3.1. Preliminary evidence: Fixed-effects regressions
To estimate the relationship between political support and public investment, we begin by estimating a prefecture-level fixed-effects model as follows:
$${public investment}_{it}={{\alpha }_{1}political support}_{it}+\beta {Z}_{i}·{\phi }_{y}+\beta {X}_{i}·{\phi }_{y}+{\theta }_{i}+{\theta }_{t}+{\epsilon }_{it}$$
1
where \({public investment}_{it}\) denotes the natural logarithm of real per capita total public investment allocated in prefecture i at time t; \({political support}_{it}\) is measured by the variables incumbent share, and victory margin in prefecture i in the last election as described in Section 2.1; \({Z}_{i}\) includes a set of geographic prefecture characteristics; \({X}_{i}\) denotes a set of predetermined prefecture characteristics measured before the beginning of our sample in order to reduce endogeneity concerns (see, e.g., Bahar et al., 2021). Interactions of fixed and predetermined controls and year dummies (φy) are included in all our estimates to flexibly account for potential differential non-parametric trends on a number of prefecture characteristics. In particular, the variables included in \({Z}_{i}\) are: (i) area pc; (ii) altitude; and (iii) capital. Moreover, following studies of the relevant literature (see, e.g., Joanis, 2011; Curto-Grau et al., 2012; Solé-Ollé, 2013), the variables included in \({X}_{i}\) are: (i) income pc; (ii) population; (iii) unemployed; (iv) illiterates (v) agricultural share; (vi) industrial share; and (vii) construction share. The variable capital is based on our own calculations; the variable income pc is measured in 1974; whereas the rest are obtained from the census of 1971. Explicit definitions, descriptive statistics and sources of the variables employed throughout the analysis are provided in Table B2 in the Appendix. The model also includes prefecture, \({\theta }_{i}\), and year fixed-effects, \({\vartheta }_{t}\), to control for time-invariant prefecture characteristics and shocks common to all prefectures. Finally, \({\epsilon }_{it}\) is the error term clustered at the prefecture i level.
To assess if regions with stronger support for the incumbent party receive more investment flows consistent with the core voter hypothesis, the coefficient on incumbent share must have a positive sign. Additional evidence to this direction can be provided if the variable victory margin returns a positive and statistically significant coefficient. Alternatively, we introduce a squared term of the variable victory margin to examine whether Greek governments opted to divert projects towards prefectures with weak support (i.e., swing prefectures) if they believe that their core supporters are going to vote for them unconditionally. If the swing voter argument can explain our results, a positive coefficient on victory margin and a negative coefficient on its squared term are to be expected (see Jablonski, 2014; Tribin, 2020).
The estimates of Eq. (1) are presented in Table 1. Odd-numbered columns report the results with prefecture, year fixed-effects and geographic controls; whereas even-numbered columns add the predetermined prefectural characteristics. As it can be seen, our empirical results are consistent with the core voter hypothesis (also see, Rodriguez-Pose et al., 2016). In particular, the coefficient of the variable incumbent share in columns (1) and (2) is positive and statistically significant at the 1 percent level. Consistent to this, in columns (3) and (4) the coefficient of the variable victory margin is positive and statistically significant at the 5 percent level. Finally, in columns (5) and (6) the coefficient on the polynomial term of victory margin is positive (and marginally insignificant), not negative and statistically significant as we would expect according to the swing voter hypothesis. Taking into account the direction of these results, in the analysis that follows we retain the political variable incumbent share.
