Tsebelis G. Veto Players: How Political Institutions Work

Подождите немного. Документ загружается.

251

values of all available indices. To accomplish this, I standardized each one of the indices and

then took the average of the standardized scores that were available for each case government.

For standardization, I used only the values of the variables for the countries covered by all three

indices. This procedure was run separately on all three range and alternation variables, resulting

in three standardized range and three standardized alternation variables. The average range and

alternation variables used all the available information in the following way: In the case that all

three indices existed, the average was calculated on the basis of all three; for countries with two

indices only the average was calculated on the basis of the two standardized indices; in the cases

covered by one single analyst (Greece), I used that one standardized score. In the regressions I

used the absolute value of alternation as calculated above, because it makes no difference

whether a left wing government is replaced by a right wing government, or vice versa.

II. VETO PLAYERS AND SIGNIFICANT LEGISLATION

In this section, I will test all the predictions made in the introduction of this chapter. I

test whether range affects negatively the number of significant laws while at the same time

producing a heteroskedastic relation. I construct the variable RANGE (average normalized

ideological distance of extreme partners of a government coalition, corrected for institutional

rules like the President of Portugal, or the Budesrat in Germany). I test whether this relationship

is both negative and heteroskedastic (see Figure 1.8). I test whether alternation and government

duration affect positively the number of significant laws by introducing a series of additional

variables: ALTERNATION (absolute value of the average normalized difference between two

successive governments), DURATION (years of a government in office), as well as others that

turn out not to be significant (as predicted).

INSERT TABLE 7.1 HERE

252

Testing for the negative effect of range and for heteroskedasticity. The expected

relationshp between RANGE and LAWS is that the first is a necessary but not sufficient condition

for the second: a large range will prevent significant legislation, but a small range will not

guarantee the existence of significant legislation. As we said in Chapter 1 the implication of this

analysis is that RANGE and LAWS will be negatively correlated, and the squared (or the absolute

value of) residuals of the estimated relationship will be also negatively correlated with RANGE.

Heteroskedasticity has a negative impact on the signifiance of statistical coefficients (since it

generates high standard errors). However, the existnace of predicted heteroskedasticity should be

in favor my theory, not against it. In other words, the appropriate test for a theory predicting a

sufficient but not necessary condition is a combination of a test of means (regression) with low

statistical significance and a test of the variance (residuals) for heteroskedasticity. If both

predictions turn out to be corroborated (as they are), then the confidence to the theory that

predicted both relationships ought to be significantly higher than the p-value of any one

coefficient.

Table 7.1 tests for these two expectations with a multiplicative heteroskedastic regression

model.

112

The following two equations are tested simultaneously by maximum likelihood

estimation:

Equation 1: LAWS = a- bRANGE +??

Equation 2: ?

???exp(p- qRANGE )

As Table 7.1, indicates both the coefficients of RANGE for equations 1 and 2 are highly

significant. However, the null hypothesis is that both coefficients are 0. For this reason I estimate

a second model omitting RANGE as an explanatory variable from both equations and perform a

112

In a previous version (Tsebelis (1999)) I had estimated three independent models: one the average number of

significant laws, one estimating the error term, and one “correcting” for heteroskedasticity. The development of

253

likelihood ratio test which provides a chi

= 17.85, which gives a p-value of .0001.

INSERT FIGURES 7.1 AND 7.2

More visual evidence is provided by Figures 7.1 and 7.2. Figure 7.1 presents the

relationship between LAWS and RANGE. In this figure I have separated minimum winning

coalition governments (indicated by o in the figure) from minority governments (? in the figure)

and from oversized coalitions (? in the figure). The reason for this distinction was explained in

Chapter 4. The political logic in all these cases indicates that the parties in government are veto

players. However, the arithmetic constraints are different between minimum winning coalitions

where the support of all parties is arithmetically necessary, minority governments where more

votes are necessary, and oversized coalitions where all the votes are not necessary. My

expectation (Chapter 4) was that the veto player argument will hold in an approximate way in the

case of minority and oversized governments. The data overall confirm my expectations. I present

four bivariate regression lines. The top line summarizes the relationship between laws and range

occurring in minimum winning coalitions. The relationship is very strong and significant. The

two intemediate lines summarize the overall relationship and the oversized coalitions. Both

relationships are statistically significant. The last line reperesnts minority governments and has a

negative but very weak and not statistically significant slope. This result indicates that knowing

the exact conditions of each one of these minority governments as Strom argues (see Chapter 3)

would be preferable.

