislator extra funds for receiving a budget they already expect.
They are only willing to pay the legislator for new promises, in addition to the real value of Bt− 1 . A
large Bt− 1 , other things equal, reduces the extra money available for new promises. This reduction, in
turn, makes multi-period budgeting less attractive, which is reflected in a larger g * .
Third, the larger is the discount rate, r , the greater is the value that g * takes. A large discount
rate, other things equal, reduces the discounted present value of the stream of future benefits to group
members. This reduction, in turn, diminishes the legislator’s income under multi-period budgeting. The
resulting lower income makes multi-period budgeting less attractive, requiring a larger value of g for it
to be preferred to single-period budgeting.
Fourth, the larger is the inflation rate the smaller is g * . Stated differently, the model predicts that
increases in the rate of inflation, ceteris paribus, will increase instances of multi-period budgeting and
diminish instances of single-period budgeting. We have no cogent verbal interpretation of the reasons for
analysis here.
18
this prediction. But our tests fail to disconfirm it. And, while we might advert to an exogenous money
illusion, we prefer not to do so, and to remain with the model as stated.
Fifth, the larger is α β , the smaller is g * , making it more attractive for the legislator to choose
multi-period budgeting. A large α β indicates a strong preference among legislators for marginal
income in relation to the additional disutility that the associated increase in the deficit provokes.
Finally, the larger is g , the more likely is the legislator to choose multi-period budgeting. An
increase in the real growth of revenues makes larger the total amount of budget funds that he can allocate
to programs that benefit interest groups, without increasing political opposition. Hence, a large real
growth rate in revenues makes the increase in funds under multi-period budgeting, and hence the
legislator's income, larger. (The larger rate of revenue growth makes the legislator's income higher under
single-period budgeting as well, but the additional income from every dollar of budget money is greater
under multi-period budgeting, because it includes the discounted value of the future income stream.) The
larger the growth in the total budget, therefore, the more likely is the legislator to choose multi-period
budgeting.
If the legislator does choose multi-period budgeting, the income he receives must be greater than
the income he would have received from single-period budgeting. Single-period budgeting always leads
to a smaller budget, a smaller deficit, and consequently, less political opposition.27 For the legislator to
be willing to bear more political opposition, he must receive more income from interest groups to
compensate for the additional cost.
27
One could argue that the lower opposition increases the legislator's expected term in office and hence his income. While
that may be true, it does not change our results, because we are only comparing short-run decisions, where by definition the
legislator does not calculate future periods’earnings. We expect, in any case, that the planning horizon of most legislators
does not extend much beyond the next election.
19
The short-run model of budget determination also predicts that the legislator grows more willing to
adopt multi-period budgeting as his winning margin or the safeness of his district increases. This
hypothesis is a direct result of two predictions of the model. First, multi-period budgeting always leads
to a larger budget than does single-period budgeting. Second, a larger budget leads to greater
opposition. The more secure the legislator is in his position, however, the less concerned he becomes
about added opposition resulting from the larger budget associated with multi-period budgeting. The
safer the legislator's seat, therefore, the more likely he is to choose multi-period budgeting.
III. Empirical Analysis
Empirical studies of the budgetary process ordinarily rely on time series regressions, where the
regressand and regressors are some combination of agency requests, appropriations, and their lagged
values.28 Regression specifications in these studies are atheoretical, and inference also remains suspect
because of statistical problems. Tohamy, Dezhbakhsh, and Aranson (1999) note several such problems.
They demonstrate that budget data series are often nonstationary, possibly containing unit roots.
Regressing one nonstationary series on another induces a spurious regression wherein standard t and F
tests may be invalid.29 Many of the inferences in empirical studies of the budgetary process are
predicated on these tests. Furthermore, the estimation procedure these studies use involves subjectively
locating the "most likely" point for a change in parameters of the relationship. The stability of such
relationships is examined through ad hoc procedures. These studies also use the coefficient of
28
See, e.g., DDW (1966a; 1966b; 1971), Gist (1974), and Ripley and Franklin (1975).
29
See, e.g., Granger and Newbold (1974) and Banerjee, et al. (1993, pp. 76-83). This is not a direct criticism of those
researchers who performed prior studies of the budgetary process, because some of these studies predate recent findings
about nonstationary regressions.
20
determination, R 2 , as the criterion for choosing among models with different break points. But the use
of R 2 for this purpose remains inadequate, particularly in the presence of serial correlation--a real
possibility for the models that these studies estimate (Pesaran 1974).
The method we use here is designed to serve three purposes: to allow an analysis of non-stationary
budget data; to provide a way to detect nonincremental outcomes without relying on spurious
regressions, R 2 , or subjectively chosen break points; and to accommodate testing the hypotheses we
derive in Section II.
