induce greater efforts at cross-district coordination (and hence lower

in¬‚ation scores).

The results were also supportive of the joint centralization

hypothesis. Horizontal and vertical centralization together determine

the size of the aggregation payoff. Aggregation is best where power is

concentrated both horizontally and vertically. The interaction models

suggested that the marginal effect of vertical centralization on in¬‚ation

25

In model 4, a rise in decentralization from 0 to 1 yields a change in the predicted in¬‚ation

of .07 (from .06 to .20). (All other variables are held at their mean “ see Table 3.9.)

Testing the Theory 65

is conditional on the level of horizontal centralization. Vertical

centralization is indeed associated with a lower in¬‚ation score, but only

where there is also substantial horizontal centralization. In short, dif-

fusion of power within the national government is suf¬cient

to undermine the effect of vertical centralization on in¬‚ation, as

hypothesized. Finally, the proxy for the size of the aggregation payoff,

the decentralization index, is positively related to party system in¬‚a-

tion, as expected.

3.4 the probability of capturing the

prize “ parliamentary systems

The size of the payoff to being the largest party in government is not the

only thing that shapes aggregation incentives. The probability that the

largest party will capture that prize is also important. If the largest

legislative party has little chance of capturing the reins of government,

particularly executive of¬ce, then even a potentially large payoff will

not be enough to induce national coordination. In parliamentary sys-

tems, the probability of capturing the prize depends on whether the

leader of the largest party becomes the prime minister. If someone other

than the leader of the largest party gets the chance to form and head the

government, then this can undermine the incentives to coordinate

across districts to forge a large national party.

Prime Minister Hypothesis: In parliamentary systems, in¬‚ation will

be negatively related to the probability that the leader of the

largest party becomes prime minister.

I operationalize the probability variable as a moving average of the

number of elections since 1970 where the prime minister has not been

a member of the largest party in the legislature (probability). The

variable is continuous and ranges from 0 (the leader of the largest

party has always been the prime minister) to 1 (the leader of the

largest party has never been the prime minister). In short, probability

captures the probability that the prime minister will n o t come from

the largest legislative party. Higher values on probability should be

associated with a higher in¬‚ation score. I believe probability is a rea-

sonable proxy for the expectation candidates and party leaders might

have when devising a coordination strategy. However, as a robustness

Building Party Systems in Developing Democracies

66

check I also calculated two dichotomous variables. Probability2 is

coded 1 once someone other than the leader of the largest party serves

as prime minister and probability3 is 1 if someone other than the

leader of the largest party served as prime minister in the last election.

These two cruder proxies produce results consistent with probability,

though they are occasionally insigni¬cant. All three variables are

calculated from data in the Database of Political Institutions (Beck

et al. 2001).

Probability-Payoff Interaction

Recall that it is the interaction of the size of the payoff and the prob-

ability of capturing that prize that together jointly shape the expected

utility of becoming the largest legislative party. In short, probability

modi¬es the effect of decentralization on aggregation incentives.

Expected Utility Hypothesis: The effect of the size of the payoff on

in¬‚ation is conditional on the probability of capturing that pay-

off. Speci¬cally, the marginal effect of decentralization on in¬‚a-

tion is increasing in probability.

I test these hypotheses using a dataset of 156 elections in 21

parliamentary democracies from 1970 to 2002. I use both cross-

sectional and pooled models. Given the structure of the data I again use

OLS with robust standard errors clustered by country for the pooled

analysis.

3.4.1. Results

Table 3.5 displays the results of the hypothesis testing “ again divided

into cross-sectional and pooled analyses. Given the small number of

observations in the cross-sectional models, the results should be

interpreted with some caution. It is heartening, therefore, to see the

cross-sectional results replicated in the pooled analysis. Columns 1 and

4 contain the results of a simple additive model. We can see that

probability is statistically signi¬cant with the right sign when added to

the model in this way. The higher is the probability that the leader of

the largest party will n o t capture the payoff, the higher is the asso-

ciated in¬‚ation score. This is true, even controlling for the size of the

table 3.5. Aggregation in Parliamentary Democracies (Dependent Variable: In¬‚ation)

Cross-Sectional Analyses Pooled Analyses

Explanatory

Variables 1 2 3 (No Thailand) 4 5 6 (No Thailand)

0.07** 0.04 0.05 0.08** 0.05 0.05*

decentralization

(0.03) (0.05) (0.05) (0.03) (0.03) (0.03)

0.23** 0.16*

probability À0.14 À0.08 À0.12 À0.03

(0.10) (0.32) (0.30) (0.09) (0.18) (0.17)