Table 1
Political support and the allocation of public investment
| (1) | (2) | (3) | (4) | (5) | (6) |
---|
incumbent share | 0.882*** | 0.976*** | | | | |
| (0.312) | (0.362) | | | | |
victory margin | | | 0.408** | 0.464** | 0.208 | 0.192 |
| | | (0.181) | (0.192) | (0.143) | (0.191) |
victory margin2 | | | | | 2.177 | 2.738 |
| | | | | (1.777) | (2.427) |
Observations | 780 | 780 | 780 | 780 | 780 | 780 |
R2 | 0.532 | 0.655 | 0.528 | 0.651 | 0.532 | 0.655 |
Prefecture FE | √ | √ | √ | √ | √ | √ |
Year FE | √ | √ | √ | √ | √ | √ |
Geographic controls | √ | √ | √ | √ | √ | √ |
Additional controls | | √ | | √ | | √ |
Notes: The table reports OLS estimates of Eq. (1). The dependent variable is measured as natural logarithm of real per capita total public investment. Incumbent share is the share of votes received by the incumbent party in the elections of 1974,1977, 1981 and 1985, whereas victory margin is the difference between the incumbent share and the share of votes received by the opposition. Geographic controls include the interaction of year dummies and (i) area pc, (ii) altitude and (iii) capital. Additional controls include the interaction of year dummies and (i) income pc, (ii) population, (iii) unemployed, (iv) illiterates, (v) agricultural share, (vi) industrial share and (vii) construction share. Income pc is measured in 1974, whereas the rest variables are obtained from the census of 1971. Robust standard errors, clustered by prefecture, are reported in parentheses. *, **, *** denote statistical significance at the 10%, 5%, 1% level respectively. |
3.2. The IV approach
The literature carefully considers endogeneity concerns suggesting that the relationship between public investment flows and electoral power may seem to exist due to confounding factors that remain unobserved, or because electoral strength is unavoidably endogenous to political “pork” (see, e.g., Matakos and Xefteris, 2016). To address such endogeneity concerns, our main empirical specification consists of an IV approach. More precisely, we estimate 2SLS models employing the political support of ancestor parties before the coup of 1967 as an instrument for the post-dictatorial political support. IV estimates can be thought of as ratios of the corresponding reduced-form and first-stage estimates. In our context, reduced-form estimates indicate the effect of political ancestors’ electoral strength on the spatial allocation of public investment, and first-stage estimates indicate the relationship between past electoral strength (i.e., electoral strength of the political ancestors) and contemporary electoral strength (i.e., electoral strength of the political descendants). All else being equal, non-null IV estimates are more plausible if they are part of a seamless set of results: results which show that the political ancestors’ electoral strength affects the allocation of public investment (reduced form), past electoral strength and the contemporary electoral strength are strongly associated, and that both push the allocation of public investment in the same direction. Equations (2) to (4) provide the reduced form, first-stage and second-stage specifications:
Reduced form:
$${public investment}_{it}={{\alpha }_{2}PA incumbent share}_{it}+\beta {Z}_{i}·{\phi }_{y}+\beta {X}_{i}·{\phi }_{y}+{\theta }_{i}+{\theta }_{t}+{\epsilon }_{it}$$
2
First-stage:
$${incumbent share}_{it}={{\alpha }_{3}PA incumbent share}_{it}+\beta {Z}_{i}·{\phi }_{y}+\beta {X}_{i}·{\phi }_{y}+{\theta }_{i}+{\theta }_{t}+{\epsilon }_{it}$$
3
Second-stage:
$${public investment}_{it}={{\alpha }_{4}\widehat{incumbent share}}_{it}+\beta {Z}_{i}·{\phi }_{y}+\beta {X}_{i}·{\phi }_{y}+{\theta }_{i}+{\theta }_{t}+{\epsilon }_{it}$$
4
The variable \({PA incumbent share}_{it}\) stands for the incumbent share of Political Ancestors. It is the share of votes received by the incumbent parties in the elections of 1961 and 1964. In particular, the incumbent share of ERE in 1961 corresponds to the incumbent share of its political descendant of ND in 1974 and 1977; whereas the incumbent share of EK in 1964 corresponds to the incumbent share of its political descendant of PASOK in 1981 and 1985. The rest of the variables included in the above specifications are identical to the preceding specification and contain the same controls, prefecture and year fixed-effects.
3.2.1. The political ancestors as an instrument
To provide some evidence regarding long-run partisan loyalty in Greece, Fig. 4 plots the incumbent share of ERE (EK) after its last electoral victory in the pre-dictatorial period in 1961 (1964) against the victory margin of ND (PASOK) in its first post-dictatorial victory in 1974 (1981). The correlation of electoral influence between the two right-wing parties in the upper part of Fig. 4 is 74 percent, leaving no doubt that ND is the political successor of the pre-junta party of ERE. One could argue though that the observed persistence in regional voting patterns is just an indication that voters in some regions prefer the political agenda put forth by one or the other party. Thus, it is important to highlight that this does not seem to be the case for the Greek parties. In the case of ND, the political agenda (i.e., the pre-electoral manifesto) in 1974 was fundamentally different from that put forth by ERE in the pre-dictatorial era across several central political aspects. In particular, in an attempt to attract voters from the pre-junta centrist parties, ND transformed fundamentally changing its stance concerning the institution of monarchy (and the referendum that led to the abolition of monarchy), the issue of language (where the vernacular demotic form was officially adopted), the decision to withdraw from NATO’s military wing and an economic policy that were characterized by extensive programs of public ownership (see, e.g., Featherstone and Katsoudas, 1985).