113

Comparison of Figure 7.1 with Figure 1.8 (that I used in Chapter 1 to

indicate what to expect from empirical results) indicates the high degree of fit between theory

and data (that also appeared in the regressions of Table 7.1).

multiplicative heteroskedastic regression models permits simultaneous testing.

113

For example, it would be interesting to know whether support for government proposals was sought most of the

time among the same parties, in which case the number of veto players increases.

254

Figure 7.2 gives a graphic representation of the absolute value of the residuals from

Model 1 in Table 7.1. I selected the absolute value for this Figure because the graphic of squared

residuals is visually misleading (it eliminates small residuals and exacerbates large ones). Again,

I divided governments into minimum winning coalitions, minority governments and oversized

coalitions, but this time there is no difference between the regression lines representing the

whole dataset and each one of the two parts. We can see that the slope is negative and very

significant, exactly as expected.

INSERT TABLE 7.2

Test for alternation and government duration. The models in this section are multivariate

and introduce a series of control variables. According my analysis two additional control

variables (DURATION and ALTERNATION) are expected to have positive signs.

ALTERNATION (the difference between the midpoints of the current from the previous

government) is one way to introduce a proxy for the status quo in case legislation was introduced

by the previous government. Of course, there is no guarantee that this was actually the case (see

above).

Model 1 in Table 7.2 introduces both control variables in their linear form. Model 2

introduces the idea of a declining rate of production of significant laws by using the natural

logarithm of duration as an independent variable. This model corroborates all the expectations

generated by the veto players’ theory. This is why I subject it to three additional tests. The first

test is to examine whether the findings hold for different subsets of the data. Models 2A, and 2B

2C separate the different governments into minimum winning coalitions (23 cases), and minority

government (15 cases), and oversized coalition (21 cases) and re-test the model for each one of

these categories. The second test is to introduce a series of control variables in order to test for

255

spuriousness of the results. Model 3 introduces three plausible control variables: AGENDA

CONTROL, CORPORATISM, and LEFT IDEOLOGY of the government. The existing literatures

suggest these as alternative explanations for the findings. The last test is to rerun the model both

with and without control variables as a negative binomial regression (given of the fact that my

dependent variable is “count” the OLS coefficients may be biased).

As Table 7.2 shows all the hypothesized relationships come out with the correct sign

(negative for range, and positive for alternation and duration). On the basis of Model 2, one can

say that the production of significant laws is affected negatively by the range of government,

positively by the difference between current and previous government (alternation), and that

duration increases the number of laws but at a declining rate.

Models 2A, and 2B, 2C replicate the analysis for minimum winning coalitions, and other

governments respectively. All the signs of the coefficients are as hypothesized, but conventional

levels of statistical significance are lost, except for the case of minimum winning coalitions.

114

Let me now discuss Model 3 that introduces three different control variables. The first of

them is AGENDA CONTROL. Doering (1995b) has identified the importance of government

agenda setting for both the quantity and quality of legislation produced in a country. In a

nutshell, his argument is that government control of the agenda increases the number of

important bills and reduces legislative inflation (few small bills). Doering defined agenda control

in two different ways: qualitatively, and quantitatively.

115

He hypothesized a positive relationship

114

I thank an anonymous referee for the suggestion of subdividing the dataset. The division applied the standard

criteria in most countries. However, in Germany I had to take into account the Bundesrat if controlled by the

opposition, in Portugal the President if his party was not included in the government, and in one case of a

government decree in France I took into account the President of the Republic. In all these cases the standard status

of the government was altered to take into account the veto players’ theory: for example, in Germany a minimum

winning coalition government was coded as oversized if the support of an opposition controlled Bundesrat was

required.

