Partisan theory and spatial interaction
effects in social expenditure policy in
the Netherlands
Abstract
This thesis investigates whether partisan effects and spatial interaction effects can be found in the social expenditure policy of municipalities in the Netherlands After analyzing data from the period 2005-2009 from over 300 municipalities, it can be concluded that there is spatial interaction but a partisan effect is not found.
1
1. Introduction
Left-wing parties have the reputation that they take care of poor people while right-wing parties are for rich people. Can we justify this with empirical evidence?
This thesis analyses1 data from Dutch municipalities looking for evidence that confirm partisan theory. The first paragraph describes partisan theory. The second paragraph
discusses partisan effect with respect to social expenditures. Municipalities are neighbors of eachother and spatial interaction could influence the policy. This chapter ends with a
discussion of spatial interaction effects and explains why these effects are expected.
1.1 Partisan theory
According to Hibss (1977,1992) , partisan theory describes two groups of people. The first group is relatively less wealthy. Therefore, this group is more concerned about unemployment than about inflation. Inflation makes the income distribution more equal because it lowers the real wage and increases that part of income that goes to the labor force. The second group is more wealthy as a result of which inflation makes this group less wealthy. Therefore, this group is more concerned about inflation than about unemployment. Hibbs (1992) argues that the constituencies of left-wing parties are less wealthy than the constituencies of right-wing parties. Therefore, the less wealthy of the economy vote for the left-wing party, while the wealthy people vote for the right-wing party.
Politicians making macroeconomic policy face a tradeoff between inflation and unemployment. This choice is best described by the Philips curve. Stimulating aggregate demand leads to lower unemployment at the cost of higher inflation. Without stimulating of aggregate demand, it is not possible to reach full employment.
To get votes, the political parties should not disappoint their constituencies. Therefore, left-wing parties make policy that stimulates demand, while right-wing political parties make policy that lowers demand in order to decrease inflation.
The original partisan theory concerns economic policy. Since part of economic policy concerns expenditures, one can expect that this theory also holds for social expenditure policy.
1 I am grateful to dr. J.P Elhorst for his advices about the analysis and writing this thesis. I also appreciated the
2
1.2 Partisan theory and social expenditures
Social expenditures are all expenditures a government has to pay for people that need
financial support or need special facilities. Among these people are the homeless, disabled or elderly people and the people that are poor or unemployed. This thesis is about social
expenditures made by the council of municipalities.
This thesis investigates whether partisan theory can be applied to social expenditures. The main assumption of the partisan theory is that poor people support left-wing parties and that rich people support right-wing parties. To satisfy their constituencies left-wing parties have to increase social expenditures. This benefits poor people and therefore increasse the popularity of left-wing parties among the poor voters. Right-wing parties do not want to increase social expenditures because rich people do not benefit from the increase in social expenditures. Furthermore, since these voters have to pay the higher tax bill, they do not support an increase of social expenditures. Therefore, to satisfy their constituencies, right-wing politicians do not support any increase of social expenditures. In this theory, political parties are being divided in left-wing parties and right-wing parties. At the time the original partisan theory was described this difference was very clear. These days it is hard to tell for most parties whether they are right-wing or left-wing. Therefore, this thesis investigates the degree of influence each different party has on social expenditure instead of only looking at a partisan effect.
1.3 Spatial interaction effects
The Netherlands is divided in 12 provinces which have their own council. These provinces are divided in municipalities with their own council. The Netherlands has over 441
municipalities. In order to estimate partisan effects, data from these different municipalities in the Netherlands is used2. It would be unrealistic and therefore wrong to assume that these municipalities operate solitary. Decisions of a municipality to increase social expenditures are noticed by voters from other municipalities. When, for example, one municipality builds new facilities for disabled people, people from other municipalities move to this municipality to enjoy these new facilities or demand their own council to spend more money on the same kind of facilities. Besides that, municipalities are bounded by the regulation of the province. That also decreases their independency. When the economic situation in the Netherlands deteriorates, unemployment in all municipalities increases. This affects the amount of
2
3 homeless people and the expenditures on facilities for these people in all municipalities at the same time. These kind of effects are called spatial interaction effects. Ignoring these effects makes the model less useful.
In sum, I pose in the following research question:
'Are there partisan effects and spatial interaction effects in the social expenditure policy of municipalities in the Netherlands?'
1.4 Relevance
The answer to this research question is not only interesting from a scientific point of view as a variant or extension of the existing partisan theory. It is also important information for voters before the elections when the politicians make promises. When voters choose to vote for a certain political party, hoping that this makes their situation better, they have the right to know what they can expect from this political party.
