Budgeting in Public Organizations: The Influence of Managerial and Political Aspects

Abstract

We study the use of budgets in public organizations with governance shaped by democratic structures. In particular, we examine political variables – electoral cycle, majority government, and ideology – and managerial variables – past performance and peer performance – in the context of city governments. Using data from 170 municipalities over a period of seven years (two electoral cycles), the study documents the simultaneous effect of these two types of variables. We find evidence that budget increases are more pronounced right before and right after elections (political). Furthermore, our findings suggest that ideology affects not only resource allocation decisions but also how municipalities manage the budgeting process; indicating the potential role of ideology in management practices (political). We also find evidence for ratcheting (managerial): the deviation between last period’s actual and budgeted performance is associated with upcoming budgets. However, this effect is symmetric, in contrast to the more common asymmetric structure previously documented in for-profit organizations. Relative target setting (managerial) is also present in public organizations and varies with the relative debt level of the city. The findings of this study are also informative to for-profit organizations as they open up their governance to more democratic structures and begin to experience the competing managerial and political logics in interacting with stakeholders.

Keywords:

• Public Management
• Informativeness
• Ratcheting
• Relative Target Setting

JEL codes:

• M41
• M48

Acknowledgements

We thank Matthias D. Mahlendorf and two anonymous reviewers for their valuable comments and guidance.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 For instance, program spending decisions across public departments has been found to be associated with individual departmental performance (Ho, 2011), suggesting that past performance affects budgeting decisions beyond political constraints.

2 The Leviathan model is similar to the ‘empire building’ argument in agency theory (Jensen, 1986), where top management deviates organizational resources for private enjoyment (the corporate jet being the typical example). While the ‘empire building’ argument is mainly an economic argument where resources are misused, the Leviathan model is about power whether is economic (misallocation of resources for personal use), social (reelection), and organizational (power within the party and the public bureaucracy). The concept of empire building is associated with a limitation of governance structures and agency costs in for profit organizations; and it is unrelated to the emergence of parties other than shareholders participating in the governance of companies.

3 This is similar to for-profit organizations where personal and organizational objectives interact in the empirical proxies used to measure them. While each proxy reflects various objectives, they weight these objectives in different proportions.

4 Brülhart and Jametti (2019) describe how municipalities with greater scope for direct-democratic participation in the tax setting process (instead of delegated fiscal authority) are more likely to correspond to the policy preferred by the median voter, which reduces budget spending. Craw (2008) argues that, in the context of delegated fiscal authority, institutions that promote greater accountability of policy makers to voters may (ironically) create opportunities for rent seeking and other distortions that result in an upward bias in government spending.

5 We interpret ratcheting as impounding past information into upcoming budgets, with no implicit interpretation on whether performing above budget indicates easier or harder future budgets. When ratcheting happens in the context of revenues and profits, performing (below) above budget is associated with (easier) harder budgets for the upcoming period. In the context of our study – management of expenses by city halls, the interpretation of ratcheting is opposite in the sense that a spending (below) above budget is associated with (harder) easier budgets for the upcoming period.

6 This city council-mayor relationship can be also interpreted as a principal-agent relationship, mayor’s control of the city council mutes the control role that principals play in this relationships. We thank one of the reviewers for suggesting the agency theory perspective to this relationship.

7 An implicit assumption in the for-profit literature is that governance structures are designed well enough for managers to use information with the objective of using resources as effectively and efficiently as possible. If Median Voter Preference mayors’ objectives are also to use resources in the same way, then we expect a similar behavior. We thank one of the reviewers for this important point.

8 Prior work (Brülhart & Jametti, 2019) has reported the comparability of municipalities’ behavior. These authors define benevolent (non-Leviathan) municipalities as those municipalities that set tax rates by direct-democratic participation of citizens, which constrains politicians to behave ‘benevolently’ (Craw, 2008). Leviathan municipalities are those with delegated fiscal authority, i.e. those that elect representatives who then, set tax rates. These authors find a relationship between budget size and cities’ competition for non-Leviathan municipalities, but non-significant for Leviathan ones. In our setting, all the municipalities work with delegated fiscal authority. However, since a sign of Leviathan is the tendency to over-spend, higher debt might signal more of a Leviathan behavior.

9 Then number of council members is 21 for cities between 20,000 and 50,000 people. For cities between 50,000 and 100,000 is 25. Above 100,000, one more member every 100,000 people or fraction; if the number is even, then an additional member is added to always have an odd number.

10 The transfers from the central government are determined following a specific formula, which gives approximately 75% weight to population, and the remaining 25% is allocated based on fiscal effort. Central government has some leeway to accommodate for the differences in cities’ size. Those from the regional government are allotted based on the city’s contribution to regional taxes and are of much less importance. As of 2009, 60% of transfers were form their central government, 25% from regional governments and the rest from other sources. Regional governments’ transfers’ main role is to level off the play field for small towns with little tax collection capacity. In our sample the correlation coefficient between city population and transfers from regional and national governments is 0.94.

