Political business cycles

There is strong cumulative evidence that incumbents in liberal democracies benefit from favourable economic conditions (Drazen Reference Drazen2000; Hibbs Reference Hibbs2006). Building on this logic, political business cycle theory specifies how incumbents can time the use of monetary, fiscal and other policy instruments to reap electoral benefits. However, the empirical evidence on electioneering in these domains has been mixed and inconclusive, especially in recent decades. Franzese and Jusko (Reference Franzese and Jusko2006) argue that institutional constraints such as transparency and accountability requirements, and consequently better informed voters, have made it increasingly difficult for politicians in advanced democracies to manipulate monetary and budget cycles (Alt and Lassen Reference Alt and Lassen2006). But if voters evaluate incumbents on their recent performance, incumbents’ incentives to manipulate public policy increase as elections approach. Since the electoral benefits of such manipulations presumably remain large, politicians can be expected to merely shift the use of cycle mechanisms towards policy domains that are easier for them to manipulate, such as public employment.

Earlier work on Germany has primarily investigated monetary business cycles, while more recent publications focus on budget cycles. Berger and Woitek (Reference Berger and Woitek1997) find no evidence for country-level monetary cycles, though results are very sensitive to data quality and estimation methodology. Jochimsen and Nuscheler (Reference Jochimsen and Nuscheler2011) find that public deficits decrease in pre-election years. Investigating 11 West German states with respect to seven different budget categories, Galli and Rossi (Reference Galli and Rossi2002) find little evidence of cycles. Potrafke (Reference Potrafke2011) finds weak evidence that politicians increase expenditures for schooling in election years and adds that government partisan ideology has no effect on any of six expenditure categories. Oberndorfer and Steiner (Reference Oberndorfer and Steiner2007) find that conservative governments or social-democratic-conservative coalitions in West German states spend more, not less, on higher education than pure social-democratic governments.

The ambiguity of these findings might appear to confirm that German politicians have little scope to manipulate monetary policies or fiscal budget decisions. This article considers a dependent variable that differs in a fundamental way from prior research, by shifting attention away from monetary or fiscal policy manipulation towards public employment policies. The only such study on Germany to date investigates teachers and finds that against a backdrop of a continuously shrinking total teachers’ pool, states have accelerated the hiring of new teachers during election periods, only to partly reverse this later in the electoral cycle (Tepe and Vanhuysse Reference Tepe and Vanhuysse2009). We argue that the allocation of legal competences among the federal and state levels in Germany also makes police employment a fruitful case for analysing political business cycles.Footnote ² The lack of room for manoeuvre in other policy fields may turn domestic security into a particularly important arena for electoral manipulation at the state level.

The institutional context of German federalism provides state incumbents with both the “opportunity” and the “means” for manipulating police employment to gain votes (Tufte Reference Tufte1978; Franzese and Jusko Reference Franzese and Jusko2006). The “motive” to do so appears strong as well. The high salience of local-level domestic security provides some circumstantial evidence for its electoral importance in Germany (Frevel and Groß Reference Frevel and Groß2008, 85; Gerber et al. Reference Gerber, Hirternlehrer and Jackson2010). Criminologists indicate that the fear of crime affects far more citizens than actual experiences with crime (Warr Reference Warr2000; Gray et al. Reference Gray, Jackson and Farrall2008). Economists show that increases in the size of police forces substantially reduce violent crime (Levitt Reference Levitt1997) and crime generally (Levitt and Miles Reference Levitt and Miles2006; Benson Reference Benson2010). Police officers generally tend to be perceived by the public as symbolic “moral guardians” of social stability and order (Jackson and Bradford Reference Jackson and Bradford2009; Norris Reference Norris2007) and confidence in police tends to reflect public perceptions of crime and property crime rates (Sindall et al. Reference Sindall, Sturgis and Jennings2012). The quality of public goods such as inner security is hard to measure. Whereas, more sophisticated policy success indicators (e.g. crime clearance rates) require substantial information search and processing efforts, changes in police employment figures are easy for politicians to communicate and for voters to register. Voters may thus use these figures as a heuristic, or proxy, for incumbents’ policy effort.

