The origins of incrementalism and its critics

The study of policy change has taken on many forms and proceeded along many paths over the decades. One approach to the study of this broad question has focussed on the nature of the systems that create policy and how these systems constrain the speed and breadth of policy change.

One strain of this research tradition is of particular interest to the study of budgets. Wildavsky (Wildavsky Reference Wildavsky1964; Wildavsky and Caiden Reference Wildavsky and Caiden1988) drew from the prior tradition on incremental policy change – most often associated with Lindblom (Reference Lindblom1959) – to argue that budgetary processes (and, in particular, the federal budgetary policy process) are incremental. Budgets represent compromises between diverse actors and any proposed large change threatens the existing coalitions. As a result, budgetary changes tend to be small and incremental.

There was an almost immediate reaction to the strong form of Wildavsky’s incrementalist characterisation of the budgetary process. Schulman (Reference Schulman1975) argued that there were important counter-examples to the notion that incrementalism was a covering law for policy change processes. He argued that certain policy domains could not proceed incrementally because they could not be subdivided into smaller parts. His signal example is space policy and NASA. One cannot build slowly towards a space programme, Schulman argued. One has to make large (in an absolute sense) and nonincremental (in a sense of changes from previous years) change to proceed in these domains of policy.

The resulting debate between proponents of incrementalism as a generally applicable covering law and critics who pointed to counter-examples continued for decades. This debate was further complicated by drift and confusion within the meaning of various terms in their debate including how one should define and measure incrementalism (Wanat Reference Wanat1974; Tucker Reference Tucker1982; Berry Reference Berry1990; John and Bevan Reference John and Bevan2012). Scholars used incrementalism as a descriptive, explanatory, normative and a predictive theory. Each of those theoretical lenses has yielded various ways for incrementalism to be measured.

Punctuated equilibrium approaches

Cross-application of a theory from palaeontology overcame the impasse between incrementalism and its critics. Baumgartner and Jones (Reference Baumgartner and Jones2009) argued that policy changed proceeded through two separate patterns. Most of the time, policy change was slow – as the incrementalists had argued. However, conditions could create a period wherein large change (punctuated change) is possible through the operation of positive attention cycles and dramatic external events – consistent with many of the critics of incrementalism. PET bridged the divide between theories of fast and slow policy change by acknowledging both dynamics exist in the policy process.

The resulting theory of punctuated policy change led to specific predictions about the distribution of policy changes. Particularly, scholars expected the distribution to have a leptokurtic shape – a tall centre and heavy tails of the distribution. Some of the earliest tests of the predictions of the punctuated equilibrium model used budgetary changes as the indicator of policy change (Jones et al. Reference Jones, True and Baumgartner1997; Jones et al. Reference Jones, Baumgartner and True1998). These investigations found support for the punctuated equilibrium model of policy change in the observed distribution of budgetary changes in federal budgetary categories.

The study of the distributions of policy-relevant outputs spreads rapidly to accommodate tests in various stages of the United States (US) federal policy-making process (Jones et al. Reference Jones, Sulkin and Larsen2003), US state-level budgetary data (Breunig and Koski Reference Breunig and Koski2006) and US school district-level data (Robinson Reference Robinson2004). Tests also spread to other political system to find similar patterns of policy change in a wide range of nations (Breunig Reference Breunig2006; Baumgartner et al. Reference Baumgartner, Breunig, Green-Pedersen, Jones, Mortensen, Nuyte-mans and Walgrave2009). The result of these various studies was a persuasive case that the mixture of incremental and punctuated policy changes characterized all policy systems studied (Jones et al. Reference Jones, Baumgartner, Breunig, Wlezien, Soroka, Foucault, François, Green-Pedersen, Koski, John, Mortensen, Varone and Walgrave2009). This empirical phenomenon was also characterized as the Dynamic Model of Choice for Public Policy (Jones and Baumgartner Reference Jones and Baumgartner2005; Jensen et al. Reference Jensen, Mortensen and Serritzlew2016).

Hypothesis testing

The distributional approach to punctuated policy change has revealed a great deal about patterns common across institutional environments. However, the approach is limited in the sorts of hypotheses it can test – in part because comparisons across data distributions call for several large databases. A separate methodological tradition emerged that tested hypotheses about how characteristics of the policy systems could lead to different expectations of change. Notably, this approach sought research designs more similar to traditional regression-based hypothesis testing.