A. Testing Marginality of Budgetary Outcomes: We provide an alternative method for measuring
incremental and nonincremental changes in the U.S. federal budget. We calculate the real budgetary
growth rate of each of the sampled agencies, for every year in the period under study. Then, we test for
the marginality of the budgetary outcome using this definition: if an agency’s real budget growth rate for
a given year is not significantly different from (either larger or smaller than) zero, then we classify the
treatment of its budget as incremental for that year. Otherwise, we classify the treatment of its budget as
nonincremental. This classification reflects the definition of multi-period budgeting, where group
members continue to expect the same real value of program budgets. We also measure the regularity of
incremental outcomes by identifying agencies that display incremental budgeting for four or more
consecutive years.30
To summarize our method, we identify agencies with budget growth rates significantly different
from zero by first drawing a confidence band [ − c ′
σ∃st ; + cσ∃st ] around zero for each year.31 σ∃st is the
30
Tohamy, Dezhbakhsh, and Aranson (1999) detail the method for identifying agencies whose budgets display incremental
(or marginal) outcomes. They construct eight different measures of incrementalism, four of which identify regularity over
an extended period of time.
31
Notice that we are imposing the hypothesized zero growth rate in our calculation.
21
cross agency standard deviation of the annual percentage change in the budget.32 c and c ′are the
critical values from the standardized empirical distribution (one for each size) of the annual budget
change.33 Any agency with a growth rate that falls outside (either above or below) this band indicates a
deviation from an incremental outcome in a given year. Then, we sum cases of deviations from an
incremental outcome over all of the agencies for each year. The resulting variable, Dt , represents the
number of agencies in the sample that deviate from a marginal, incremental outcome in a given year. We
use eight different measures of Dt : DI1t , DI 2t , DI 3t , DI 4t , DR1t , DR2t , DR3t , and DR4t . The
second letter in the variable name, I or R , indicates whether the variable measures deviations from the
incremental one year outcome, or the regular outcome, where the budget is incremental for four or more
consecutive years, respectively. A value of one for the subscript indicates a small positive forty percent
band and a value of two indicates a two-sided forty percent band. Similarly, a value of three for the
subscript indicates a large positive 45 percent band and a value of four indicates a two-sided 45 percent
band. By constructing Dt from a cross sectional inference, we avoid the statistical problems mentioned
earlier, caused by identifying nonincremental deviations through time series regressions with
nonstationary variables.
The model presented earlier suggests several variables that may affect Dt . We test the hypotheses
of the model by regressing the variable Dt on the set of exogenous variables that the model suggests may
32
Small agencies may experience exceptionally large growth rates, particularly soon after their inception. To prevent such
outliers from skewing the results, we first divide the sample into three sets according to agency size, in the year 1970 (the
midpoint of the time series). Then we construct three different bands for the three different sizes, using the three different
sample variances.
33
Because the distribution is far from normal, instead of using standard normal critical values for calculating the statistical
band, we standardize the distribution of growth rates by subtracting from the median and dividing by the standard
deviation of the pooled rates. Notice that the two critical values are different from each other because they are derived from
a skewed distribution. We use four pairs of critical values: plus or minus forty percent around zero, plus or minus 45
22
influence budgetary outcomes. We use this functional relationship to test these hypotheses:
(1)
where
i,
Dt = g ( i t , rt , g t , ∆d t , du∆d t , Democratst , εt ) ,
r , g , and ∆d are the inflation rate, the discount rate (marginal rate of time preference), the
real rate of revenue growth, and the change in the nominal value of the deficit (the change in the nominal
value of revenues minus the nominal budget), respectively. 34 εt is an error term.
The variable du∆d is the product of a binary variable and ∆d . The binary variable takes a value of
one if the share of the deficit in the total budget is greater than ten percent for two consecutive years.
Otherwise it takes a value of zero. The justification for creating this dummy variable is that budgeters’
attitudes toward the deficit change over the sample period. In the early period the deficit is not very
large as a share of the budget and a majority of budgeters appear to have ignored it. Hence, budget
expansion was mostly deficit financed, which rendered the real rate of revenue growth a non-binding
constraint. In the later period the deficit becomes larger and more noticeable, and hence the model’s
assumption about the real rate of revenue growth imposing a constraint on budgetary expansion becomes
applicable. If this suggested description of the budgetary process is accurate, the du∆d t variable will
have a positive effect on the total number of deviations from incremental changes, reflecting a positive
shift in the deficit coefficient in the later period.35 We refer to this variable as Persistent-Deficit.
The political party variable, Democrats , takes a value of three if the Democrats control the White
percent around zero, zero to forty percent and zero to 45 percent.
34
We use the deficit level to test the effect of both revenues and expenditures on the number of deviations from
incrementalism. The coefficients for revenues and the total budget have predicted opposite signs, which allows us to test
their effect jointly (to avoid multicollinearity in the independent variables). We also use differences rather than levels, to
reduce the time trend in that variable.
35
Notice that the small size of the sample for the later period does not allow a meaningful estimation of separate
relationships for the two periods.
23
House, Senate, and House of Representatives. The variable takes a value of two, one, or zero if
Democrats control two, one, or none of these institutions. Our model does not directly predict the effect
of this variable on our measure of deviations. We use it here as a proxy measure for α / β in Table 1. It
tries to measure different preferences for additional income relative to increased political opposition by
using party affiliation of both the president and members of Congress. We nevertheless expect
Democrats to have a positive coefficient, because our data set contains the budgets of nondefense
agencies only.