A key distinction within German police employment is the one between civil servants (Beamter) and public employees (Angestellten). Here we focus on civil servant positions, which are numerically the most important category by far as they accounted for 84 per cent of all police officers on average in 1992–2008. We expect electioneering to be especially attractive for these jobs as they are tenured and financially most coveted. Moreover, since civil servant officers are highly street-visible “bobbies on the beat”, they are more likely to make electioneering pay off politically. Seeing regular police patrols on the street, and feeling informed about police activities, is associated with higher opinions about police effectiveness (Bradford et al. Reference Bradford, Jackson and Stanko2009). Hiring civil servant officers is therefore more likely to sway voters who care strongly about state-level crime and security at the time of elections. This is a “street-visible security provision” effect. Moreover, this effect on voters generally goes hand in hand with separate patronage motives towards police officers specifically. Above and beyond the visibility to voters of more officers in the streets, civil servant jobs are coveted benefits in and of themselves to officers. Being newly hired, or upgraded, into a tenured position is likely to influence the voting behaviour of officers. This is a straightforward “patronage” effect, and it regards relatively few voters, albeit ones that may be pivotal. Hence we hypothesise:

By way of initial exploration of H1, Table 1 compares average annual growth rates for civil servant police employment in election versus non-election years, excluding Eastern states in the immediate post-unification period 1992–1995. These descriptive statistics show very different dynamics in the two parts of Germany. In Western Germany, growth rates of officers are more positive or less negative in election years (0.69 per cent versus 0.10 per cent for civil servants). This is especially the case in Saarland and Niedersachsen, though there are deviations from this pattern in states such as Baden-Württemberg and Hessen.

Table 1

Average annual growth rates of civil servant police officers in election and non-election years, and by partisan governments

Note: Election is dummy coded and equals 1 in election years. 1993–1995 excluded for Eastern states. CDU/SPD 50plus = 50 per cent or more of the cabinet ministers belong to the CDU, respectively SPD.

Partisanship

Partisan theories in the spirit of Tufte (Reference Tufte1978) and Korpi (Reference Korpi1983) assume that politicians are not purely opportunistic vote-seekers, but decide the course of policy, at least partly, according to their ideological preferences and how policies benefit their core constituencies. Traditional partisanship theory predicts left-wing governments to pursue more expansive or Keynesian strategies when it comes to public employment and social policy, and right-wing governments to aim primarily at fiscal conservatism and low tax burdens. At least when it comes to social policy, empirical support for the partisan thesis is increasingly mixed or altogether waning (Huber and Stephens Reference Huber and Stephens2001; but see Tepe and Vanhuysse Reference Tepe and Vanhuysse2010, Reference Tepe and Vanhuysse2012). Prior research on Western Germany has found that state governments manipulate the budget cycle in pre-election periods by increasing transfer payments on social security (Schneider Reference Schneider2010). Oberndorfer and Steiner (Reference Oberndorfer and Steiner2007) find that state governments under conservative parties or coalitions between social democrats and conservatives spend more, not less, on public higher education. Potrafke (Reference Potrafke2011) finds that leftist governments slightly increase spending for schooling, whereas right-wing governments spend more on universities and cultural affairs.