0.21 0.10 0.18 0.07

decent*prob

(0.17) (0.17) (0.13) (0.12)

0.36*** 0.32** 0.30** 0.53*** 0.51*** 0.48**

ef

(0.11) (0.11) (0.12) (0.16) (0.16) (0.17)

log_govage À0.07** À0.07*** À0.07*** À0.03** À0.02** À0.02**

(0.02) (0.02) (0.02) (0.01) (0.01) (0.01)

0.01 0.02 0.02 0.01 0.01 0.01

log_districts

(0.02) (0.02) (0.02) (0.02) (0.03) (0.03)

0.11 0.18 0.18

Constant À0.05 À0.01 À0.01

(0.13) (0.14) (0.14) (0.10) (0.10) (0.10)

0.74

R-squared .74 .65 .66 .67 .64

19 19 18 101 101 98

Observations

* p < .1, ** p < .05, *** p < .01; standard errors in parentheses

67

Building Party Systems in Developing Democracies

68

payoff (decentralization). Note too that the effect of decentralization

on in¬‚ation remains signi¬cant in these models.

Columns 2 and 5 summarize the results of the interactive models.

To get a better sense of the interactive dynamics at work in these

models, I once again include marginal effects graphs. In this case, the

graphs display the marginal effect of decentralization as probability

changes. Figures 3.2a and 3.2b show that the marginal effect of de-

centralization on in¬‚ation is always positive, though when proba-

bility is at or near 0 this positive effect is not statistically signi¬cant at

the 95% level (though it is at the 90% level).26 This suggests that

the marginal effect of the size of the payoff is weakest where actors

are virtually certain that the largest party will capture the

premiership. However, as the probability that someone other than the

leader of the largest party will become prime minister increases, the

marginal impact of decentralization in¬‚ation increases as well. As

hypothesized, the marginal effect of decentralization increases in

probability. As I argued in Chapter 2, aggregation appears to be the

poorest where both the size of payoff is small (decentralization is

high) and the chance of capturing that payoff is low (probability is

high). More concretely, the model in column 5 suggests that where

power is highly decentralized (decentralization ¼ 2) and there is little

chance of capturing the payoff (probability ¼ 1), the in¬‚ation score is

a very high .42.27 By contrast, where there is certainly that the largest

legislative party will capture a valuable prize (probability and

decentralization are both 0), the predicted in¬‚ation score is .08 (see

Table 3.9).

Plots of the observations suggest that Thailand is an outlier with both

an unusually high probability score and an equally high in¬‚ation score.

Even though this result is consistent with the theory (as I will demonstrate

in Chapter 5), I want to be certain the results are not being driven entirely

by the Thai case. As a remedy, I reran the interactive models with

Thailand excluded and display the results in columns 2 and 4. We can see

that dropping the Thai case does indeed reduce the size of the coef¬cients

on both probability and the interaction term. The marginal effects graphs

26

In the pooled analyses, substituting expenditures for revenues in the decentralization

index produces consistent but weaker results. Substituting expenditures makes no

difference in the cross-sectional analyses.

27

Holding all other variables at their means.

Testing the Theory 69

95% Confidence Interval

Marginal Effect of Decentralization on

Marginal Effect of Decentralization on (b)

(a)

Inflation as Probability Changes (Pooled)

Inflation as Probability Changes (XS)

Marginal Effect of Decentralization

Marginal Effect of Decentralization

.6 .5

.4

.4

.3

.2

.2

0 .1

0

“.2

0 .5 1 0 .5 1

Probability of NOT Capturing Payoff

Probability of NOT Capturing Payoff

(c) (d)

Marginal Effect of Decentralization on Marginal Effect of Decentralization on

Inflation as Probability Changes““No Thailand (XS) Inflation as Probability Changes““No Thailand (Pooled)

Marginal Effect of Decentralization

Marginal Effect of Decentralization

.4 .3

.3

.2

.2

.1

.1

0

0

“.1

“0.1

0 .5 1 0 .5 1

.5

Probability of NOT Capturing Payoff Probability of NOT Capturing Payoff

(e) Marginal Effect of Probability on Marginal Effect of Probability on Inflation

(f)

Inflation as Decentralization Changes (XS) as Decentralization Changes (Pooled)

Probability of NOT Capturing Payoff

Probability of NOT Capturing Payoff

.5

1

.5

0 0

“.5

“.5

“.1

0 1 2 3 0 .5 1 2

1.5

Decentralization Decentralization