Because Dt is an integer-valued variable, we use the Poisson regression to estimate (1). 36
Consider the Poisson probability model:
(2)
e − λt ( X t ) ( λt ( X t )) dt
,
Prob( Dt = d t ) =
dt !
where d t = 0, 1, 2,..., T , λt(.) is finite and positive and Xt is a set of explanatory variables. We wish
to estimate E(DtXt) = λt(Xt), which is the moment conditional on the explanatory variables.37
Following statistical convention, we assume that λt(.) is log-linearly dependent on the model’s
explanatory variables. Therefore, we specify the expected regression relation:
(3)
ln E ( Dt ) = β1 i t + β2 rt + β3 g t + β4 ∆d t + β5 du∆d t + β6 Democratst ,
where E denotes expectations conditional on the regressors.
Based on the theory’s predictions (see Table 1) we offer several conjectures about equation (3).
36
See, e.g., Hausman, Hall, and Griliches (1984).
37
The Poisson model imposes the restriction that the conditional mean and variance of the count variable are equal.
Cameron and Trivedi (1990) explain that the model is inappropriate if the data exhibits overdispersion--variance exceeding
the mean. In such cases one should use the negative binomial model or other mixture models, as discussed in Gurmu and
Trivedi (1994). We apply a test proposed by Cameron and Trivedi, and the results suggest no overdispersion in our data.
So, we proceed with the Poisson model.
24
By definition, the larger the variable Dt , the more deviations there are from marginal outcomes, or
equivalently, the more the occurrences there are of single-period (nonincremental) budgeting. The lower
the value of Dt , the more the occurrences there are of multi-period (incremental) budgeting, where
legislators promise to maintain for interest groups a constant stream of benefits over time. The model
then suggests these hypotheses: first, the larger the inflation rate, the more likely will legislators adopt
multi-period budgeting and the lower will be Dt ; second, the larger the discount rate, the more likely
will legislators adopt single-period budgeting and the higher will be Dt ; third, the higher the real rate of
revenue growth, the more likely will legislators adopt multi-period budgeting; fourth, the larger the
deficit level of the previous year (and the change in the deficit as well), the more likely will legislators
adopt single-period budgeting. (The larger the total budget of the previous year, and the smaller the
revenues of the previous year, the more likely will legislators adopt single-period budgeting). These
hypotheses predict that β1 , and β3 will be negative and β2 and β5 will be positive. The sign of β4 is
not relevant to the model because it measures the effect of the deficit throughout the sample period,
when we know that its role as a constraint on spending only becomes relevant during the later period.
Even though the model does not directly predict the sign of β6 , we expect party affiliation to affect
program funding preferences. The increasing strength of a party would lead to a shift in program
funding to more of the party’s preferred programs and will therefore lead to more deviations from
incremental outcomes.
B. Empirical Results: We estimate the empirical model using real rather than nominal data, to reflect
the model's emphasis on the legislator’s budgeting to maintain a constant level of service--the definition
of incrementalism we use. We only use outlays data, because the emphasis of the model is on the actual
25
budget funds that programs receive and not on agency requests, or appropriations.38
We collect data on all 115 nondefense agencies that DDW (1971) study.39 We expand DDW's
sample period to 1946-1994, to provide a stronger asymptotic justification for the statistical results. We
drop the agencies for which data are not reported for some of the years. For agencies that merge after
1963, we add their expenditure data for the years before the merger and create a new time series; then,
we use the resulting series as if the two agencies were one throughout the period. We exclude agencies
that merge with other agencies that are not in the original sample. We add zeros for agencies that come
into existence after 1946 or go out of existence before 1994, to reflect that they receive no funds in the
years described.40 The resulting number of series is 93. We use the twelve-month Treasury Bill rate to
measure the discount rate. We also use the Consumer Price Index to transform expenditure figures into
real figures.
We estimate equation (3) for each of the eight measurements of the dependent variable Dt , using
the maximum likelihood (ML) estimation method for the Poisson model. Tables 2 and 3 summarize the
results. The dependent variables in Table 2 reflect only year-to-year changes, leading to placement
outside the bands. . We denote these as the incremental variables. The dependent variables in Table 3,
by contrast, measure regularity of the incremental outcome, as a small change that last for at least four
years, which we denote as the regularity variables.
Tables 4 and 5 use the same set of independent variables used in Tables 2 and 3, together with
38
We believe that outlays are a better measure than are appropriations because appropriations are not always given for one
year and future appropriations are sometimes used as a way to appease agencies that do not receive a large increase in their
budgets in a given year.
39
We use data from the Budget of the United States Government: Fiscal Years 1946-1996.
40
This process will tend to increase the extent of reported incrementalism.
26
three additional dummy variables: ElecYear, PresSwitch, and CongSwitch, which indicate a presidential
election year, a switch in the party affiliation of the president, and a switch in the majority party in
Congress, respectively. Separating the House and the Senate switch dummies gave similar results. The
regression equation is given by:
(4)
ln E ( Dt ) = β0 + β1i t + β2 rt + β3 g t + β4 ∆d t + β5du∆d t + β6 Democratst
+ β7 ElecYeart + β8 Pr esSwitcht + β9 CongSwitcht
We expect all three dummy variables to have a positive effect on deviations from incrementalism
( β7 , β8 , and β9 are positive) because they indicate a change in program priorities in the executive or
the legislative branches of government. A change in program priorities will lead to greater deviations
from an incremental path as legislators reduce funding of some existing programs, so that they can
increase funding to their (or their constituents’) preferred programs. Both groups of affected programs
are them more likely to experience a nonincremental change in their funding level.