Regarding the relationship between PASOK and EK, a number of scholars have placed the political origins of PASOK firmly in the traditional centre (see, e.g., Mavrogordatos, 1983b); a view that is also in line with more recent studies suggesting that the political power of PASOK was basically based on the pre-junta interpersonal patronage networks of the EK (see, e.g., Pappas, 2009a; 2009b). However, it is important to note that the socialist party of PASOK cannot be considered as an ideological descendant of the pre-junta centrist party mostly because of its more radical political agenda and the extensive renewal of ideas and practices it brought to the Greek political arena (see Elephantis, 1981; Lyrintzis, 1984). Specifically, PASOK called itself a socialistic movement based on Marxist methodology and theory (although it resolutely rejected Leninist principles and the bureaucratic state socialism of the Eastern bloc). According to PASOK, Greece was economically underdeveloped and politically subordinate because it belonged to the ‘capitalist periphery’ and it suffered from its foundation, exploitation and domination from the imperialist countries that were its ‘protectors’ (see, e.g., Elephantis, 1981). Although, this radical political agenda was fundamentally different from that which was put forth by the centrist party of EK, the majority of the centrist voters turned to PASOK in the election of 1981 (see Mavrogordatos, 1984; Nicolacopoulos, 2005). Thus, the observed correlation of 72 percent between the victory margins of the centrist pre-dictatorship party of EK and the post-dictatorship party of PASOK clearly reveals a strong persistence on voting patterns that surely are not due to similarity in the political manifestos between the pre-junta and the post-junta parties.
Our aim is to use the electoral strength of political ancestors as instrument. To do so, it must satisfy both the exclusion restriction and the relevance requirements. We provided initial evidence of instrumental relevance in Fig. 4 showing graphically the strong positive correlation in the electoral strength of political ancestors and descendants. Moreover, in the regression tables below, the Kleibergen Paap F-statistic of excluded instrument for each regression it is consistently much greater than 10. Next, the exclusion restriction requires that past electoral strength operates solely through the contemporary electoral strength channel on public investments’ flows. This assumption cannot be tested directly and there is always a concern that the electoral strength of the political ancestors may be correlated with factors other than contemporary electoral strength which could be unobserved. Considering that both ND and PASOK were not ideological descendants of the pre-junta parties -since their political agendas were fundamentally different from that of their ancestors- one could argue that the observed persistence in voting preferences is due to loyalty. To further support this view and mitigate concerns about potential factors that may undermine the exclusion restriction, in Table B3 in the Appendix we investigate whether the electoral strength of the two ancestor parties (i.e., ERE and EK) are correlated with our set of confounding factors. Empirical results reported in Panels A and B show that past electoral strength is not correlated with any of our controls. This finding strengthens our priors that employed instruments operate solely through a contemporary electoral strength mechanism on the geographical allocation of public investment.
3.2.2. Baseline Results
The IV estimates examining the core voter hypothesis on the allocation of public investment funds are reported in Table 2. The OLS result in column (1) of Table 2 corresponds to the specification with the full set of controls in column (2) of Table 1. Column (2) of Table 2 reports the reduced form estimates, column (3) provides the first-stage estimates and column (4) reports the IV estimates. As can be seen in column (2), the electoral strength of political ancestor parties is positively related to the allocation of public investment funds at the 5 percent level of significance. These estimates highlight the long-term influence of political preferences on public policy (see also Joanis, 2011). Moreover, consistent with the exploratory evidence in Fig. 4, the first-stage estimates, reported in column (3), reveal significant persistence in political preferences. The value of the first-stage Kleibergen Paap F-statistic for the exclusion of instrument is reported in column (4) and is equal to 51, confirming that past political preferences do matter for the observed variation in the post-1974 period.