At the same time, by increasing the transparency of governmental policy, this thesis forces politicians to increase their reliability. When politicians know that scientists are doing research about the promises of politicians, they think twice when they make promises and take decisions because they know that the public finds out that they are unreliable. Therefore, this thesis not only benefits economic science but also democracy. The next chapter gives an overview of the relevant literature. Chapter 3 and 4
4
2. Literature overview
This chapter discusses relevant articles about partisan theory and interaction effects. It starts with a discussion about the development of partisan theory. This development started with Hibbs (1977) describing the original partisan theory, then Alesina (1989) adapted partisan theory to the rational expectations hypothesis, and three years later, Hibbs (1992) took care of the comeback of the original partisan theory.
All the literature discussed in paragraph 2.1 is about economic policy while this thesis is about a part of that policy: expenditures. Therefore, Paragraph 2.2 focuses on evidence of partisan effects in expenditures. Doing analysis about municipalities, one has to take into account the interaction effects between the municipalities. This is clear from the literature about spatial interaction effects. Part of that literature is discussed in paragraph 2.3. This chapter ends with a summary.
2.1 First research about the Partisan theory
There are a lot of articles written to answer the question whether there is a partisan effect or not. One of the first is Hibbs (1977) who used data from 12 countries. He found that
countries with a high percentage of socialist parties in the government in a certain period had high inflation and low unemployment in that period, a finding that supported the partisan theory. Besides this analysis he paid special attention to the United states of America (USA) and Great-Britain. Both countries have a two party system that consist of a left-wing and a right-wing party. For both countries his analysis showed that unemployment was relative low under a left-wing government and relative high under a right-wing government. So, for both countries he found evidence of partisan theory.
The introduction chapter of this thesis describes the partisan theory as introduced by Hibbs (1977). Just before Hibbs (1977) was published, the rational expectations revolution started. Because of the rational expectations economic policy became ineffective (the so-called policy ineffectiveness hypothesis). This made the partisan theory useless, because when
5 right-wing government. When the people make expectations that appear to be wrong later, there is a partisan effect possible. There are two situations that can occur: people expect a left-wing government but end up with a right-left-wing government and vice versa. Both situations are discussed here.
When people expect the left-wing party to win the election they estimate a high
inflation rate. In order to remain at the same real income level they demand a higher wage rate when they negotiate their contracts. The demand for labor decreases, as a result output
decreases and the economy might go into recession. The next time they sign a contract they know that there is a right-wing government that takes care of inflation and, realizing that, they do not demand high wages. Output returns to a natural rate and, at the end of the period, expectations are conform reality and the economy is recovering from the recession. When people expect the right-wing party to win the election they estimate a low inflation rate and demand a low nominal wage rate. When they see that the left-wing party has won the election, they cannot change the contract anymore and have to work for the low wage rate. Firms produce more and output increases to a level above the natural rate. The economy is booming. This boom is temporary, because when the next contract is signed the estimated inflation is high and output decreases to its natural rate. However, inflation does not decrease to a lower level because people expect it to be high. This decreases aggregate demand. To prevent a recession, the government has to stimulate the economy and this brings inflation up to the level of expected inflation. So even when there are rational expectations there is a partisan effect possible.
Two different partisan theories are discussed. As Alesina (1989) pointed out, there is an important difference between the partisan theory of Hibbs (1977) and the rational partisan theory he presents. The effects of a new government on real output are permanent during the election period in the theory of Hibbs (1977), while the effects on real output are transitory in the theory of Alesina (1989). Alesina (1989) found weak evidence for his rational partisan theory.
6 Hibbs (1977).
Hibbs (1992) discussed studies applying to partisan theory in the period of fifteen years after the publication of Hibbs (1977). The conclusion of Alesina and Roubini (1992) that they have found empirical evidence in favor of rational partisan theory, is criticized by Hibbs (1992). He disagreed with their conclusion because they did not use the strict version of the rational partisan theory. The strict version of the rational partisan theory assumed that only unexpected policy can have real effects, but they assumed instead that also expected policy can have real effects. According to Hibbs (1992), the effects they described could be an indication of a rational partisan effect but were also consistent with a gradual steepening of the Philips curve that is caused by a change in partisan goals. Hibbs (1992) argued that the data that Alesina and Roubini (1992) have used, could contain random walk effects and he concluded that that the original partisan theory cannot be rejected.
Erlandson (2004) tested the partisan theory in two ways using data from Sweden for the period 1985 to 2000. He started with a model that used the real output as dependent variable and then tested a model in which unemployment was the dependent variable. When he used output he found a significant partisan effect while when he used unemployment he did not find a partisan effect.
The partisan theory and the rational partisan theory are discussed so far. One of the main differences between these two theories is that the rational partisan effects are transitory and the partisan effects are not. The subject of this thesis is the partisan effects that can be found in the social expenditure policy of municipalities. An important question is whether this partisan effect can be seen as a rational partisan effect or a regular partisan effect. The social expenditure effects are not transitory because they do not depend on contracts. The agents are not bounded by contracts. Therefore, this thesis discusses a social expenditure version of the original partisan theory of Hibbs (1977).