Jordan’s (Reference Jordan2003) work on local government expenditures represents an early example of this tradition. Jordan defined a threshold beyond which a change could be categorized as nonincremental (a “punctuation”). She then compared different types of local government expenditures (police, highways, public buildings, etc.) to identify which most closely fit the expectations of PET.

Robinson adopted a similar approach to studying local government expenditures because of the leverage they provide on questions of institutional determinants of policy change. The earliest work adapted the distributional methodology to direct hypothesis testing (Robinson Reference Robinson2004) comparing school districts with high and low values of bureaucratisation. The later work adopted a more direct categorisation of each budgetary outcome as incremental, moderate or large changes (Robinson et al. Reference Robinson, Caver, Meier and O’Toole2007). This hypothesis testing approach allowed for multivariate testing within a regression context and compared the influence of bureaucratisation and organisation size on the propensity of an organisation to experience punctuated budgetary change – while allowing for some control variables as well. The approaches of Jordan and Robinson opened up a methodological space for testing hypotheses for factors related to increased or decreased rates of nonincremental policy change.

Recent innovations

Recent work has further refined and extended the hypothesis testing approach to studying punctuated or nonincremental budgetary change.

Some work has built directly on the distributional analysis methodology to construct comparisons more amenable to formal hypotheses testing – notably through the use of l-kurtosis measures. L-kurtosis provides a normed characterisation of a distribution that facilitates comparison across samples – a necessity for comparison of samples of budgetary change. The resulting methodology has undergirded studies of American state budgets (Breunig and Koski Reference Breunig and Koski2006, Reference Breunig and Koski2012) and cross-national comparisons (Breunig Reference Breunig2006; Baumgartner et al. Reference Baumgartner, Breunig, Green-Pedersen, Jones, Mortensen, Nuyte-mans and Walgrave2009; Jones et al. Reference Jones, Baumgartner, Breunig, Wlezien, Soroka, Foucault, François, Green-Pedersen, Koski, John, Mortensen, Varone and Walgrave2009).

Others have further developed our understanding of the institutional characteristics that facilitate or impede nonincremental budgetary changes. Ryu (Reference Ryu2011a, Reference Ryu2011b, Reference Ryu2009) used US state sub-function expenditure patterns to assess the role of institutional friction and legislative professionalism on budgetary change. Epp and Baumgartner (Reference Epp and Baumgartner2017) examined how institutional complexity and capacity influence budgetary punctuations. In the analysis of US budget authorities from 1947 to 2012, the authors find that complexity leads to more punctuations and capacity leads to more incremental budgetary changes Epp and Baumgartner (Reference Epp and Baumgartner2017).

A limitation of comparing distributions is that the time-series elements of the data are removed. With regards to punctuations, the atemporal approach left questions as to whether the large budgetary changes occurred multiple years in a row or randomly distributed over time. Robinson et al. (Reference Robinson, Flink and King2014) examined this question by proposing two theoretical models of budgetary punctuations – the institutional and error accumulation models. Empirical analyses showed that budgetary punctuations occur in groupings. In other words, the probability of a budgetary punctuation is positively related to having had a recent punctuation.

In her work Flink (Reference Flink2017), incorporated theories from public administration literature to predicting policy change. Endogenous organisation change (personnel instability) and policy feedback (organisation performance) are examined as predictors of categories of budgetary changes. Findings show that high organisation performance and low personnel turnover lead to more incremental budgetary changes. Furthermore, Flink (Reference Flink2017) demonstrates that magnitudes of budgetary changes (incremental, medium, punctuated) should be analyzed as positive and negative categories, something scholars have not done with consistency throughout the literature. The empirical results in Flink (Reference Flink2017) reveal that medium size budgetary changes, when split between positive and negative changes, give competing expectations in the results.

Budgetary punctuations as corrective reactions

For this study, we focus on budgets as indicators of policy attention and budgetary change as measures of policy change – an approach common within the PET literature (Breunig and Koski Reference Breunig and Koski2006; Ryu Reference Ryu2011a; Robinson et al. Reference Robinson, Flink and King2014). Public organisations may experience budgetary punctuations that work as corrective processes – a drastic change followed by another rapid change in the opposite direction. The budget keeps a general equilibrium point. We describe this pattern as a corrective reaction of budgetary punctuations. It assumes that managers are able to take control of budgetary allocations to make major alterations each fiscal year. The back and forth pattern of punctuations could be evident in public organisations for a number of reasons: major fluctuations in overall resources, a reprioritisation of policy objectives, environmental shocks, performance goals and general mismanagement of financial resources may contribute to a corrective reaction in the budget.