Both the reduced form and first-stage estimates are promising conditions for the IV analysis of the effect of political support. As can be seen in column (4), the coefficient of the latter is positive and statistically significant at the 1 percent level. Our results reveal a larger magnitude of the IV relative to the OLS estimates in column (1). This could indicate that omitted variable bias weight heavier than the impact from reverse causality. Yet, an alternative explanation is that the instrument picks up a Local Average Treatment Effect (LATE). While we cannot provide direct evidence against a LATE interpretation, a negative bias is more consistent with the fact that OLS estimates increase in magnitude when a large set of controls is included (see columns (1) and (2) of Table 1). Qualitatively, IV estimates suggest that a one SD increase in the incumbent share (or 7.7 percent) leads to a 11.5 percentage points increase in the allocation of public funds.
Table 2
Political support and the allocation of public investment: IV estimates
|
(1)
|
(2)
|
(3)
|
(4)
|
|
OLS
|
Reduced form
|
First-stage
|
IV
|
incumbent share
|
0.976***
|
|
|
1.219**
|
|
(0.362)
|
|
|
(0.507)
|
PA incumbent share
|
|
0.627**
|
0.515***
|
|
|
|
(0.265)
|
(0.072)
|
|
Observations
|
780
|
780
|
780
|
780
|
R2
|
0.655
|
0.653
|
0.742
|
0.654
|
Kleibergen Paap F-statistic
|
|
|
|
51.751
|
Prefecture FE
|
√
|
√
|
√
|
√
|
Year FE
|
√
|
√
|
√
|
√
|
Geographic controls
|
√
|
√
|
√
|
√
|
Additional controls
|
√
|
√
|
√
|
√
|
Notes: Columns (1), (2), (3) and (4) report estimates of Equations (1), (2), (3) and (4) respectively. The dependent variable is measured as natural logarithm of real per capita total public investment. Incumbent share is the share of votes received by the incumbent party in the elections of 1974, 1977, 1981 and 1985. PA incumbent share is the share of votes received by the incumbent party in the elections of 1961 and 1964. The incumbent share of ERE in 1961 corresponds to the incumbent share of ND in 1974 and 1977. The incumbent share of EK in 1964 corresponds to the incumbent share of PASOK in 1981 and 1985. Geographic controls include the interaction of year dummies and (i) area pc, (ii) altitude and (iii) capital. Additional controls include the interaction of year dummies and (i) income pc, (ii) population, (iii) unemployed, (iv) illiterates, (v) agricultural share, (vi) industrial share and (vii) construction share. Income pc is measured in 1974, whereas the rest variables are obtained from the census of 1971. Robust standard errors, clustered by prefecture, are reported in parentheses. *, **, *** denote statistical significance at the 10%, 5%, 1% level respectively. |
3.2.3. Robustness checks
Our first robustness check is to experiment with an alternative political support measure to add validity in our argument that our findings are explained by the core voter hypothesis. In particular, we replace the continuous variable incumbent share with a dichotomous classification that allows us to identify the most traditional political strongholds of the two parties (see, e.g., Tribin, 2020). The variable stronghold takes the value 1 if ND and PASOK have won at the prefecture level for all four elections from 1974 to 1985. Alternatively, the less restrictive variable stronghold 2 takes the value 1 in prefectures that ND and PASOK won at least three out of four elections from 1974 to 1985. Our reasoning for the latter variable is that at least one of these victories comes in an election that the party lost at the national level. Panels A and B of Figures B2 in the Appendix provide exploratory evidence illustrating that areas in northern Greece (like Serres, the place of origin of Konstantinos Karamanlis) and south Peloponnese voted strongly in favor of ND, whereas prefectures in the Crete island and south Peloponnese (like Achaia, the place of origin of Andreas Papandreou) are political strongholds of PASOK.
The empirical results using the alternative variables are presented in Table 3. As can be seen in columns (1) and (4), the coefficients of OLS estimates for the variables stronghold and stronghold 2 are positive but marginally insignificant at the 10 percent level. However, they turn positive and statistically significant at the 5 percent level in columns (3) and (6) which report the IV estimates. As before, the instrument employed is the voting share of political ancestor parties. According to the results provided in columns (2) and (5), the latter is strongly correlated with the prefectures identified as strongholds for the two parties with a Kleibergen Paap F-statistic always very large. The results suggest that the stronger political bailiwicks identified by the variable stronghold receive approximately 35 percent more public investment. This percentage drops to 23.5 percent when we use the more relaxed definition of the variable stronghold 2. Overall, the empirical findings from the two alternative variables employed, as well as the increase in the magnitude for the strict definition of strongholds, are consistent with the core voter hypothesis.
