Budget Transparency in Local Governments: An Empirical Analysis

Abstract

The aim of this paper is to shed additional light on the determinants of budget transparency in local governments. Our work is based on a Likert-type survey questionnaire specifically designed to measure budget transparency in small municipalities. The questionnaire is based on the IMF's revised Code of Good Practices on Fiscal Transparency (2007). Results from 33 Galician municipalities are used to assess its internal consistency and to test a battery of hypotheses on the determinants of budget transparency. While several previous findings of the literature are confirmed, some new results are also obtained.

Key Words:

• Government accounting
• local public finance
• municipal budgeting
• budget transparency

Notes

1. In this article, the word “scale” usually refers to “an instrument made up of multiple items that have a relationship to each other as well as to the concept of interest [budget transparency, in our case]” (Colton and Covert 2007, p. 249). A Likert scale is a scale composed of Likert-type items, as we explain in subsection 3.1. Although there are precedents for using Likert-type items in measuring budget transparency at the national level (Lavielle, Pérez and Hofbauer 2003, Pérez 2005), leading studies on determinants of budget transparency have tended to rely on the above-mentioned kinds of data.

2. According to Colton and Covert (2007, p. 68), content validity is the degree to which an instrument is representative of the topic and process being investigated.

3. For instance: (i) inclusion of comparative and non-financial information, (ii) explanation of deviations, (iii) consistency in format between the budget and the final report, (iv) issuing of monthly reports within four weeks, and (v) audit and issuing of the year-end report within of six months.

4. The second pillar also requires “clear mechanisms for the coordination and management of budgetary and extra-budgetary activities within the overall fiscal policy framework”. The third and fourth pillars incorporate innovative provisions such as:

i.	publication of a periodic report on long-term public finances;
ii.	openness in purchase and sale of public assets;
iii.	mechanisms to monitor follow-up actions recommended by the national auditor;
iv.	assessment of the fiscal forecasts, the macroeconomic forecasts, and their underlying assumptions by independent experts; and
v.	institutional independence of the national statistical body.

5. See also Jarmuzek (2006). Although he does not detail the effects of his instrumental variables on fiscal transparency, the variables used in the model include a political competition index (i.e. past turnover), the rule of law, and the media freedom index compiled by the Freedom House.

6. The authors base this hypothesis on model implications from Ferejohn's (1999).

7. As Islam (2006, p. 154) acknowledges, transparency indicators “could be strengthened by considering not just the frequency and availability of data but also the quality of the data produced by governments.”

8. Although we are well aware of the threats that public corporations and other entities depending on a municipality may pose to fiscal transparency, these issues have been considered beyond the scope of the more restricted notion of budget transparency adopted in our empirical research. All in all, revenues and expenditures of the autonomous bodies or agencies depending on a municipality are included in the consolidated budget of local governments. Regarding public corporations, only three of the municipalities under study have any of them.

9. Plotting the magnitude of the successive eigenvalues and applying the Cattell's (1966) Scree test, a sharp drop in eigenvalues from component one is observed.

10. Within a hierarchical factor model framework, Zinbarg et al. (2005) demonstrate that α underestimates reliability in the first three of four theoretical scenarios: (1) unidimensional scales with unequal general factor loadings, (2) multidimensional scales with equal general factor loadings, and (3) multidimensional scales with unequal general factor loadings. Only in the fourth case, unidimensional scales with equal general factor loadings – i.e., essential tau equivalence – is α as appropriate as ω to measure reliability. Vehkalahti, Puntanen and Tarkkonen (2006, p. 2) also use alpha's underestimating bias to justify the search for and proposal of a new estimator suggesting that this estimator, called Tarkkonen's rho, provides a better alternative for Cronbach's α.

11. According Revelle (2010, p. 216), the correlations associated with an ordinal scale are not Pearson's but Spearman's. However, Garson (2008) explains that ordinality in Likert scales refers only to an ordinal relationship of values within a single item: Likert response values are ordinal within any given item but sets of Likert items are not necessarily ordinal with respect to each other, and they can be used to form indexes.

12. When calculated from the correlation matrix (Table 4), a standardised alpha of 0.849 is obtained – the same could be calculated from the covariance matrix by standardizing and summing all items in the scale. In our case, the raw alpha provided by SPSS (0.850) is practically equal to the standardised alpha. In Table 5, raw alphas from SPSS are reported.