Applying partisan theory to police employment, we would expect conservative voters to be especially likely to reward increases and to punish reductions in police employment levels. Moreover, more conservative parties proposing “law-and-order” platforms can naturally be assumed to benefit from a higher value proximity, both among conservative voters generally and among police officers specifically. Figure 1 shows the policy position of federal parties in Germany towards law and order, taken from the Party Manifesto project (Klingemann et al. Reference Klingemann, Volkens, Bara, Budge and Mc-Donald2006).Footnote ³ Despite differences in the magnitude of this issue dimension over time, there is a rather stable preference order throughout the post-unification period: the mainstream conservative party union CDU/CSU most clearly and consistently emphasises inner security issues. If conservative parties can visibly show themselves to be systematically in favour of more police hiring, law-and-order sensitive voters’ attitudinal support can be turned into actual votes, both from such voters generally (a straightforward “competence signalling” effect) and from law-and-order producing officers specifically (a mixed “patronage and competence signalling” effect).

Figure 1

German parties’ policy position towards law and order.

Note: Law and order (per605) is defined as “Enforcement of all laws; actions against crime; support and resources for police; tougher attitudes in courts; importance of internal security” (Source: https://manifesto-project.wzb.eu/data/MPDataset_full_codebook.pdf). Dots represent years in which federal elections were held.

Figure 2 provides evidence for why the first strategy, straightforward “competence signalling”, is electorally important. It shows the party-specific average perception of inner security as a salient policy issue, with data from GESIS (2010) for the years 1993–2007. Choosing from a list of public policy issues, participants were asked to name their first priority policy issue. Since participants were also asked to name the political party they would vote for in the next election, we could calculate the percentage of respondents considering inner security as the single most important issue for each political party.Footnote ⁴ Supporters of the mainstream conservative CDU/CSU and the extreme right-wing parties DVU and Republikaner consider inner security as an important policy issue more often than supporters of left-wing (SPD, Linke) and green parties (Grüne). Even supporters of the liberal party (FDP), the natural coalition partner for conservative parties, seem not to have particularly strong attitudes towards inner security.

Figure 2

Inner security preferences among German voters by party affiliation. Question: What do you consider as the most urgent issue in Germany? … Inner Security.

Sample: Politbarometer 1993–2007 (GESIS 2010), Obs.: 146,831. Percentage points represent the share of party supporters within parties considering inner security as the most urgent issue, using all other policy issues (e.g. unemployment, economic growth, education, etc.) as the reference category.

The second partisan strategy, mixed “patronage and competence signalling” towards police officers, is important also because the conservative CDU/CSU party family clearly appears to be the “police officers party”. Relying on survey data from Socio-Economic Panel Study (SOEP) (2008) for the waves 2003–2007, Figure 3 shows the party preference among all 91 respondents in the Panel who were police officers or detectives employed as civil servants.Footnote ⁵ Officers clearly lean much more towards CDU/CSU (37 per cent) than towards the other major mainstream party SPD (13 per cent) or any of the smaller parties (close to 0 per cent). A CDU/CSU effort to prove itself the “police-hiring” party would thus promise to be effective in consolidating this natural electoral base. We therefore hypothesise:

Figure 3

Party preferences among German police officers.

Question: Many people in Germany lean towards one party in the long term, even if they occasionally vote for another party. Do you lean towards a particular party? Towards which party do you lean?.

Sample: SOEP 2007–2003 (SOEP 2008), Obs.: 91, respondents are working in the Civil Service (oeffd) as Police, Inspector Detective (ISCO88 Code 3450). Responses refer to the most recent panel wave.

Table 1 compares average annual police employment growth in states where the share of ministers from CDU/CSU (top panel) and SPD (bottom panel) in cabinet was 50 per cent or higher and where it was not (again excluding East German states immediately after re-unification). These “partisan majority growth rates” can only be compared if there has been a change in state-level political majorities within the observation period. Hence, states such as Bavaria and Baden-Württemberg (where the CDU/CSU held the absolute majority over the entire observation period) and Rheinland-Pfalz (idem for the SPD) are excluded. Police employment growth is more positive or less negative with a strong CDU government in just five out of ten relevant states. But in seven out of nine relevant states employment growth is less positive or more negative under a strong SPD government. This simple 50 per cent dichotomy does not account for potential government coalition dynamics: these will be analysed in the multivariate dynamic panel analysis below.