2.2. Looking at the expenditures
Tufte (1978) discussed different articles and books about the partisan theory. He also investigated partisan theory by counting certain words in party information and official announcements of the president of the USA. Left-wing politicians use the word
7 of the message. For example, a president who is explaining why he is satisfied with the level of unemployment at the moment uses the word unemployment very often. However, he does not change do something about it because he is satisfied with it. Therefore, counting words does not prove anything.
Until now, articles are discussed that investigated partisan theory by looking at the unemployment rate and inflation rate. The difference between right-wing politicians and left-wing politicians is not limited to the priority politicians give to unemployment and inflation. Tufte(1978) added to this difference that left-wing politicians, besides low unemployment, also want high expenditures and an equal income distribution. Right-wing politicians want low expenditures and do not focus on an equal income distribution. It is important for this thesis that Tufte (1978) made the partisan theory wider than Hibbs (1977) because this thesis also uses partisan theory to explain differences in expenditures. Two examples of other research about partisan effects that were found in expenditure policy are discussed now. Petry et al. (1999) focused on the partisan effects in the government budget of the provinces of Canada. They estimated a separate model for each province and found mixed results: Some provinces showed a partisan cycle, while others do not. Besides testing for the partisan theory, they also tested whether there is interaction between the electoral cycle (also called political business cycle) and the partisan cycle. They found evidence for this
interaction. This means that in non-election years politicians focus their policy on ideology to satisfy the constituencies, while in election years they focus on winning the election by stimulating the economy to keep as much voters satisfied as possible.
8 comments about the fact that there is no partisan effect in the public administration
expenditure. These expenditure consist of the loan of the workers at the public institutions and investments by the federal government. One could expect that left-wing parties, the parties that want to improve the income distribution, want to increase the loans of the employees of public institutions and want better public facilities (for example libraries). Therefore, a partisan effect in this kind of expenditures is likely, even though they did not find one.
2.3 Spatial interaction
Leenders (2002) described two different forms of interaction: communication and
comparison. Comparison between units leads to the same behavior but not to the same belief. This means that when units do something because other do it, they do not necessarily know what the idea behind the behavior is. Communication leads to the same belief but does not always lead to the same behavior. It is not possible to determine empirically which caused the behavior of the unit. Manski (1993) described three kinds of interaction effects: endogenous effects, exogenous effects and correlated effects. When the propensity to behave in a certain way depends on the behavior of the group then there is an endogenous interaction effect. When the exogenous characteristics of the group influence the behavior of the individual there is a exogenous interaction effect. Manski (1993) defined correlated effects as the effect that individuals in the same group behave similarly because they have similar characteristics or face similar institutional environment. He proved in a theoretical model that it is not possible to estimate three interaction effects in one model and he also showed that someone never can determine which interaction effect causes the behavior that we see without knowing how the individual form a reference group and how he perceives reference group outcomes.
2.4. Evidence of spatial interaction between municipalities
In the literature three sources of interaction were described : spillovers, yardstick competition and welfare competition. See Ermini and Santolini (2010). Spillovers can be found in the area of policy, security and infrastructure. When a municipality increases the spending on police, crime rates in other area’s increases because criminals move to another area. Spillover effects can be positive as well as negative. Voters have little information to assess the achievements of the local government. Therefore they compare their facilities with the facilities of
9 municipalities. This makes the local economy wealthier. They also want citizens to live in their area because that increases the tax base. Therefore, municipalities have to take into account that people and companies can move out of the area.
Allers and Elhorst (2005) found strong evidence in favor of spatial interaction between the municipalities in the Netherlands when they analyzed the property taxes. Municipalities can not set property taxes without taking into account that voters compare taxes and public expenditures between municipalities. Besides this interaction effect, they also found that local governments with a high proportion of right-wing politicians set lower taxes. This is in accordance with partisan theory because right-wing voters are wealthier than left-wing voters and are owners of property. Therefore they do not want high property taxes. It is clear that a partisan effect can be seen here.
Heyndels and Van Uchelen (1998) found evidence of tax mimicking in Belgium. They discovered that first order and second order neighbors influence each other. Ermini and Santolini(2010) analyzed data from municipalities in Italy and found interaction effects in the (total) expenditure policy between municipalities. When they focused on separate
expenditure area’s they didn’t find interaction effects in the area of social expenditures. Case et. al.(1993) found spatial interaction between the social expenditure policy of different states in the USA.
Summary
This chapter first showed the development of the partisan theory and the evidence of the partisan theory others found in the expenditure policy. The second half of this chapter
discussed the sources of spatial interaction effects and articles that showed that one can expect to find these kind of effects between municipalities. The next chapter contains a description of the model that is used.