13. Although the widely-accepted social science criterion is that alpha should be 0.70 or higher for a set of items to be considered a scale, and even a lenient cut-off of 0.60 is common in exploratory research, many researchers require a score of 0.80 for a “good scale” (Garson 2008, 2010). On the other hand, a high alpha may also suggest a high level of item redundancy, wherein essentially the same item is rephrased in several different ways (Boyle 1991). Thus, for example, Fitzpatrick et al. (1998, p. 23) consider that alpha values should not be higher than 0.9 for scales which are used as research tools to compare groups. A very high α suggests that there is some redundancy among items, and the possibility that the items together are addressing a rather narrow aspect of an attribute.

14. In particular, we detected a high correlation between aging, population size, municipal public expenditure over municipal GDP, and per capita GDP. Youth migrate from backward municipalities in terms of GDP, bringing about both the ageing and reduction of total population. Insofar as local expenditure is partially financed by equalisation grants, the public expenditure/GDP ratio tends to be higher in the backward municipalities. Dropping per capita GDP and ageing from the specification substantially reduced multicollinearity among the independent variables.

15. Alt, Lassen and Rose (2006) find a non-significant effect. Esteller-Moré and Polo-Otero (2008) obtain a negative effect, but only for municipalities with over 5,000 inhabitants.

16. LaFaive (2009) detects a negative relationship between unemployment and the transparency of a Michigan's program for economic development. However, Peixoto (2010) finds no correlation between the US states’ levels of unemployment and the transparency of their recovery websites. Moreover, Andersen and Nielsen (2010, p. 28) suggest that the extremely damaging nature of procyclical fiscal policies during recessions may trigger reforms that increase the degree of fiscal transparency. Other studies on transparency also consider unemployment (Alt, Lassen and Skilling 2001, Rosendorff and Vreeland 2009), but not as a determinant of transparency.

17. In Alt, Lassen and Rose's (2006) empirical analysis, the resulting sign is positive for imbalance (both surplus and deficit), implying that a greater deficit leads to greater fiscal transparency.

18. La Porte, Demchak and Jong (2002) fail to obtain empirical support for their hypothesised positive effect of government size on openness, although in some of their tests the “government size” variable does not remain after removing accounting for obvious collinearity. Bastida and Benito (2007, p. 690–691) observe that the relative size of central government presents a low significant (p-value = 0.079) negative relationship (–0.260) with budget transparency. According to the authors, although the study shows that larger central governments are linked to lower levels of transparency, the low significance prevents using the evidence to draw a strong conclusion.

19. Alt, Lassen and Rose (2006) find that per capita general revenues are not significant in most estimates. In Esteller-Moré and Polo-Otero (2008) taxes are only significant at 10% in a cluster model, with the expected positive sign.

20. Although La Porte, Demchak and Jong (2002) find no empirical support for the hypothesized impact of democracy on web site openness, they appear not include the “voting” variable in testing. Esteller-Moré and Polo-Otero (2008), however, confirm a positive relationship between electoral participation and fiscal transparency, and Rosendorff and Vreeland (2009) do the same for democracy and transparency. Finally, Bastida and Benito (2007, p. 690) find no significant correlations between their democratic-level variables and budget transparency, although they suggest that the direction of the relationships supports the notion that greater levels of political and civil liberty correspond to higher fulfilment of the OECD Best Practices for Budget Transparency.

21. As pointed out in the above mentioned section, Alt, Lassen and Rose's (2006) econometric analysis corroborates this positive relationship when controlling for the effects of debt, fiscal imbalance, and polarisation, as well as when using an Arellano-Bond GMM first-difference estimation of their main equations. By contrast, Esteller-Moré and Polo-Otero (2008) find the variable is not significant, and Bastida and Benito (2007, p. 692) do not find any relationship between ideology and budget transparency.

22. Variables defined as those with a t-statistic below unity (in absolute value) in the first step: 4 variables in the case of equation [3][3] and up to 13 variables in the case of [4]. A Wald test on the joint significance of those variables clearly failed to reject the null hypothesis in both cases (p-values = 0.73 and 0.42, respectively)

23. RESET is a general test for the following types of specification errors: Omitted variables, incorrect functional form, and correlation between the exogenous variables and the random term which may be caused by measurement error or simultaneity, among other things.

24. Following a referee's suggestion, we also included the rate of female aldermen in the council as explanatory variable in specification [3], but it was not statistically significant (p-value > 0.50). In this sense, it is interesting to note that Piotrowski and Van Ryzin (2007) find that this variable is relevant to explain the demand for safety transparency, but not to explain differences in fiscal transparency.