This model yields two hypotheses:

Figures 1 and 2 illustrate both of the hypotheses. Each figure shows the budget amount, from x to x+n, for a programme over the course of five fiscal years (t to t+4). Figure 1 has a budget series consistent with the expectations of the positive correction from Hypothesis 1. From year to year, funding levels experience incremental changes. However, in Figure 1, there is a drastic budget decrease from t+1 to t+2. That is followed by a positive budgetary punctuation in the next fiscal year to restore funding near original levels.

Figure 1 Positive correction.

Figure 2 Negative correction.

Figure 2 displays the other prediction of the corrective model, a negative correction. Over the course of the five fiscal years in the figure, the funding level at the start (t) is about the same amount as at the end (t+4) for the programme. In the middle of this budget series is a cluster of punctuations. A positive punctuation occurs at t+2. The financial growth is short-lived, though. A negative budgetary punctuation occurs in the next fiscal year.

In both Figures 1 and 2, the major financial alterations are not sustained over multiple fiscal years. Figure 1 has a one-time funding cut, while Figure 2 has a single influx of money. The corrective model theorizes that budgetary punctuations are temporary shocks to policy systems – budgetary punctuations are not step changes to a new budgetary base level of funding. We theorize it maintains a normal or equilibrium funding level over time. For illustrative purposes, we assume that the corrective punctuations are of the same magnitude as the initial punctuation, though this model can have punctuations that are not fully (or even over) corrective.

We observe these patterns in our dataset of school districts by analyzing our budgetary variable of interest, instructional expenditures per student. For example, a positive correction was observed when a district experienced a drop in instructional expenditures coupled with a rise in enrolment. This drastically reduced the instructional expenditures per student. The next year instructional expenditures were returned to a higher level and total student enrolment had a slight decrease. This was a corrective punctuation back to near the original funding level before the cut.

Budgetary punctuations as trends

The other model we propose is the trending model. This model predicts multi-year punctuations that repeat in either the positive or negative direction. The grouping of punctuations yields a budget that is trending upwards or downwards. This pattern of budgetary punctuations could result from a major change in overall financial resources in a period of several years – economic prosperity can lead to climbing budgets, while hard financial times can shrink funds. Trending budgets could also suggest a reprioritisation of a programme or policy area. Sustained dedication to a programme could yield multiple positive punctuations over several years. Negative budgetary punctuations signal a sustained shift away from a policy, mission or goal of the organisation. New leadership in the organisation could lead to different policy priorities, manifesting as drastic shifts in a programme’s monetary resources – but shifts spread out over several years of implementation. In any of these cases, the financial pattern over time would show a separate clustering of positive and negative punctuations.

The hypotheses for the trending model are:

Figures 3 and 4 illustrate each of the hypotheses, similarly to the figures from the corrective model. Figure 3 demonstrates Hypothesis 3, a positive trend. This figure shows how positive budgetary punctuations group together. The budget keeps rising in drastic increments, resulting in a more highly funded programme. On the other hand, Figure 4 displays negative budgetary punctuations over multiple fiscal years. This is referred to as a negative trend in Hypothesis 4. Financial resources for this programme take a massive cut from t to t+4. Overall, these figures show positive budgetary punctuations clustering separately from negative budgetary punctuations.

Figure 3 Positive trend.

Figure 4 Negative trend.

Unlike the corrective model that portrays budgetary punctuations as shocks to the equilibrium of a programme’s budget, the trending model reveals budgetary punctuations as a path to a new base level of funding. In Figures 1 and 2, the budget returns to a similar level as it began in the series. In Figures 3 and 4, the programme budget adjusts to a new high or low level of funding. For example, in our sample of school districts, a positive trend occurred in the instructional expenditures per student variable for a district that experienced slight declines in enrolment in addition to increases in instructional expenditures. This resulted in much higher funding for instructional expenditures per student in a short timespan of only 2 years.

Distribution of budgetary changes

How do budgets respond after a punctuation? The empirical models in the previous section show that punctuations lead to more punctuations. The comparison of the magnitude of the punctuations reveals more about the dynamics of sequential punctuations.

The corrective model suggests big moves in one direction (positive or negative) leads to more changes (of any size) in the other direction. The distributions are expected to illustrate this by having a number of punctuations in the opposite direction of the previous punctuation. Furthermore, the distribution should be shifted towards the opposite direction of the punctuation – more observations should be observed in the medium size changes as well. A tall central peak centred at zero in the distribution means that the punctuation level was kept – the next fiscal period did not work to correct the punctuation.