11 For instance, the southern region of Andalusia, with an area that represents a 17.25% of Spain, has 778 municipalities; 25 with a population larger than 40,000; 123 with a population between 40,000 and 10,000; and 626 municipalities with less than 10,000, of which 20 have less than 1000 citizens. On the opposite extreme, La Rioja, in the Northeast, represents only 1% of total area of the country and has a total of 174 municipalities, only one with more than 40,000 inhabitants and three additional ones with a population over 10,000. The total number of municipalities in Spain is 8124, but 4998 have less than 1000 inhabitants, and only 172 have more than 40,000 (Source: The Spanish Statistical Office (INE)). Finally, the GDP of Spain as of 2016 was €1,118,522 million (current prices) and its share in the total EU GDP was of 7.5 % (Source: Eurostat).

12 The size of the third larger city (Valencia is only 787,807) and the fourth and fifth (Sevilla and Zaragoza) are 689,434 and 664,938 respectively. Source: Statistical Spanish Office (2016).

13 An interesting set of control variables are non-financial performance variables (Liguori et al., 2012) and examine how they affect target setting. We examined existing databases and talked with city managers, a consistent set of non-financial measures for public services does not exist. Moreover, such a research question would have most power at the department level rather than the overall city level. At this latter unit of analysis, the influence of service performance is relevant if over (under) performance of different services is correlated and, if this correlation exists and has a significant degree of autocorrelation, fixed effect variables capture part of their effect.

14 The dummy takes a value of one for budgets approved in years right before an election year. Election years are 2011, 2015 and 2019. That is the dummy takes one for years 2010, 2014, and 2018. Elections take place always in May and, therefore, politicians have little room for using budgets of election years as a political tool. If any political management of budgets is to take place, it should be done in years prior to elections.

15 We also had information on mayors’ level of education for 87% of the sample observations. We create a categorical variable, with values ranging from 9 (PhD holders) to 1 (elementary studies). In robustness checks, we ran all models including dummies for education and they were statistically insignificant.

16 Alternative control variables to unemployment rate were considered. For instance, we run all of our models with income per capita instead. It has a correlation coefficient of -0.34 with unemployment. Results remain consistent.

17 For the electoral cycle (2015–2018), cities start reporting whether city councils were governed in coalition or minority. We ran a means t-test comparing target revision in majority and coalition governments against the base group of minority governments. We compare means per year to control for general differences in economic conditions across time. Consistent with H2, we find that cities with majority and coalition increase their budget significantly more than the ones with minority in all years of the last electoral cycle except for 2017.

18 We further test the proportion of cities with a majority (coalition) government with a change in budgeted expenses higher than the average change in budgeted expenses in cities with a minority government. If there were no differences in terms of spending associated with government strength, we would expect to find a percentage not significantly different from 50%. We use asymptotically normally distributed test statistic calculated as $z=\frac{\hat{p}-p_0}{s_0}$, where p₀ is the hypothesized proportion, $\hat{p}$ is the observed proportion, p₀, and s₀ is the standard error $\hat{p}$ under the null hypothesis of p = p₀ = 0.5. The p-value is computed as $1-\varphi \left(z\right)$, where $\varphi (\cdot )$ is the cdf of a standard normal distribution. For 2016 and 2018, the proportion of cities governed by majority that experience a higher increase in their budgets is statistically higher than 50% as predicted in H2.

19 We further explored whether ideology had an impact on majority. In unreported results, we find this interaction term to be marginally significant: left wing governments are more likely to use majority to increase budgets to a larger extent. We thank a reviewer for suggesting this test.

20 According to Leone and Rock (2002), Aranda et al. (2014, 2017), the coefficient in for-profit organizations moves within the interval 0.60–0.80.

21 We thank one of the reviewers for suggesting the analysis of ideology on the budgeting process.

22 The autocorrelation of our dependent variable (change in budget) is –0.07. In unreported analysis, we built a histogram with the autocorrelation of each city. It shows that the majority of cities have an autocorrelation coefficient close to zero. Our inference does not appear to suffer from inefficient estimators; nevertheless, we use robust standard errors to correct for any potential autocorrelation in the error term.

23 In Panel B of Table 6 we run the same models with peers selected using similarity along structural determinants using a k-means clustering. Results are consistent.

24 ${\eta }_k^2$ measures the proportional reduction of the WSS for each cluster solution k compared with the total sum of squares (TSS).

Additional information

Funding

This work was supported by the Spanish Ministry of Science and Innovation [grant number P160133071].