The estimated coefficients of the discount rate, the deficit, the presidential election year, and the
switch years are significant and positive, again as expected, in all variants of the model. The estimated
coefficient of inflation is significant and negative, as expected, in all variants of the model. These results
strongly support the model’s predictions.
The estimated coefficient of the real rate of revenue growth is positive and significant for the
measures using a positive band for both the incremental and regularity variables (Tables 2 and 3). This
result is opposite to the predicted effect. It arises, we believe, because the real rate of revenue growth
matters in the model. This indicates an increase in revenue, and therefore legislators could fund
programs at increasing levels, without affecting political opposition. Using the positive band to measure
a marginal outcome results in a positive effect of the growth rate on deviations because a low growth
rate of revenues makes it difficult to keep previous real funding levels. The reduction in the real budget
27
(to make sure that funding to most programs is close to its previous level) would be measured as a
nonincremental change when the positive band is used, even if the nominal budget level were constant.
This result disappears for the two sided band measures. It also disappears when we use the three
dummy (political) variables to measure the effect of changing program priorities. The estimated
coefficient of the real rate of revenue growth becomes insignificant throughout. The positive significant
estimated coefficients in Tables 2 and 3 merely may result from the omission of the dummy variables in
equation (3) and the resulting mis-specification of the model.
While this discussion may explain why the significant positive coefficient for the real rate of
revenue growth does not provide evidence against the model, we still need to discuss the insignificant
coefficients of the real rate of revenue growth in the estimation results of equation (4). The insignificant
coefficients for all eight measures may arise from our inability to separate the sample period into two
different subperiods. In the earlier period, revenues, their growth rate, or the size of the deficit do not
place a binding constraint on budget expansion because politicians relied on deficit-financed spending
without facing any political opposition. (Politicians used deficit financing, inter alia, to stimulate the
economy.) In the later period, deficit financing generates political opposition and revenue growth
becomes a restraining factor for additional expansion of the budget. The later period is a better
reflection of the model’s assumptions regarding the increase in political opposition as a result of an
increase in the deficit.
The results show high overall significance for all eight measures of incrementalism, as indicated by
the log-likelihood function. The results are slightly stronger for the positive band compared with the two
sided band measures. This additional strength indicates that incrementalism is best described as having a
real rate of budget growth that is greater, without being much larger, than zero. Any negative deviations
from a zero real rate of growth of budgetary benefits, even if they are not statistically different from zero,
28
appear to constituent groups as broken incremental promises. So, they hurt the legislator’s ability to
receive compensation from the interest groups affected, groups with which he might wish to deal. The
results also show that the regularity measures outperform the incremental measures. Both results jointly
indicate that we best may measure incrementalism using a positive band around zero real growth for a
given agency for at least four consecutive years.41
IV. Extensions, Implications, and Concluding Remarks
We extend the model to include the legislator’s choice of budget strategy in the presence of
ideologically satisfying projects that he supports for their intrinsic value. The literature refers to this kind
of legislative activity, which detracts from payoffs to the legislator’s clients, as “shirking” (see, e.g., Kalt
and Zupan 1984; Kau and Rubin 1979; 1993; Nelson and Silberberg 1987). The extended model shows
that ideologically motivated legislators will not choose multi-period budgeting as often as will
nonshirking legislators.
We also extend the model to allow the legislator to maximize his utility over his full tenure in
office. This extension leads to imperfect multi-period budgeting, because group members adjust their
expectations downward, as legislators sometimes break their multi-period promises. The results of the
basic model are reinforced.
Two additional results also emerge. First, the legislator’s credibility (the probability with which
interest group members expect him to keep his multi-period budgeting promises) affects his choice of
strategy. Indeed, the increase in the legislator’s credibility, either over time or compared with previous
41
This may be the result of the basic model’s hypothesis of “perfect” MPB in the theoretical model. One way of checking
the robustness of the results is to test the predictions of a model that introduces uncertainty and less than perfect MPB.
29
legislators, makes him more likely to choose multi-period budgeting. Second, legislators who expect to
remain in office for a very short period will choose single-period budgeting. As their expected tenure
increases, they choose multi-period budgeting more often. Yet, as their expected tenure increases even
more, they may switch back to single-period budgeting.
Our theory of the budgetary process holds five implications. First, the total payment to each
interest group, across all agencies and programs, is the best unit to use to analyze budgeting. The
literature on incrementalism concentrates on agency budgets, with some reference to program budgets
and functional components of the budget. Our model, by contrast, rests on the notion that while these
classifications may be relevant to analyses of budgeting, they only remain so as long as they correspond
to, or provide an approximation for, how much specific groups that are beneficiaries of these agencies
and programs receive. We must view with some caution, therefore, studies of budgeting (including our
own) that are based on agency, program, or functional allocations. The model uses agency allocations
(or outlays) as an approximation for interest-group benefits. We believe, however, that the interest
groups that benefit from federal funding are the proper units of analysis to use when constructing models
to explain budgetary allocations, both over time and among different agencies and programs. We also
believe that interest groups’benefits are best reflected through agencies, and not through specific agency
programs or functional components.42
Second, the budget tends to expand in real terms over time. Legislators weigh the benefit of
expanding the budget against the political opposition from doing so. The increase in the deficit over the
42
Natchez and Bupp (1973) show that while the Atomic Energy Commission's overall budget remains stable over the period
studied, there are significant variations over specific Commission program budgets. Concentrating on these program
budgets might lead one to believe that Congress acted in a nonincremental fashion. But exploring benefits to recipient
groups might lead to a different conclusion, because shifts across programs might leave the members of these groups
largely unaffected by such changes.