3. Model
10
3.1 spatial interaction models
To estimate a spatial interaction model, a choice have to be made between the spatial lag model, the spatial error model and the spatial Durbin model. All these models use a spatial weights matrix W. There are many possible variations of W. For example, the elements of this matrix could be the distance between two points, the inverse distance or W could be a
contiguity matrix. When W is a contiguity matrix the elements in the matrix are equal to 1 when the units are neighbors and the elements are 0 when units are not neighbors. By assumption, an unit is no neighbor of itself so the diagonal elements are 0. According to Leenders (2002), the matrix should be normalized to allocate the influence the units have on each other. It is possible to normalize across columns or rows. When each unit is divided by the column sum then the matrix is column normalized. After column normalization, each unit has the same influence on the other units if they are neighbors. Row normalization is dividing each element by the sum of the row elements. After W is row-normalized then each unit is under the same influence from each other unit (when they are neighbors otherwise the
influence is zero). Sometimes different spatial weight matrix can be used. To choose the right spatial weight matrix, the loglikelihood function value is used.
The first model that is discussed is the spatial lag model. Elhorst (2010) describes the spatial lag model as equation 3.1
3.1
In which Y is a N × 1 vector consisting the dependent variable. The first part at the right of the this equation is the endogenous interaction part. It contains of the spatial weight matrix W, the dependent variable Y and the spatial autoregressive coefficient p. When p > 0, the value of the dependent variable depends on the value of the dependent variable of neighbors. The second part is the N × 1 vector of 1’s times the constant α. The third part consist of the N × K matrix of independent exogenous variables times K × 1 vector β, in which there is a different value of β for each explanatory variable. The last part of the equation is the independently
distributed vector of error terms ε. These error terms have zero mean and variance δ2 and are equally distributed.
The second model, the spatial error model, is used to estimate the effect of the
11
3.2
3.3
When spatial autocorrelation coefficient λ>0, then the error term of each individual has a correlation with the error term of other individuals.
The spatial Durbin model can be changed into the spatial lag or spatial error model and contains an extra part WXθ . This part estimates the exogenous interaction effects. Equation 3.4 is the spatial durbin model.
3.4
3.2 the ols model
After discussing the general forms of the spatial interaction models the remainder of this chapter describes the model that is used in this thesis to analyse the social expenditure policy. The dependent variable of this model are the amount of social expenditures (SE). Social expenditures are expenditures for different social activities and payments like activities to help unemployed people find work again, social benefits (which are payments for the poorest people), costs of advice for people that have social problems, costs of help for disabled people or people that do not have the Dutch nationality. The social benefits are part of the social expenditures, but because the amount that each unemployed person receives is determined by the central government in The Hague, the expenditures to social benefits can not be
influenced directly by policy of municipalities. Therefore, it is better to withdraw the social benefits from the social expenditures. To see the difference, the model is estimated with and without social benefits.
Social expenditures depend on the amount of unemployed people in a country.
Therefore, the amount of unemployed people is one of the independent variables of the model and a positive relationship is expected between the social expenditure and the amount of unemployed people.
12 that do not have seats in every municipality. These seats are added into four groups: the local Christian Parties, local independent parties, local progressive parties and rest parties. The last group consist of all seats of parties that could not be placed in one of the other categories.
The following ols model is used:
it it 66it it it ! it "
# $% & !$ ' %( % % )*!+ , 3.5
In which
ESit = total social expenditures
u
it = amount of people that are unable to do workcdait = Number of seats in the municipality council that belong to the political party CDA
d66it = Number of seats in the municipality council that belong to the political party D66
vvdit = Number of seats in the municipality council that belong to the political party VVD pvda it = Number of seats in the municipality council that belong to the political party PVDA
spit = Number of seats in the municipality council that belong to the political party SP
cuit = Number of seats in the municipality council that belong to the political party CU
glit = Number of seats in the municipality council that belong to the political party
Groenlinks
sgpit = Number of seats in the municipality council that belong to the political party SGP
liit = Number of seats in the municipality council that belong to local independent parties.
lc it = Number of seats in the municipality council that belong to local Christian parties
lpit = Number of seats in the municipality council that belong to local progressive parties
restit = Number of seats in the municipality council that belong to political parties that do not belong to one of the other groups
µ1 = municipality specific effects. λ1 = time specific effects
ε it = error term with average = 0 and standard deviation δ2
13
3.3 fixed effects and specific-to-general method
To estimate the model, it could be necessary to use time fixed effects or spatial fixed effects or both. This is necessary when some period or area has certain characteristics that influence social expenditures. This could be a period of an economic depression (time) or an area in which the inhabitants are wealthier than in other area (space). To determine whether the fixed effect are needed the Likelihood Ratio test is used. Using fixed effects has influence on α. It is only possible to estimate α when ∑ μ 0 and when ∑ λ 0 . When α is not estimated then these restrictions can be ignored.