Table 3
Political strongholds and the allocation of public investment: IV estimates
|
(1)
|
(2)
|
(3)
|
(4)
|
(5)
|
(6)
|
|
OLS
|
First-stage
|
IV
|
OLS
|
First-stage
|
IV
|
stronghold
|
0.149
|
|
0.299**
|
|
|
|
|
(0.095)
|
|
(0.122)
|
|
|
|
PA incumbent share
|
|
0.469***
|
|
|
0.515***
|
|
|
|
(0.068)
|
|
|
(0.072)
|
|
|
|
|
|
0.136
|
|
0.211**
|
stronghold 2
|
|
|
|
(0.082)
|
|
(0.082)
|
Observations
|
780
|
780
|
780
|
780
|
780
|
780
|
R2
|
0.645
|
0.750
|
0.640
|
0.651
|
0.742
|
0.648
|
Kleibergen Paap F-statistic
|
|
|
73.071
|
|
|
64.500
|
Prefecture FE
|
√
|
√
|
√
|
√
|
√
|
√
|
Year FE
|
√
|
√
|
√
|
√
|
√
|
√
|
Geographic controls
|
√
|
√
|
√
|
√
|
√
|
√
|
Additional controls
|
√
|
√
|
√
|
√
|
√
|
√
|
Notes: Columns (1), (2) and (3) report estimates of Equations (1), (3) and (4) respectively. Columns (4)-(6) follow the same structure. The dependent variable is measured as natural logarithm of real per capita total public investment. The variable stronghold in columns (1)-(3) takes the value of 1 if ND and PASOK have won at the prefecture level the elections of 1974,1977, 1981 and 1985. The variable stronghold 2 in columns (4)-(6) takes the value of 1 if ND and PASOK have won at the prefecture level at least three of the four elections of 1974,1977, 1981 and 1985. The rest information is similar as in Table 2. Robust standard errors, clustered by prefecture, are reported in parentheses. *, **, *** denote statistical significance at the 10%, 5%, 1% level respectively. |
Our second robustness check is to employ an alternative measure of public investments flows. More precisely, given that public investment is measured at the regional level on a yearly basis it can be very volatile over time. To this end, in Table B4 in the Appendix we aggregate our variables per term-of-office. Once again, results are very similar to those obtained in Table 2. Finally, our third robustness check is to remove from our in-sample the period 1975 to 1977. The reason is that these are the first years after the transition to democracy combined with a call by the prime minister for an early election in 1977. It would be problematic if our findings are mainly driven by this very turbulent period of Greek politics. As can be seen in Table B5 in the Appendix, our results remain intact even by omitting the first period in our estimations.
3.2.4. Placing the spotlight on the mechanism: The key role of powerful MPs
In an effort to explain why governments prioritize swing or core/loyal constituencies, a strand of the literature places the spotlight on political institutions and, especially, on the electoral system (see, e.g., McGillivray, 2004; Kemmerling and Stephan, 2015). A major characteristic of the political system that is expected to affect the allocation of public investment is the degree of personalism in the electoral contests (see, e.g., McGillivray, 2004; Golden and Picci, 2008) and, consequently, the relative strength of the party vis-à-vis local legislators (i.e., MPs).
Practically, party strength can be thought as the degree of central party control over candidate selection. McGillivray (2004) contends that in the context of a majoritarian voting system, weak parties are not able to effectively discipline their MPs. Therefore, MPs with greater seniority or greater influence within the party direct resources to their home constituencies even when the latter appears to be safe and not marginal. Similarly, Golden and Picci (2008) suggest that also under an OLPR system, powerful MPs typically favor their home constituencies. This is due to the fact that OLPR induces intra-party competition -as candidates compete over their co-partisans in order to be elected- and this may lead candidates to cultivate their own personal reputation among the voters (see, e.g., Carey and Shugart, 1995) as well as to develop clientelistic linkages between them and the electorate (see, e.g., Ames, 1995; Kitschelt, 2000).