30
real value of the deficit of the previous year affects legislator’s utility less than does the increase in the
budget. Hence the resulting deficit will tend to increase over time, at least in nominal terms. Both the
expansion of revenues, because of the growth of national income, and the expanding deficit, lead to an
expanding total budget. The model predicts that the budget will tend to expand in real terms over time.
It also predicts that it will expand more under multi-period, incremental, budgeting than under singleperiod budgeting.
Third, our earlier remarks in this section concerning legislative tenure and the choice of a budgetary
strategy hold implications for the imposition of term limits. Very short term limits will induce legislators
to act nonincrementally, with a resulting smaller budget. Increasing the term limit up to a point leads to
more occurrences of multi-period budgeting, a larger budget, and less legislator shirking. If the term
limit is very long, and if the legislator expects to remain in office during his natural life, more instances of
single-period budgeting eventually will occur. Because multi-period budgeting, other things constant,
leads to a larger federal budget, if one seeks a lower rate of expansion of the federal budget and if a
reduction in the federal deficit is considered a high priority, one could design a very short term limit to
induce legislators to choose more single-period budgeting. Similarly, the model predicts more rentseeking under multi-period budgeting. And, if one views reducing rent-seeking as a priority, then a very
short term limit might lead to less rent-seeking than will a term limit of intermediate length.
Fourth, incrementalism should not be without its defenders. If legislative outcomes arise out of
disequilibrium preference profiles, then the only way for the budgetary outcome to be an equilibrium
outcome is through issue-by-issue voting, or some other structurally induced equilibrium. Yet, if
legislators circumvent issue-by-issue voting through logrolling, there may exist no equilibrium
(McKelvey and Ordeshook 1984; Shepsle and Weingast 1984). Incrementalism would then have the
virtue of minimizing the resulting chaos by reducing variations in the budgetary outcomes from one year
31
to the next. If this were true, then it may be in the interest of stability of the whole budgetary system to
introduce moderate term limits, which lead to more occurrences of incrementalism, and hence reduce the
possible chaos resulting from disequilibrium.
Fifth, our theory points to and explains cases where incremental budgeting breaks down. The
hypothesis of legislative incrementalism predicts that legislators seldom reconsider the total budget and
only focus on small year-to-year additions: they treat the budget of the previous year as given. Scholars
since Lindblom suggest that legislators revert to incremental decision making to simplify an
unmanageably complex, extensive, and conflicted budgetary process. Our theory rejects the complexity
of the budgetary process as an explanation for the incremental outcomes observed. It suggests, instead,
that in periods of crises, when revenues are shrinking, for example, legislators will reconsider the total
budget. They will indeed revert to cutting program budgets, to be able to finance new programs and
remain within spending targets. Hence, the incremental outcome of the budgetary process is not the
result of an incremental decision process, where decision-makers do not examine the base. Instead,
when it occurs, incrementalism is the result of a rational, utility-maximizing process that leads to
marginal budgetary outcomes.43 That is, the result of the metadecision of “deciding how to decide” is
itself a rational choice.
A logical extension of our research is to test some of the model's untested predictions. These
include the effects of a legislator's winning margin, credibility, and expected term in office on his
budgetary decisions. This test would require identification of a bill where votes reflect the choice
between single- and multi-period budgeting. We tried using votes on indexation of Social Security
43
There have been very few additions to the literature on budgetary incrementalism in recent years. Perhaps this is because
of the absence of a theory. More likely, it reflects the conventional wisdom, since the Reagan years, that members of
Congress do adopt nonincremental changes.
32
benefits. But the data set could not allow statistical inferences, because of the almost unanimous vote for
indexation. It also would be fruitful to link this model to the existing literature on legislative shirking and
to derive a statistical test of the model's hypotheses about this phenomenon.
We also could apply the theory to several budget determination cases, to compare the model's
predictions in different political, constitutional, and economic environments. Even though we restrict the
analysis to budgets of nondefense agencies in the United States, it would be interesting to test the model
using both defense and nondefense agencies and to extend it to regulatory and taxation decisions. It also
would be interesting to analyze its applications to other countries. It may be useful to compare some of
the results among developed countries, on the one hand, and developing countries, on the other, to
identify the effects on public sector budgeting of constitutional variations and interest-group politics in
developing countries and to find where they differ from the results reached here. We also can test the
model using state budgets, to analyze any differences in governmental structures and interest-group
politics at the state and federal levels.
33
Bibliography
Banerjee, A., J. Dolado, J. Galbraith, and D. F. Hendry. 1993. Co-integration, Error
Correction, and the Econometric Analysis of Non-stationarity. New York: Oxford
University Press.