Combining the spatial- and time period fixed effects with spatial interaction leads to biased estimates of the parameters. The bias depends on which fixed effect is used and the number of observations and number of years that is used. To correct the bias, Elhorst (2011) uses a bias correction procedure and in this thesis the same procedure is used.
One of the spatial interaction models that could be used is the spatial durbin model. When the spatial durbin model is used, two effects are estimated: direct effect and an indirect effect. The direct effect is the effect that a change in the independent variable of a
municipality has on the dependent variable of the same municipality. The indirect effect is the effect the a change of the independent variable of a municipality has on the value of the dependent variable of one other municipality. Elhorst (2010) shows that the direct and indirect effects differ per unit, so in this case, each municipality has his own direct and indirect effect. The estimates of the direct and indirect effects in this thesis are averages of the effects of all municipalities. The difference between direct and indirect effect only applies to spatial Durbin and spatial lag models. The spatial error model only estimates direct effects which are equal to the parameter estimates.
14
4. Data
The data of expenditures and inability-to-work comes from the website of the central bureau of statistics of the Netherlands3. The political data comes from the university of Groningen. In 2009 the Netherlands was divided in 441 municipalities. Unfortunately, due to missing data it is only possible to use 307 municipalities.
To estimate spatial interaction effects, normally in this case, a contiguity matrix is used that indicates for each two municipalities whether these are neighbours or not. Because of the missing data, in this thesis a contiguity matrix is used that indicates municipalities as
neighbours when they are in the same province. This leads to a special spatial weigh matrix with the following characteristic: When there is a ‘1’ in W then the row-sum is equal to the column-sum. This is caused by the definition of neighbours that is used in this thesis. When municipality A is neighbour of B and C because they are in the same province, then A , B and C all have two neighbors. W(A,B) =W(A,C)=1.This means that the column sum of A=column sum of B=column sum of C =2. The row-sum of A=the row sum of B =row sum of C=2. Dividing each element by the column sum does not differ from dividing each element by the row-sum. Also assume there is a municipality D which is no neighbour of the other
municipality then W(A,D)=0. For all the municipalities that are not neighbour of A then the element is ‘0’. Dividing by row or column sum does not influence this value. This means that the row-normalised W is exact equal to the column-normalised W. The method of choosing, based on the loglikelihood function value is not necessary to use here. In the analysis the column normalised contiguity matrix is used. Table 4.1 gives more information about the data that is used.
3
15
Table 4.1. More information about the data.
Variable min max Average Standard
dev. Social expenditure
per capita (€)
SE 0,12 2,31 0,51 0,29
16
Table 4.1 (continued)
5. Results
This chapter discusses the results of the analysis. The first paragraph shows the results of the ols-model. The second paragraph describes results of the spatial interaction model. The last two paragraphs describe respectively the two-wings model and the three groups model. Some of the tables that present the results are very large. Therefore, the reader can find these tables in the appendix.
5.1 The ols model.
The starting point of the analysis is the ols model containing all the parties separately using a dependent variable that contains social benefits. In the appendix the reader can find table A.1. This table gives the results of this model using different combinations of fixed effects. The second column of table A.1 gives the result of the ols model without fixed effects. A clear partisan effect can be seen here because the left-wing party SP has a positive effect on social expenditures and the right-wing party VVD has a negative effect on social expenditure. To see whether it is necessary to use fixed effects the Likelihood ratio test is used. From the results of this test we see that spatial fixed effects are needed. (The Likelihood Ratio test- value is 3329.6528 with 307 degrees of freedom and probability 0.000). Time-period fixed effects are also necessary. (The Likelihood Ratio test- value is 342.1971 with 5 degrees
4
The council of municipality Lith consist totally of local independent parties.
Variable min max Average Standard
dev.
Local Independent Parties (%)
li 0,00 1004 24,34 17,43
Local Christian Parties (lc) (%)
lc 0,00 33,33 0,29 2,43
Local Progressive Parties (%)
lp 0,00 45,45 2,21 6,81
Rest Parties (%)
17 of freedom and probability 0.000).The Hausmann test gives an indication whether random effects are better than fixed effects. The Hausman test-statistic value is 92.6928 with 27 degrees of freedom and probability 0.000 ) so we have to use fixed effects. The results of the ols model with spatial- and time period fixed effects can be found in the last column of table A.1. We see that in the ols model with both kind of fixed effects, none of the political variables are significant. We also see that there is a significant influence of the amount of people unable to work on the social expenditures. In each of the models discussed in this thesis this variable is significant. Because this thesis is about partisan effects I this variable is not mentioned in the
discussion.
18 Box 5.1 mathematical answer
Mathematics is used to answer the following question:
Leads a decrease in the dependent variable to a direct loglikelihood effect?
To answer this question, a model is needed of which the loglikelihood function was given and properly described. That is why the model of Valavanis (1959) is used here. This section starts with a description of the model and loglikelihood function. Then partial derivatives are derived. At the end of this section the conclusions are discussed.