The political landscape in Greece is characterized by a substantial degree of electoral personalism (see, e.g., Mavrogordatos, 1983a). This is due to the following two reasons: (i) the electoral system which was a standard OLPR encouraged this environment, and (ii) also because Greece during the period under investigation was a ‘new democracy’ characterized by infant and relatively weak political parties that were drawing electoral support from powerful MPs and their personalistic patron-client networks. To investigate this possibility, we employ data from Tziovaras and Chiotis (2006) and construct the variables powerful MPs, ministers and deputy ministers.
The variable minister (deputy minister) equals to 1 when an appointed minister (deputy minister) of the government has been elected in prefecture i and year t, and 0 otherwise. Moreover, the variable powerful MPs equals to 1 when an appointed minister (or deputy minister) of the government has been elected in prefecture i and year t, and 0 otherwise. The idea behind this is that we expect popular MPs of both parties to be assigned in ministerial positions, whereas, at the same time, these positions offer them discretion over the allocation of investment funds to benefit their home constituencies in an attempt to get re-elected. It should be noted that simple fixed-effect regressions, reported in Table B6 in the Appendix, indicate a positive relationship between electoral strength at the prefecture level and the probability to have a powerful MP as minister or deputy minister. Interestingly, this relationship seems to be driven mostly by deputy ministers rather than ministers.
Table 4 presents the results that test the mediating role of powerful MPs on the geographical allocation of public investment funds. For brevity, we present only the OLS and the IV results -odd and even columns, respectively- for specifications employing the three variables described above. As can be seen in column (2), the interaction term between the incumbent share and powerful MPs of the IV estimates is positive and statistically significant at the 10 percent level. This is consistent with our expectations that loyal prefectures receive more public investment when they have powerful MPs appointed as ministers or deputy ministers, since the latter attempt to support their patron-client linkages in order to increase their re-election prospects in an OLPR system. In columns (3)-(4) and (5)-(6) we split powerful MPs to ministers and deputy ministers, respectively. Interestingly, the previously discussed relationship seems to be driven by the deputy ministers. A possible explanation is that high-profile elected ministers are not in need to use their discretion over public investment funds as their position is anyway secure. In contrast, deputy ministers, who are lower in the ranks, attempt to support their candidacy against competitors from the same party by allocating more public investment funds within their prefecture; in turn parties profit from their electoral influence.
Table 4
Political support and the allocation of public investment: The role of powerful MPs
|
(1)
|
(2)
|
(3)
|
(4)
|
(5)
|
(6)
|
|
OLS
|
IV
|
OLS
|
IV
|
OLS
|
IV
|
MPs variable
|
powerful MPs
|
ministers
|
deputy ministers
|
incumbent share
|
0.760**
|
0.705*
|
1.078***
|
1.272**
|
0.655**
|
0.636
|
|
(0.339)
|
(0.428)
|
(0.392)
|
(0.531)
|
(0.322)
|
(0.439)
|
MPs variable
|
-0.335
|
-0.581*
|
0.093
|
-0.041
|
-0.563*
|
-1.054**
|
|
(0.236)
|
(0.327)
|
(0.250)
|
(0.201)
|
(0.309)
|
(0.525)
|
incumbent share · MPs variable
|
0.742
|
1.386*
|
-0.539
|
-0.179
|
1.454*
|
2.706**
|
|
(0.581)
|
(0.835)
|
(0.642)
|
(0.508)
|
(0.762)
|
(1.341)
|
Observations
|
780
|
780
|
780
|
780
|
780
|
780
|
R2
|
0.659
|
0.657
|
0.660
|
0.658
|
0.661
|
0.655
|
Kleibergen Paap F-statistic
|
|
28.784
|
|
25.711
|
|
14.060
|
Prefecture FE
|
√
|
√
|
√
|
√
|
√
|
√
|
Year FE
|
√
|
√
|
√
|
√
|
√
|
√
|
Geographic controls
|
√
|
√
|
√
|
√
|
√
|
√
|
Additional controls
|
√
|
√
|
√
|
√
|
√
|
√
|
Notes: Columns (1) and (2) report estimates of Equations (1) and (4) respectively, augmented with the variable powerful MPs and its interaction with the incumbent share. Columns (3)-(4) and (5)-(6) follow the same structure for the variables ministers and deputy ministers, respectively. The dependent variable is measured as natural logarithm of real per capita total public investment. The variable powerful MPs equals to 1 when an appointed minister (or deputy minister) of the government has been elected in the prefecture, and 0 otherwise. The variable ministers (deputy ministers) equals to 1 when an appointed minister (deputy minister) of the government has been elected in the prefecture, and 0 otherwise. Incumbent share is the share of votes received by the incumbent party in the elections of 1974, 1977, 1981 and 1985. The rest information is similar as in Table 2. Robust standard errors, clustered by prefecture, are reported in parentheses. *, **, *** denote statistical significance at the 10%, 5%, 1% level respectively. |
[13] It is important to note that the elections of 1961 and 1964 were the last elections that took place before the military junta (1967-1974). In the elections of 1961, ERE was the first party (and formed a government) whereas in the elections of 1964, EK was the first party (and formed a government).