Bozeman, Barry. 1977. "The Effect of Economic and Partisan Change on Federal
Appropriations." Western Political Quarterly 30 (March), pp. 112-24.
Bureau of the Census. 1975. Historical Statistics of the United States: Colonial Times to 1970
(volumes 1 and 2). Washington D.C.: U.S. Department of Commerce.
Cameron, A. Colin. and Pravin K. Trivedi. 1990. Regression-Based Tests for Overdispersion in
the Poisson Model. Journal of Econometrics 46: 347-64.
Davis, Otto A., M.A.H. Dempster, and Aaron Wildavsky. 1966a "A Theory of the Budgetary
Process." American Political Science Review 60 (September), pp. 529-47.
Davis, Otto A., M.A.H. Dempster, and Aaron Wildavsky. 1966b. "On the Process of Budgeting:
An Empirical Study of Congressional Appropriation." in Gordon Tullock, ed., Papers on
Non-market Decision Making. Charlottesville, Va.: Thomas Jefferson Center for Political
Economy, pp. 63-132.
Davis, Otto A., M.A.H. Dempster, and Aaron Wildavsky. 1971. "On the Process of Budgeting
II: an Empirical Study of Congressional Appropriation." In R. F. Byrne, et al., eds., Studies
in Budgeting. Amsterdam: North Holland Publishing Co., pp. 292-373.
Davis, Otto A., M.A.H. Dempster, and Aaron Wildavsky. 1974. "Towards a Predictive Theory
of Government Expenditure: U.S. Domestic Appropriations." British Journal of Political
Science 4 (October), pp. 419-52.
Gist, John R. 1974. "Mandatory Expenditures and the Defense Sector: Theory of Budgetary
Incrementalism." In Randall B. Ripley, ed., A Sage Professional Paper pp. 5-39.
Gist, John R. 1982. “’Stability’and ‘Competition’in Budgetary Theory.” American Political
Science Review 76 (December), pp. 859-72.
Granger, C.W.J., and P. Newbold. 1974. "Spurious Regressions in Econometrics." Journal of
Econometrics 2, pp. 111-20.
Gurmu, Shiferwa and Pravin K. Trivedi. 1994. Recent Developments in Models of Events
Counts: A Survey. Discussion Paper #261, Indiana University.
Hausman, Jerry, Brownwyn H. Hall, and Zvi Griliches. 1984. "Econometric Models for Count
Data with an Application to Patents-R & D Relationship." Econometrica 52 (4), pp. 909938.
34
Heiner, Ronald A. 1983. "The Origins of Predictable Behavior." American Economic Review
73, pp. 227-61.
Jackson, John E. 1972. "Politics and the Budgetary Process." Social Science Research 1
(April), pp. 35-60.
Kalt, Joseph P., and Mark A. Zupan. 1984. "Capture and Ideology in the Economic Theory of
Politics." American Economic Review 74 (June), pp. 279-300.
Kamlet, Mark S. and David C. Mowery. 1985. “The First Decade of the Congressional Budget
Act: Legislative Imitation and Adaptation in Budgeting.” Policy Sciences 18 (December),
pp. 313-34.
Kamlet, Mark S., David C. Mowery, and Tsai-Tsu Su. 1988. "Upsetting National Priorities? The
Reagan Administration's Budgetary Strategy." American Political Science Review 82
(December), pp. 1293-307.
Kanter, Arnold. 1972. "Congress and Defense Budget: 1960-1970." American Political
Science Review 66 (March), pp. 129-43.
Kau, James B. and Paul H. Rubin. 1979. "Self-interest, Ideology and Logrolling in
Congressional Voting." Journal of Law and Economics 22 (October), pp. 365-84.
Kau, James B., and Paul H. Rubin. 1993. “Ideology, Voting, and Shirking.” Public Choice 76,
pp.151-172.
Landes, William M., and Richard A. Posner. 1975. "The Independent Judiciary in an Interestgroup Perspective." Journal of Law and Economics 18 (December), pp. 875-901.
Lane, Jan-Eric, Anders Westlund, and Hans Stenlund. 1981. "Analysis of Structural Variability
in Budget-Making." Scandinavian Political Studies 4 (2), pp. 127-49.
LeLoup, Lance T. 1978. "The Myth of Incrementalism: Analytical Choices in Budgetary
Theory." Polity 10 (Summer), pp. 488- 509.
Lindblom, Charles E. 1959 "The Science of ‘Muddling Through’." Public Administration
Review 19 (Spring), pp. 79-88.
McKelvey, R. D. and P. C. Ordeshook. 1984. “An Experimental Study of the Effects of
Procedural Rules on Committee Behavior.” Journal of Politics 46 (1) pp. 182-205.
Moreland, William B. 1975. "A Nonincremental Perspective on Budgetary Policy Actions." In
Randall B. Ripley and Grace A. Franklin, eds., Policy-making in the Federal Executive
Branch. New York: The Free Press, pp. 45-60.
Natchez, Peter B., and Irvin C. Bupp. 1973. "Policy and Priority in the Budgetary Process."
American Political Science Review 67 (September), pp. 951-63.
Nelson, Douglas, and Eugene Silberberg. 1987. "Ideology and Legislator Shirking." Economic
35
Inquiry 25 (January), pp. 15-25.