Perceived demand-side salience

Moving beyond the standard political economy questions that are the main motivation of our article, we also query how demand-side salience might affect incumbents’ police employment strategies. On the demand side of law and order, voters are likely to perceive some alleged “causes” and some “forms” of crime as particularly worrying (Benson and Zimmerman Reference Benson and Zimmerman2010). Regarding causes of crime, the popular perception that immigrants are disproportionately crime prone may be no more than a myth – but it is a persistent one (Bersani Reference Bersani2012; Fitzgerald et al. Reference Fitzgerald, Curtis and Corliss2012). Vote-seeking incumbents may therefore respond to such voter perceptions by increasing police employment because of two possible mechanisms, which we refer to as reactive functionalism and opportunistic prejudice. Policy-makers tend to respond to public attitudes on migration issues (Hartmann Reference Hartmann2009). One particularly salient issue refers to the link between migration and crime. In 2007, 21.4 per cent of all criminal suspects in Germany did not have a German passport (Hartmann Reference Hartmann2009, 191). As non-Germans are just 9 per cent of the total population, they appear to be highly overrepresented in crime statistics. Even if violations against the immigration law were excluded from the list of delinquencies, non-Germans have a higher likelihood of becoming crime suspects.Footnote ⁶ German citizens appear to blame migrants when anxious about crime (Fitzgerald et al. Reference Fitzgerald, Curtis and Corliss2012). This “genuine” perception of a link between migrants and crime could then lead incumbents to increase street-visible police employment as a functional reaction to larger shares of non-Germans in the population. Alternatively, increased police hiring, especially of street-visible officers, could result from opportunistic prejudice, whereby politicians understand, but need not share, their voters’ cognitive linkage between crimes and the share of non-Germans and accommodate these beliefs by politically motivated hiring. In both cases we would expect that:

Voters are also likely to perceive some forms of crime as more salient than others (Gerber et al. Reference Gerber, Hirternlehrer and Jackson2010). Clearly, violent crimes such as murder, manslaughter, rape and sexual assault can be expected to have a particularly high salience among voters, intermediated by media coverage (if it bleeds, it leads). But as broken-windows theory indicates, also other forms of subjectively and locally “immediate” street-level crime, such as robberies of cash and valuables, bag-snatching and serious injury on roads, paths and squares, tend to be highly salient as they are visible in the local neighbourhood and are prominently covered in local newspapers (Wilson and Kelling Reference Wilson and Kelling1982). Vote-seeking policy-makers should thus put particular emphasis on being seen to respond strongly to increases in both forms of salient crime statistics, by means of street-visible police hiring. We therefore hypothesise:

To present an overview of the distribution of crime burdens among the 16 German states, Figure 4 shows the frequency of three types of crime measured as an overall period mean, with data from various issues of German Crime Statistics (see Appendix Table A.1). Economic crime is more frequent in states bordering another country, especially Poland. All three forms of crime are particularly high in the city-states, Bremen, Hamburg and Berlin (Birkel Reference Birkel2008, 91). Moreover, there is a clear North–South divide in crime rates for violent crime and street crime, with southern states such as Baden-Württemberg, Bayern and Thüringen showing systematically lower crime rates.Footnote ⁷

Figure 4

Types of crime across German states (1993–2010).

Note: The five shades of grey represent five quantiles. Darker shades represent higher crime rates, measured as the frequency of registered cases per 100,000 inhabitants.

Baseline regression analysis

The baseline dynamic panel data model explores the impact of partisanship and electioneering on the number of civil servant officers using the LSDVc estimator with AB for initial bias correction and O(1/NT) to determine the accuracy of the approximation (Table 2). Both the Sargan's test for over-identifying restrictions and the AB test of second-order autocorrelation give reason to accept the model specifications.Footnote ¹⁰ In the log-lin specification, the β-coefficients express the constant proportional change in the dependent variable for a given absolute change in the value of the independent variable. The percentage point change in the dependent variable associated with a unit change in the independent variable can be calculated as (e^β−1) × 100.