As starting point, the model of Valavanis (1959) page 29 is used.
1 ∝ 34 5.1
Equation 5.1 s the consumption function. In which Ct is consumption per time, Zt is income. α
en γ are respectively the constant term and the coefficient of the variable income. ut is the
normal distributed error term with constant variance. The error term is independent in time. Valavanis (1959) derives the following loglikelihood function Valavanis (page 31):
log 8 9:log 2 < 9:log =>>9 ?=>> @A ∑ B 5.2
In which S is the seize of the sample. when log L reaches the maximum, =>>
:∑?1B9 9 34B @ 5.3
This information from Valvanis(1959) is used here to derive the derivative of log L with respect to Cs.
Substituting 5.3 into 5.2:
log 8 9:log 2 < 9:log ?:∑?1B9 9 34B @ @ 9 C:∑?1B9 9 34B @ D A
∑ B 5.4
taking the derivative
E FGH I
EJK 9
:
FL ' ?∑?1B9 9 34B @A ∑?1B9 9 34B @A ! ∑ B 5.5
19
In this paragraph it becomes clear that it is necessary to use spatial and time fixed effects and that the social benefits has to be withdrawn from the social expenditure because of theoretical reasons. It is proven that it could be misleading to use the loglikelihood function value as indicator whether the model improves or not when one changes the dependent variable.
5.2 spatial interaction model
Municipalities possibly interact with each other. Therefore, it could be necessary to test for spatial interaction.
The likelihood maximum test results indicate that spatial error can not be ignored (value 550.1421, prob. 0.000 ). The robust LM-test of spatial lag gives no significant spatial lag ( value 0.1611, prob. 0.688).
Starting point here is the spatial Durbin model. The Wald-test indicates whether this model can be simplified to spatial lag or spatial error model. Table 5.1 gives the result of the Wald test. Both the spatial lag and spatial error model are rejected in favor of the spatial durbin model. That is why the spatial durbin model is the model that is used here. Table A.3 gives the results of the spatial durbin model.
Table 5.1 the Wald test for spatial lag and spatial error.
The second column of table A.3 gives the Spatial durbin model with bias corrected values. The third column gives the values that are not bias corrected.We see that none of political variables have a coefficient that significant differ from zero. We also see that the bias correction leads to an decrease of the t-value. Table A.4 gives the direct and indirect effects. We see that only local Christian parties have an indirect effect. This means that when Christian parties in a municipality gets more seats in the council, that the social expenditures in neighboring municipalities increases. Other political parties do not have a direct or indirect influence on social expenditure. The spatial lag coefficient is significant so there is en endogenous interaction effect.
We have found that none of the political parties have direct influence on the social
expenditure. In the Netherlands, political parties have to form coalitions otherwise there is no majority in the council. The next two paragraphs discuss models in which the political parties are combined.
Wald-test statitistic
Probability
spatial lag 34.9964 8.4775e-004
20
5.3 the two-wings model
The second model has two different political variables, right-wing and left-wing. The political parties are divided in right-wing and left-wing based on the direct effects that are estimated in the spatial durbin model. (see table A.4)
For this model, the political parties that have a negative direct effect are called right-wing and political parties that have a positive direct effect are called left-wing. Dividing the political variables leads to the following division:
Left-wing parties (direct effect in parentheses)
CU (-0,0035), Rest (-0,0023), SGP (-0,0012), VVD (-0,0007), CDA (-0,0006), Groenlinks (-0,0006), D66 (-0,0003), Loc. Prog (-0,0003) and Local Ind (-0,0003)
Right-wing parties (direct effect in parentheses) SP (0,0007), PVDA (0,001) and Local Chr (0,0014) The following model is estimated.
∝ M N 8 5.6
In which
Eit = social expenditures without social benefits.
Uit
= amount of people that are unable to do workRWit = Number of seats in the municipality council that belong to Right-wing political parties
LWit = Number of seats in the municipality council that belong to Left-wing political
parties
The Likelihood-Ratio test indicates that time period and spatial fixed effects are needed. (time period fixed effects 652.5974, degrees of freedom: 5, prob. 0.0000 and spatial fixed effects: 2649.3088, degrees of freedom 307, prob. 0.0000). Thereofre, the model is estimated with both fixed effects. This makes the ∝ disappear and leads to the following model:
90.060931∗ M 90.000328 N 0.000959 8
(-5.0924) (-0.4899) (1.1772)
R2= 0.0247
21 We see that none of the political variables are significant. The robust LM test of spatial lag and spatial error indicate that spatial lag can be ignored (value = 0.8307, prob. 0.362) and spatial error can not be ignored (value = 39.8863, prob. 0.000). The Wald test indicates that the spatial durbin model can be simplified to the spatial error model (value 5.7247, prob. 0.1258) or spatial lag model (value 5.2438, Prob. 0.1548) . Combining the robust LM Test and the Wald test leads to the conclusion that the spatial error model is the model that should be used. Testing the spatial error model gives the following model.