115

The qualitative measure of agenda control is the first of the indicators defined by Doering (see discussion in

Chapter 3). The quantitative comes from my calculations in the same chapter and can be found in Table 4.1. In my

256

between significant laws and agenda control. However, Doering was discussing countries as

units of analysis, and his measures (which I use) refer to countries. Consequently, the variance of

significant legislation within each country cannot possibly be captured by Doering’s variables. I

will discuss his analysis more in detail in the next section.

CORPORATISM was the second variable introduced for control purposes. I introduced

CORPORATISM both as a trichotomous variable (with Belgium, Germany, Luxembourg and

Switzerland as ambiguous cases), and as a dichotomous one (with the above countries considered

corporatist). Like AGENDA CONTROL, it is considered constant by country. In this respect, I

follow most of the literature on corporatism, despite the fact that contemporary research finds

significant fluctuations in the variables that comprise the concept over time (Golden et. al.,

1999).

In corporatist countries, the argument goes, peak associations of employers and unions

negotiate the subjects covered by this article; and only if they do not agree does parliament step

in and legislate or the Government issues decrees. Because of this, corporatist countries (where

government ranges are generally high) presumably produce less significant labor legislation.

There are two problems with this argument. First, in corporatist countries legislation is produced

whether the social partners agree or not. If they agree, the parliament or the government issues

legislation or decrees confirming the agreement. If they disagree, the legislative institutions of

the country decide on the disagreement. For example, at the end of the 1980s the problem in both

Norway and Sweden was the need to cut wages to prevent unemployment from rising. In

Norway, the social partners (unions and employers) agreed to a wage freeze and asked the social

democratic minority government to put it into legislation so that it would be universally binding.

The legislation was passed by Parliament, while the independent unions (i.e., those not affiliated

analyses I used both measures with similar results.

257

with the main confederation) complained that they were being victimized, wages declined and

unemployment did not grow very much. In Sweden, at the same time, the social partners failed

to agree to control wages, so the social democratic minority government introduced legislation to

freeze wages. In the Swedish case, all of the unions protested, the proposal was defeated, the

government fell, wages continued to rise rapidly and unemployment went much higher than in

Norway. Second, if the argument were correct, one would expect less overall labor legislation in

corporatist countries, not just less significant legislation. However, corporatist countries have

more overall legislation in the area of working time and working conditions.

The third variable was the ideology of each government. Since the dependent variable is

labor legislation, one may assume that left-wing governments produce more of it. In my view,

this interpretation ignores the possibility of right-wing governments repealing labor laws, or

undoing what left-wing governments have done. LEFT IDEOLOGY was measured exactly the

same way as RANGE and ALTERNATION, so it varies by government and the empirical results

will be conclusive.

As Model 3 indicates, none of the three control variables above has any impact on the

results of Model 2. The additional three variables come out very close to zero and completely

insignificant. In addition, there is no increase in the R

of the model, and the adjusted R

shrinks.

It is safe to say that statistically these variables do not explain anything (although conceptually

one has to refer back to the discussion in the previous paragraphs to understand why it is so).

In order to make sure that these results are not generated because of peculiarities in any

one country, I examined the points of highest leverage (the four cases (three points, but one of

them is double) in the upper left quarter of Figure 7.1) in order to make sure that they do not

reflect unusual situations. These four points represent governments of Belgium, Sweden, Greece

258

and the UK. In the case of the first two countries, the governments produced an extraordinary

amount of laws because their range was unusually small. In the case of the second two countries,

the rule was single party governments, and two of them (both comprised of two or more actual

governments) produced a high number of significant laws. Even without these cases, the

negative relationship between range and significant laws is preserved, although statistical

significance is lost.

Finally, the negative binomial regression models do not alter in any way the conclusions

of the previous models. Given that the interpretation of linear coefficients is easier and more

intuitive, and given that additional control variables do not improve upon it, one should draw

conclusions from Model 2. I caution the reader that the numbers will appear “small” because I

am dealing with a single area of legislation. One would have to aggregate across different areas

to find the overall effect.