Table 2

Determinants of civil servant police officers: general models

Note: LSDVc using Arellano-Bond (Reference Arellano and Bond1991) for initial bias correction and O(1/TN) to determine the accuracy of the approximation (Kiviet Reference Kiviet1995), standard errors are bootstrapped with 1,000 repetitions, standard errors are in brackets.

LSDVc = Bias-corrected least-squares dummy variable.

***p < 0.01, **p < 0.05, *p < 0.1.

The Wooldridge test (Reference Wooldridge2002, 283) for autocorrelation in panel data strongly rejects the null hypothesis of no first-order autocorrelation (F(1,12) = 26.184, p = 0.0003). This suggests the use of a lagged dependent variable as a technical means to deal with a violation of the estimator assumptions (see Keele and Kelly Reference Keele and Kelly2006 for discussion). In addition, the lagged dependent variable enables us to model the dynamics of the independent variables in terms of short-run (β ₂) and long-run (β ₂/[1−β ₁]) effects (Cochrane and Orcutt Reference Cochrane and Orcutt1949; De Boef and Keele Reference De Boef and Keele2008). The lagged dependent variable is statistically significant at the highest margin and associated with a 38–49 percentage point increase in officer numbers in the last period, ceteris paribus (see Table 2). Among the three macro-structural variables, only population density systematically exerts a statistically significant impact on the number of police officers. In the short run, a unit increase in the change of population density is associated with a 16–21 percentage point increase in officers in the last period, ceteris paribus. In the long run, such an increase is associated with a 28–36 percentage point increase in officers. Assuming a certain rigidity and stickiness of changes in police employment, it is plausible to find larger long and smaller short-run effects. The positive and highly significant coefficient for population density is largely driven by findings for Eastern Germany (Model 8).Footnote ¹¹ We caution, however, that the split sample approach utilised in Models 7 and 8 is a rather rough approach to account for East–West differences since the size and statistical power of the two sub-samples varies. Hence the findings on our electioneering and partisanship hypotheses from these models also need to be taken with caution. To account for adjustment in police staffing due to German re-unification, Models 1–6 include a reunification dummy equalling 1 for Eastern states in 1992–1995, which slightly fails to reach conventional significance levels.Footnote ¹²

The coefficient for Election Year slightly fails to reach conventional levels of significance in Model 2 (p = 0.149), but it is positive and statistically significant in the full Models 5 and 6 (supporting H1). In the short run, a unit increase in Election Year, equivalent to moving elections from the first to the last day of the year, is associated with a 1.4 percentage points increase in officer numbers over its value in the last period, while in the long run, the association increases to 2.3 percentage points. Models 7 and 8 indicate that this electioneering result stems from Western Germany, where it is associated with a 2.3 percentage point increase in the short run and 3.7 percentage point increase in the long run.

Models 3–6 provide evidence supporting H2. In the short run, a unit increase in the change of the share of CDU/CSU cabinet ministers is associated with a 1.6–1.7 percentage point increase in officers in the last period, ceteris paribus, and in the long run with a 2.6–2.8 percentage point increase. Conversely, for a unit change in the share of SPD ministers the association in the short run is −1.6, and −2.6 in the long run. Substantively, the electioneering and partisanship effects are rather small. The preliminary exploration of East–West differences with respect to partisanship shows no significant differences (Models 7 and 8). We therefore continue to explore the full state sample, while controlling for different temporal dynamics with a re-unification dummy.