R2=0.9066
90.068707∗ M 90.000512 N 0.000638 8
(-4.9179) (-0.7350) (0.7515)
u =0.590003* Wu + ε (15.7371)
t-value in parentheses * significant at 1%.
From this model it becomes clear that there is no partisan effect. There is a significant positive spatial error effect. This means that municipalities increase the social expenditure when other municipalities in the same province do the same. This could be an indication of the influence that the council of the province has on the social expenditure policy of municipalities.
5.4 the three groups model
This paragraph discusses the last model that is estimated. This model contains three political variables: Rigth-wing, center and left-wing. Like the second model. the division between these categories is also based on the direct effect estimated from the spatial durbin model Right-wing parties have a direct effect coefficient less than -0.0007. Left-wing parties has a direct effect coefficient larger than 0.0007. The parties with a coefficient between -0.0007 and 0.0007 are in the center category.
Left-wing (direct effect in parentheses)
SP (0,0007), PVDA (0,0010) and Local Chr (0,0014)
Center (direct effect in parentheses)
22
Right-wing (direct effect in brackets)
CU (-0,0035), Rest (-0,0023), SGP (-0,0012), VVD (-0,0007),
Using that division of parties, the ols model becomes
∝ M 8 1*W+*) N 5.7
In which
Eit = social expenditures without social benefits.
Uit = amount of people that are unable to do work
LWit = Number of seats in the municipality council that belong to Left-wing political
parties
Centerit = Number of seats in the municipality council that belong to central political parties
RWit = Number of seats in the municipality council that belong to Right-wing political parties
The result of the Likelihood Ratio test is that spatial fixed effects ( value 2640.2768, degrees of freedom 307, Prob. 0.000) and time-period fixed effects (value 658.4298, degrees of freedom 5, prob. 0.000) can not be ignored.
This leads to the following ols model: R2=0.0284
90.060497∗M 0.000886 8 90.000337 1*W+*) 90.001618 N
(-5.063686) (1.088660) (-0.504427) (-1.893196)
t-value in parentheses * significant at 1%.
23 R2= 0.9070 90.068081∗ M 90.001628 8 90.000474 1*W+*) 0.000625 N (-4.8800) (-1.8675) (-0.6818) (0.7387) u =0.592988* Wu + ε (15.9303)
t-value in parentheses * significant at 1%.
The results make clear that there is no partisan effect. A spatial error effect is significant.
5.5 sensitivity analysis
the analysis so far is done with all kind of municipalities. Some municipalities are little villages and other are large cities. To investigate whether this influences the results, the third model is estimated with only the municipalities having less than 50,000 inhabitants. The second column of table 5.2 repeats the results of the third model while the third column gives the results of the model in which only the smallest municipalities are used.
We see in table 5.2 that the coefficient estimates of the right-wing and left–wing political parties do not have the same sign. For example, left-wing parties have a negative
24 Table 5.2 sensitivity analysis
Using all municipalities
Using only
municipalities with less than 50,000 inhabitants. Unable to work -0.068081* (-4.8800) -0.073183* (-5.0275) Left-wing -0.001628 (-1.8675) 0.000907 (0.9845) Central -0.000474 (-0.6818) -0.000413 (-0.5178) Right-wing 0.000625 (0.7387) -0.001644 (-1.7208) Spatial error autocor..coeficient 0.592988* (15.9303) 0.589003* (15.6707) t-values in parentheses. * is significant at 1%
6. Conclusion
After reading the literature and analyzing the data, it is time to look back at the question posed in the introduction: 'Are there partisan effects and spatial interaction effects in the social expenditure policy of municipalities in the Netherlands?'
25 and two of them are discussed here.
This thesis uses data from just one election and it is possible that one election is not enough to show the influence of political parties on social expenditure policy. The only way to prove this, is to wait ten years and do the same analysis again. It is also possible that political parties really do not have any significant influence on social expenditures because of rules and orders from higher governments. This could be a subject of further research.
The three models that are estimated all use spatial interaction effects. The spatial durbin model combines the spatial lag model and the spatial error. The coefficient of the spatial lag variable is significant so there is a endogenous interaction effect between municipalities. The two-wings model and the three groups model both are spatial error models. So, besides endogenous interaction there is also a correlated error term.
What can we say about Dutch politicians and their reliability? Right-wing politicians do not have a negative influence on social expenditures while their constituencies are wealthy people that do not want to pay a high tax bill. Left-wing politicians do not have a positive influence on social expenditures while their constituencies expect them to make the situation of poor and helpless people better. Politicians do not keep their promises about social expenditure policy. We should that keep in mind when we listen to politicians and when we use the red pencil….