Table 3 reports a three-by-three table of the categories of punctuations (none, positive, negative) for year t−1 and year t. By the numbers, it is observed that positive punctuations in t−1 yield negative punctuations 14.5% of the time compared to another positive punctuation 6.8% of the time. The most common occurrence though is a single positive punctuation at 78.7%. For negative punctuations in t−1, positive punctuations occur 49.6% of the time in response, while negative punctuations occur 3.4%. Negative punctuations are followed by no punctuation at 47.1% of the time. This illustrates that even among the raw numbers of the categorical variables, the corrective model is supported in positive versus negative punctuations.

Table 3 Categories of punctuations in time t−1 and time t

Table 4 and Figure 6 examine the distribution of budgetary changes and descriptive statistics 1 year and 2 years after a positive and negative punctuations. The patterns provide context for the statistical model. In the year following a punctuation, the mean values of budgetary changes are altered from the mean of all budgetary changes, around 4%. One year after a positive punctuation, the mean shifts down to negative 4% – a modest change of 8%. In a more drastic shift, 1 year after a negative punctuation, the mean budgetary change is 44%. Considering all of Figure 6, second punctuations are more likely than the initial ones, but they are still relatively rare. Instead, a punctuation leads to a shift in the mean and a widened range of budgetary changes – largely mirroring the overall distribution of budgetary changes.

Figure 6 Annual per cent budgetary change after punctuations.

Table 4 Annual per cent budgetary change after punctuations

Another interesting result in Figure 6 is that the distributions after positive punctuations seem to cluster around zero and in the shoulders of the distribution more than the negative punctuation distributions. This illustrates that a greater portion of the positive budgetary changes are being retained for some time. The negative budgetary punctuations lead to distributions that are more of uniform shape than a normal distribution.

Lastly, it is interesting that most corrective punctuations are occurring in the year after the punctuation, as opposed to 2 years after the punctuation. This is evident by Figure 6 – there are many more observations of punctuations in the top row than the bottom row. This tells us that corrections are immediate.

Corrective punctuations

Given the similarity of the postpunctuation distribution of budgetary changes, it is important to assess the degree to which punctuations are fully “corrected” by the compensating changes. Are the second punctuations of similar sizes to the original punctuations? This could be an indication that the punctuations are temporary exceptions in an equilibrium seeking system. In a strong version, an equilibrium seeking system would react to a punctuation with an equal and opposite budgetary change – resulting in no net change in the budget. This is the model illustrated in Figures 1 and 2. The data allow us to directly assess the extent to which punctuations are “corrected” to their original budget by the corrective punctuation pattern.

Table 4 and Figure 7 present the results for this assessment. It is important to keep in mind that out of the hundreds of punctuations, only 59 and 58 follow the alternating pattern. While this represents more secondary punctuations than one would expect if there were no relationship between punctuation propensity, these are still rare events – even rare in the context of punctuated change. We are discussing 117 instances of paired punctuations out of the 600 total punctuations.

Figure 7 New budget’s per cent of original budget.

Figure 7 shows the percentage of budgetary change from the original budget level (before punctuations) to the level after two punctuations (initial and corrective, positive and negative). A score at 0% means that the budget was returned to its original level – the punctuations were perfectly corrective. A score in the negatives means that the new budget level is less than the amount before punctuations. A positive budgetary change means that the punctuations resulted in a higher budget than before the punctuations (Table 5).

Table 5 New budget’s per cent of original budget

The information on Table 4 suggests that second punctuations are, on average, fully corrective. The net result of a negative-then-positive pair is a reduction of a mere 0.39% in the budget. This is effectively a full correction. However, the mean conceals a great deal of variation around that mean. The distribution (illustrated on the top half of Figure 7) is almost uniformly distributed around the mean with a range from –80% to 73% and a SD of 28. While the average pair is corrective, the average value is not as likely an observation in these data with a more uniform shape, as compared to a normal distribution.

The same pattern appears in positive-then-negative pairs. The net change of these pairs is, on average, a reduction of 1.66%. This is not as close to zero as the other alternating pair, but still a small net change. Again the distribution has a large standard deviation (27) with a remarkable range around this mean – with the negative side of the distribution weighed down by a single observation at a complete net reduction of the budget (–100%). In both cases, the compensatory nature of paired punctuations is accurate on average but conceals a great deal of variation in these budgetary systems.