In line with our auxiliary H3, a larger share of non-Germans is systematically positively associated with an increase in police officers. In the short run, a unit increase in the change in this variable is associated with a 2.0–2.8 percentage point increase in officers in the last period, ceteris paribus. In the long run, the association amounts 3.4–4.6 percentage points. While the statistical association is robust, this analysis cannot determine whether reactive functionalism or opportunistic prejudice lie behind it. Concerning our auxiliary H4, we find that z-standardised frequency indicators of an additive index of violent and street crime, and of economic crime, both exert a statistically significant positive impact on police employment.Footnote ¹³ Consistent with H4, a unit increase in the change of violent and street crime and economic crime are associated with, respectively, a 2.4 and 0.9 point increase in officers in the short run, and 4.1 and 1.5 point increases in the long run. These differences in effect sizes support our theoretical expectation, as the former type of crime is hypothesised to be politically more salient. Moreover, fighting violent and street crime may be more personnel intensive than fighting ‘white-collar’ economic crime.

Models 5 and 6 thus provide tentative support for our four hypotheses. As the next step in our identification strategy we now use this model specification to explore the robustness of findings with alternative estimators (Appendix Table A.3). Substantial findings remain unchanged using the (uncorrected) LSDV estimator (Model 1). The same applies to the LSDVc estimator with an accuracy approximation of O(1/T) rather than O(1/NT) and AB for initial bias correction (Model 2). Using the LSDVc estimator with the AH estimator for initial bias correction causes the variables for economic crime and electioneering to slightly fail reaching the conventional level of statistical significance (Model 3). Applying the Arellano-Bond (Reference Arellano and Bond1991) GMM approach (Models 4–7), rather than bias correction suggested by Kiviet (Reference Kiviet1995), provides almost similar results to the LSDVc estimator, when the GMM(AB) specification uses the maximum number of instruments (17 lagged values of the dependent variable). In Model 4, electioneering exerts a statistically significant effect on officer numbers while partisanship remains insignificant. Reducing the number of instruments provides stable results for electioneering and population density, but weaker findings for alternative determinants. This points out to prior research on the poor finite sample performance of GMM estimators in the context of macro-panels (Baltagi Reference Baltagi2008, 164). In sum, the analysis of alternative estimators generally supports our primary findings but also points to caution about the marginal significance and substantial impact of the political determinants (Table 3).

Table 3

Determinants of civil servant police officers: stepwise

Note: LSDVc using Arellano-Bond (Reference Arellano and Bond1991) for initial bias correction and O(1/TN) to determine the accuracy of the approximation (Kiviet Reference Kiviet1995), standard errors are bootstrapped with 1,000 repetitions, standard errors are in brackets.

LSDVc = Bias-corrected least-squares dummy variable.

***p < 0.01, **p < 0.05, *p < 0.1.

Probing deeper: issues of endogeneity

The next step in the identification strategy is to take into account that findings on the effect of electioneering and partisanship might suffer from endogeneity problems. These could stem from interactions between different levels of government and electoral competition in German federalism. Alternatively, our partisan variable may represent background social and economic factors influencing police employment. To address this issue we re-run Models 3 and 4 from Table 2 and stepwise include the social and economic controls. The analysis shows three things. Firstly, the direction of the impact of the social and economic controls remains stable. Secondly, the effect of partisanship vanishes once we exclude social and economic controls. Thirdly, substantively it is the social and economic control variables that contribute most in explaining variance in the dependent variable. This shows that controlling for relevant confounders is essential to reveal the relevance of political determinants (Cameron and Trivedi Reference Cameron and Trivedi2005, 37). Only after taking into account structural social and economic controls do the partisan composition of the government helps to explain the remaining variance in civil servant police employment.