Olson, Mancur. 1977. The Logic of Collective Action: Public Goods and the Theory of Groups.
New York: Schocken Books.
Pesaran, M. Hashem. 1974. "On the General Problem of Model Selection." Review of Economic
Studies 41 (1), pp. 153-171.
Ripley, Randall B., and Grace A. Franklin. 1975. "The Empirical Analysis of Federal Policy
Making." In Randall B. Ripley and Grace A. Franklin, eds., Policy-making in the Federal
Executive Branch. New York: The Free Press, pp. 171-85.
Ripley, Randall B., William M. Holmes, Grace A. Franklin, and William B. Moreland. 1975.
"Explaining Changes in Budgetary Policy Actions." In Randall B. Ripley and Grace A.
Franklin, eds., Policy-making in the Federal Executive Branch. New York: The Free Press,
pp. 145-70.
Shepsle, Kenneth A. and Barry R. Weingast. 1981. “Structure-Induced Equilibrium and
Legislative Choice.” Public Choice 37 (3) pp. 503-19.
Shepsle, Kenneth A. and Barry R. Weingast. 1984. “When Do Rules of Procedure Matter?”
Journal of Politics 46 (1) pp. 183-221.
Shepsle, Kenneth A., and Barry R. Weingst. 1987. “The Institutional Foundation of Committee
Power.” American Political Science Review 81 (March), pp. 85-127.
Simon, Herbert A. 1955. "A Behavioral Model of Rational Choice." Quarterly Journal of
Economics 60, pp. 99-118.
Simon, Herbert A. 1959. "Theories of Decision Making in Economics and Behavioral Science."
American Economic Review 49, pp. 253-283.
Tohamy, Soumaya M., Hashem Dezhbakhsh, and Peter H. Aranson. 1999. " A New Approach
for Testing Budgetary Incrementalism." Manuscript, Department of Economics, Emory
University, Atlanta, Georgia.
United States Office of Management and Budget. 1946-1995. The Budget of the United States
Government: Fiscal Years 1947 - 1996. Washington D.C.: Executive Office of the
President, Office of Management and Budget.
Wagner, Adolph. 1877. Finanzwissenschaft, Part I. Leipzig: C. F. Winter.
Wagner, Adolph. 1890. Finanzwissenschaft, Part II. Leipzig: C. F. Winter.
Weingast, Barry R. and William Marshall. 1988. “The Industrial Organization of Congress.”
Journal of Political Economy 96(1), pp. 132-63.
36
Table 1: The Effect of Different Variables on the Choice of Strategy in the Basic Model
Variable
Total Revenues, Rt− 1
Total Budget, Bt− 1
Discount Rate, r
Inflation Rate, i
Relative Valuation of Income and
Political Opposition, α β
Real Rate of Revenue Growth, g
Effect on the Choice
of MPB
+
−
−
+
+
+
Effect on the Choice
of SPB
−
+
+
−
−
−
Notes: MPB and SPB refer to multi-period and single-period budgeting respectively. A “+” in the second
column signifies that an increase in the variable in the first column, other things equal, makes multiperiod budgeting more preferred relative to single-period budgeting. The effect of all the variables
other than g is determined through the g * expression. The effect of g is determined through the
computation of a critical discount rate, r * . Because the analysis of r * is exactly the same as that
of g * , we do not repeat it here.
37
Table 2: Regression Results for equation (3) – Incremental Variables
Explanatory Variables
ECONOMIC VARIABLES:
INFLATION RATE
DI1t
Dependent Variable
DI2t
DI3t
DI4t
-11.056***
(0.872)
-11.118***
(0.957)
-9.796***
(0.917)
-8.403***
(0.921)
DISCOUNT RATE
30.610***
(0.990)
28.781***
(1.090)
29.426***
(1.039)
24.684***
(1.234)
REVENUE GROWTH
0.784**
(0.370)
0.639
(0.398)
0.790**
(0.393)
0.3397
(0.445)
DEFICIT
-0.000005***
(0.000002)
-0.000004**
(0.000002)
-0.000006**
(0.000002)
-0.000005**
(0.000002)
PERSISTENT-DEFICIT
0.00001***
(0.000002)
0.00001***
(0.000002)
0.000012***
(0.000002)
0.000012***
(0.000002)
1.192***
(0.0157)
1.172***
(0.0169)
1.146***
(0.0167)
1.123***
(0.0193)
-1619.30
47
-1345.90
47
-1383.07
47
-1027.42
47
POLITICAL VARIABLE
DEMOCRATS
Log-likelihood
Number of observations
Notes:
Standard errors are reported in parentheses.
‘***’ and ‘**’ indicate significance at the one and five percent levels, respectively.
DI1t measures the number of incremental cases using a small positive band.
DI2t measures the number of incremental cases using a small two-sided band.
DI3t measures the number of incremental cases using a large positive band.
DI4t measures the number of incremental cases using a large two-sided band.