26
References
Alesina, Alberto. (1989). “Politics and business cycles in industrial democracies”, Economic policy, Vol. 8 (apr), 55-98.
Alesina, Alberto and Nouriel Roubini. (1992). ”Political cycles in OECD economies”, Review of economic studies, Vol. 59, 663-688.
Allers, Maarten A. and J. Paul Elhorst. (2005). “Tax mimicking and yardstick competition among local governments in the Netherlands”, International tax and public finance; Vol. 12, 493-513. Case, Anne C. and Harvey S. Rosen. (1993). “Budget spillovers and fiscal policy interdependence”,
Journal of public economics, Vol. 52, 285 -307.
Dalen van, Hendrik P. and Otto H. Swank. (1996). “Government spending cycles: Ideological or opportunistic?”, Public choice, Vol. 89, No. 1/2, 183-200.
Elhorst, J.Paul. (2010). “Applied spatial econometrics: Raising the bar”, Spatial economic analysis, Vol. 1, (March).
Elhorst, J. Paul (2011). “Matlab software for spatial panels”, presented at world conference of the spatial econometrics (June 2010). Downloadable from
www.regroningen.nl/elhorst/software.shtml.
Erlandsson, Mattias. (2004). “Partisan differences in Swedish macroeconomic policy”, Public choice, Vol. 120, No. 1, 205-220.
Ermini, Barbara and Raffaella Santolini. (2010). “Local expenditure interaction in Italian
municipalities: Do local council partnership make a difference?”, Local government studies, Vol. 36, No. 5, 655-678.
Heyndels, Bruno and Jef Vuchelen. (1998). ” Tax mimicking among Belgian municipalities”, National tax journal, Vol. 51, No. 1, 89-102.
27
Hibbs, Douglas A. (1992). ”Partisan theory after 15 years”, European journal of political economy, Vol. 8 (Oct), 361-373.
Leenders, Roger, Th. A.J. (2002). “Modeling social influence through network autocorrelation: constructing the weight matrix”, Social networks Vol. 24, 21-47.
Manski, Charles F. (1993). “Identification of endogenous social effects: The reflection problem”, Review of economic studies, Vol. 60, No. 204, 531-543.
Petry, François, Louis M. Imbeau, Jean Crête and Michel Clavet. (1999). “Electoral and partisan cycles in the Canadian provinces”, Canadian Journal of Political Science,
Vol. 32, No. 2 (June), 273-292.
Tufte, Edward R. (1978). Political control of the economy, Princeton University Press, New Jersey. Valavanis, Stefan (1959), Econometrics, an introduction to maximum likelihood methods, McGraw Hill, New York.
28
Appendix 1. the results of the ols model.
Tabel A1. The results of the ols model including social benefits
t-value in parentheses * significant at 1%.
No fixed effects Spatial fixed effects Time fixed effects Spatial and time fixed effects
R2 0.4148 0.3889 0.4456 0.0319
Loglikelihood 7.2180e+003 8.8010e+003 7.3073e+003 8.9721e+003
29
Table A.1 continued.
t-value in parentheses * significant at 1%.
30
Table A.2 ols model without social benefits in the social expenditures
No fixed effects Spatial fixed effects Time fixed effects Spatial and time fixed effects
R2 0.3345 0.5237 0.3948 0.0322
Loglikelihood 7.8356e+003 9.0377e+003 8.0761e+003 9.3491e+003
31
Table A.2 (continued).
t-value in parentheses * significant at 1%.
32
Table A.3 spatial durbin model
t-value in parentheses * significant at 1%.
Bias corrected values No bias correction.
33
Table A.3 (continued)
W*Unable to work 0.071154 (1.8799) 0.071154 (2.1018) W*CDA -0.000231 (-0.0219) -0.000231 (-0.0245) W*D66 -0.007434 (-0.3457) -0.007434 (-0.3864) W*VVD 0.006421 (0.3218) 0.006422 (0.3596) W*PVDA 0.008140 (0.7456) 0.008140 (0.83301) W*SP 0.013825 (0.4739) 0.013825 (0.5298) W*CU -0.006108 (-0.3929) -0.006108 (-0.4392) W*Groenlinks 0.018625 (0.9287) 0.018625 (1.0378) W*SGP 0.020256 (1.6213) 0.020256 (1.8107) W*Local Ind. 0.001345 (0.1374) 0.001345 (0.1536) W*Local Chr. 0.120615* (3.1484) 0.120617* (3.5061) W*Local Pro. -0.000559 (-0.0586) 0.000559 (-0.0655) W*Rest 0.005879 (-0.3151) -0.005879 (-0.3521) W*Y -0.236038* (-2.2300) -0.236068* (-2.2322)
34
Table A.4 direct and indirect effects
t-value in parentheses * significant at 1%.