[14] During that period, the Hellenic Federation of Enterprises (SEV) coined the term “social-mania”, accusing the prime minister Konstantinos Karamanlis of implementing a radical left-wing economic policy. For more details on this episode, see To Vima (6 March 1976).
[15] Building on the seminal paper by Altonji et al. (2005), Oster (2019) assesses how large the bias due to unobservables should be, in comparison to that of observables, in order to explain away the estimated effect. The ratio between the two components of the bias is denoted as 𝛿. This exercise returns a ratio that is (above 1 and) negative, which indicates a negative correlation between observables and unobservables and implies that the coefficient increases when controls are added to the model. It also indicates that for the results to become economically insignificant, the selection on unobservable factors would have to dominate the selection on the included observables and work in the opposite direction.
[16] It should be noted that according to both definitions, the strongholds include prefectures in the top quartile of the distribution of electoral strength.
[17] The reduced form estimates correspond to those already reported in column (2) of Table 2.
[18] The same applies when we drop from the sample all subsequent terms in office (1978-1981, 1982-1985 and 1986-1989) by the governments that followed. Results are available upon request.
[19] Building on the pioneer studies of Lizzeri and Persico (2001) and Milesi-Ferretti et al. (2002), this literature argues that under majoritarian (or plurality) voting systems, politicians prefer to distribute benefits through geographically targetable public goods directed to marginal electoral districts. This is because legislative seats require plurality of votes in each electoral district and, therefore, votes in contested districts matter more to politicians than votes in safe districts. As a result, politicians face incentives to behave along the predictions of the ‘swing voter’ hypothesis. In contrast, proportional voting systems are less vulnerable to regional pork-barrel strategies (see, e.g., Persson and Tabellini, 2002).
[20] More precisely, these patron-client linkages form pyramids with powerful MPs at the apex, local party bosses in the middle and individual voters at the base (see, e.g., Mavrogordatos, 1983a). Traditionally, these networks of local bosses and middlemen belonged personally to the powerful MPs, and it was a common practice to be transmitted as inheritance -or even as dowry- within the same family from one generation to the other. It is obvious that in such political context, powerful MPs were the ultimate centre of political power and, consequently, the political parties were built structurally around these networks of local notables (see Meynaud, 2002, for more details on this). The absence of effective party organization and mass membership constituted to the party’s parliamentary group being extremely powerful; this situation is often described as vouleftokratia (‘rule of the MPs’) in the relevant literature.
[21] It should be noted that our classification considers ministers that have been elected in the prefecture and also plan to be candidates in the next election.
[22] It has been argued that, since 1974, the pre-junta party system, which was based on traditional interpersonal patron-client relationships, was starting to be transformed. In particular, the new parties that emerged during Metapolitefsi were developing stronger organizational structure, mass membership and a new type of clientelistic networks described as bureaucratic clientelism (or machine politics) instead of traditional patronage (see, e.g., Lyrintzis,1984; Mavrogordatos, 1983a, 1997; Kammas et al., 2021). This transformation of the political system is a stylized fact. However, this does not mean that the interpersonal clientelistic relationships of the past disappeared. In contrast, at least during the first years of Metapolitefsi, the newly established political parties were trying to maintain the political networks of their (pre-junta) political ancestors that were structured around powerful MPs and, at the same time, to develop a strong organizational structure (see Kammas et al., 2021, for more details on this).