38
Table 3: Regression Results for equation (3) – Regularity Variables
Explanatory Variables
ECONOMIC VARIABLES:
INFLATION RATE
DR 1t
Dependent Variable
DR 2t
DR 3t
DR 4t
-10.677***
(0.793)
-10.184***
(0.828)
-10.260***
(0.816)
-8.853***
(0.931)
DISCOUNT RATE
31.051***
(0.916)
30.456***
(0.956)
30.777***
(0.938)
27.560***
(1.080)
REVENUE GROWTH
0.726**
(0.338)
0.315
(0.361)
0.641*
(0.352)
-0.031
(0.405)
DEFICIT
-0.000006***
(0.000001)
-0.000007***
(0.000002)
-0.000006***
(0.000002)
-0.000008***
(0.000002)
0.000013***
(0.000002)
0.000014***
(0.000002)
0.000014***
(0.000002)
0.000015***
(0.000002)
1.249***
(0.0144)
1.225***
(0.015)
1.223***
(0.0149)
1.178***
(0.0169)
-2048.88
47
-1845.10
47
-1884.76
47
-1337.93
47
PERSISTENT-DEFICIT
POLITICAL VARIABLE:
DEMOCRATS
Log-likelihood
Number of observations
Notes:
Standard errors are reported in parentheses.
***’ and ‘**’ indicate significance at the one and five percent levels, respectively.
DR1t measures the number of incremental cases using a small positive band.
DR2t measures the number of incremental cases using a small two-sided band.
DR3t measures the number of incremental cases using a large positive band.
DR4t measures the number of incremental cases using a large two-sided band.
39
Table 4: Regression Results for equation (4) – Incremental Variables
Explanatory Variables
ECONOMIC VARIABLES:
INFLATION RATE
DI1t
Dependent Variable
DI2t
DI3t
DI4t
-5.828***
(0.870)
-6.745***
(0.956)
-4.775***
(0.907)
-4.916***
(1.048)
DISCOUNT RATE
21.854***
(1.020)
21.550***
(1.118
20.752***
(1.066)
18.698***
(1.240)
REVENUE GROWTH
0.289
(0.356)
0.081
(0.387)
0.254
(0.378)
-0.245
(0.436)
DEFICIT
-0.000016***
(0.000002)
-0.000015***
(0.000002)
-0.000016***
(0.000002)
-0.000014***
(0.000002)
0.000030***
(0.000002)
0.000028***
(0.000002)
0.000029***
(0.000002)
0.000026***
(0.000003)
1.186***
(0.014)
1.156***
(0.0155)
1.143***
(0.0149)
1.104***
(0.0178)
PRESIDENT-SWITCH
0.251***
(0.061)
0.225***
(0.067)
0.275***
(0.064)
0.243***
(0.075)
CONGRESS-SWITCH
1.201***
(0.062)
1.035***
(0.070)
1.172***
(0.659)
0.879***
(0.081)
ELECTION YEAR
0.608***
(0.0479)
0.609***
(0.0513)
0.592***
(0.0511)
0.576***
(0.0578)
-1408.41
47
-1203.37
47
-1202.29
47
-938.93
47
PERSISTENT-DEFICIT
POLITICAL VARIABLES:
DEMOCRATS
Log-likelihood
Number of observations
Notes:
Standard errors are reported in parentheses.
***’ indicates significance at the one percent level.
DI1t measures the number of incremental cases using a small positive band.
DI2t measures the number of incremental cases using a small two-sided band.
DI3t measures the number of incremental cases using a large positive band.
DI4t measures the number of incremental cases using a large two-sided band.
40
Table 5: Regression Results for equation (4) – Regularity Variables
Explanatory Variables
ECONOMIC VARIABLES:
INFLATION RATE
DR 1t
Dependent Variable
DR 2t
DR 3t
DR 4t
-4.843***
(0.781)
-4.691***
(0.812)
-4.542***
(0.799)
-4.151***
(0.897)
DISCOUNT RATE
21.579***
(0.925)
21.47***
(0.962)
21.394***
(0.946)
19.722***
(1.065)
REVENUE GROWTH
0.507
(0.320)
0.176
(0.342)
0.453
(0.333)
-0.203
(0.386)
DEFICIT
-0.000017***
(0.000002)
-0.000017***
(0.000002)
-0.000017***
(0.000002)
-0.000017***
(0.000002)
PERSISTENT-DEFICIT
0.000032***
(0.000002)
0.000031***
(0.000002)
0.000031***
(0.000002)
0.000030***
(0.000002)
1.244***
(0.0127)
1.217***
(0.0133)
1.218***
(0.0131)
1.163***
(0.0151)
PRESIDENT-SWITCH
0.196***
(0.0562)
0.189***(0.059) 0.183***
(0.058)
0.206***
(0.066)
CONGRESS-SWITCH
1.356***
(0.055)
1300***
(0.058)
1.351***
(0.057)
1.181***
(0.066)
ELECTION YEAR
0.581***
(0.0442)
0.561***
(0.0462)
0.573***
(0.0456)
0.560***
(0.0516)
-1750.08
47
-1593.13
47
-1607.00
47
-1166.23
47
POLITICAL VARIABLES:
DEMOCRATS
Log-likelihood
Number of observations
Notes:
Standard errors are reported in parentheses.
***’ indicates significance at the one percent level.
DR1t measures the number of incremental cases using a small positive band.
DR2t measures the number of incremental cases using a small two-sided band.
DR3t measures the number of incremental cases using a large positive band.
DR4t measures the number of incremental cases using a large two-sided band.
41