116

Given that the coefficient of the natural logarithm of duration is positive we can say that

the effect of duration on government legislation is twofold. On the one hand, duration has a

positive effect on legislation; on the other, the rate of law production declines with duration.

Let us examine the policy significance of these findings. The empirical findings taken

together indicate that large range coalitions are unlikely to produce significant legislation, while

small range coalitions and single party governments may or may not produce such laws and

decrees. In other words, policy stability is the characteristic of the first, while the possibility of

significant policy change is the characteristic of the second. So, the findings of this section are

116

In this aggregation, one would have to replicate the logic of this analysis, not extrapolate mechanically the

results. For example, the positions of government parties on environmental issues should be considered in order to

predict environmental legislation, not the left-right scale used here. So, it is perfectly reasonable to expect that a

government composed of parties close to each other on the left-right scale and produces many significant laws on

labor may produce few significant environmental laws if the veto players are far away from each other in the

environmental policy dimension. Alternatively one would have to perform a multidimensional analysis (Chapter 8).

259

that, depending on government composition (or on institutional structures that consistently

produce single or multiple veto players), one can get either policy stability or the potential for

policy change, but not both.

117

III. VETO PLAYERS AND INCREMENTAL LEGISLATION

Having established the relationship between veto players, range, and significant

legislation (lack of policy stability), we now turn to the incremental (non-significant) legislation

and the total number of laws. My expectation is: “Ceteris paribus, significant and non-

significant laws should vary inversely, because of time constraints. The ceteris paribus clause

assumes that the parliament has limited time and uses it to pass legislation (either significant or

trivial). If there are other uses of time like questions to ministers, general debates, etc., or if the

time of meetings is itself variable, controls must be introduced for these factors” (Tsebelis

1995a: 104). In this section, the unit of analysis becomes the country instead of the government,

because data at the government level are not available.

We have already discussed Doering’s analysis of government agenda setting in

parliamentary democracies in Chapter 3. Doering (1995b) has established the importance of

government agenda setting for both the quantity and quality of legislation produced in a country.

In a nutshell, his argument is that government control of the agenda increases the number of

important bills and reduces legislative inflation (few small bills).

Doering (1995 c) used actual legislative data to test his insights. The tests not only

corroborated the intuition that government agenda control reduces legislative inflation, but they

also eliminated other plausible explanations (including a country’s population size, electoral

117

Unless a single-party government finds a technology to commit credibly: by appointing an independent agency

and assigning jurisdiction, or by claiming that the status quo is its own ideal point, etc. I am not going to enter into

the discussion of commitment technologies, but the bottom line is that multiparty governments have difficulty

changing the status quo, while single-party governments do not (see discussion on taxation in Chapter 8).

260

barriers for party entry, number of parties in parliament, additional legislation at the sub national

level, etc.) along the way. So, from a different perspective, Doering has come to the same

conclusions as the veto players theory: as the number of significant laws increases, legislative

inflation goes down; but for him, it is government control of the legislative agenda that connects

the two.

In order to revisit Doering’s findings I will use the composite measure of government

control of the legislative agenda that I created in Chapter 4 out of Doering’s seven indicators.

(See Table 4.1). However, everything I say holds also if one uses Doering’s first indicator of

government agenda control, or his quantitative index.

INSERT TABLE 7.3

Table 7.3 introduces a series of variables for each country: the number of significant laws

and decrees (the sum of such legislative instruments for the whole period under examination),

the average number of laws per country (from Table 18.1 of the Doering (1995a) book), an

average number of veto players per country, as well as my index of government control

calculated on the basis of Doering’s seven indicators.

Some explanations of these data are necessary. For the number of laws per country

variable, Sweden has two numbers: one taken from Table 18.1 of Doering and the other, the

average number of new laws provided by Ingvar Mattson of the Department of Political Science

of Lund University (Doering 1995: ***). Apparently in Sweden they count each amendment as a

separate law. So the number in Doering’s Table, as he discusses in his text, although technically

correct, is inflated by comparative standards. In terms of veto players, I have used two different

variables. One provides the average of veto players for the period I examine. The other is a

qualitative measure of veto players established the following way: countries with single-party