The next step in the identification strategy is to explore the robustness of findings on partisanship and electioneering by using refined measures of electioneering and partisanship that more explicitly take into account the context of German federalism. Since state-level elections in Germany are run according to a mixed electoral system that emphasises proportional representation, most states have coalition governments. In Table 4 we therefore go beyond single party power to analyse the joint share of cabinet positions of the “natural” coalition partners in Germany: CDU/CSU with FDP (the “centre-right camp”), and SPD with Greens and Linke (the “centre-left camp”). Models 1 and 2 again support H2, with negative signs for the left-camp and positive signs for the right-camp variable. In the short run, a unit increase in the change of “centre-right camp” is associated with 1.5 percentage point increase in officers in the last period, and a unit increase in the change of “centre-left camp” is associated with a 1.2 percentage point decrease in officers in the last period. In the long run, these associations are, respectively, +2.4 and −1.9 percentage points. Models 3 and 4, which explore the effect of absolute majorities for the main centre parties CDU/CSU and SPD, find that absolute majority power for the main centre-right party CDU/CSU is associated with a 1.4 percentage point increase in police in the short run and a 2.3 point increase in the long run. This further corroborates H2. The specific allocation to left-wing or right-wing parties of the Interior Ministry portfolio, which is directly responsible for police staffing, may be as important as general left-wing or right-wing party power in government. We find that in the short run a CDU/CSU Interior Minister is associated with a 1.5 (Model 5) percentage point increase in police officers, and an SPD minister with a 1.2 (Model 6) point decrease. This again supports H2.

Table 4

Determinants of civil servant police officers: further partisan and electoral effects

Note: See Table 2.

We also probe further into the role of electioneering by taking into account early elections and an alternative measure for electoral timing. Where incumbents can call early elections, they might more easily schedule elections to coincide with economic expansions (Tufte Reference Tufte1978, 14, fn. 16; Franzese Reference Franzese2002, 381). Evidence on such endogenous timing of elections is still mixed (Smith Reference Smith1996; Palmer and Whitten Reference Palmer and Whitten2000). For an established federal democracy such as Germany, we distinguish regular scheduled elections from irregular (snap) elections. Irregular elections, however, are not easily timeable by state incumbents. Rather, they are usually the result of unexpected coalition breakdowns, which leave little time for manipulating public employment. Therefore, we would expect that irregular elections have no effect on police employment numbers. This expectation finds support in Model 7: only regular elections are positively and significantly associated with civil servant employment. These results, however, need to be interpreted carefully, since irregular elections are very rare events in state-level politics.Footnote ¹⁴ We also use a variable measuring the number of years left in the current term in Model 8, rather than the Franzese measure. As expected, the coefficient is negative, though it fails to reach conventional levels of significance (p = 0.296).

The next step in the identification strategy is to explore the potential role of interactions between federal- and state-level elections and partisan strength (Table 5). In electoral terms, this sort of interaction has been explored mainly with respect to federal government popularity, on the grounds that stronger declines in this latter variable may lead to bigger losses of federal government parties in state elections (Jeffery and Hough Reference Jeffery and Hough2001; Burkhart Reference Burkhart2004, 3; Berry and Gersen Reference Berry and Gersen2011). There are no studies, to our knowledge, of the effect of federal-local interactions on state employment policies. Nor do we expect to find such an interaction. As argued earlier, state-level employment policies in police, education and cultural affairs are constitutionally largely independent from the federal level in Germany. We constructed categorical variables accounting for whether the PM in the state belongs to the same party as the PM at the federal level (Model 1) and for whether state and federal elections were held simultaneously (Model 2). In neither case did we find evidence supporting the role of federal–state-level interactions when it comes to partisanship and electioneering in police employment.

Table 5

Interactions: federal/state party strength and federal/state elections

Definition: CDU PM state = 1 if the PM of the state (Land) is from the CDU, otherwise 0; CDU PM state–federal = 1 if the PM of the state (Land) and federal (Bund, resp. Bundeskanzler) are from the CDU, otherwise 0; federal election = 1 in federal election years (Bundestagswahl), otherwise 0; state election = 1 in state election years (Landtagswahl), otherwise 0; federal–state election = 1 if federal and state elections take place in the same year, otherwise 0.

Note: LSDVc using Arellano-Bond (Reference Arellano and Bond1991) for initial bias correction and O(1/TN) to determine the accuracy of the approximation (Kiviet Reference Kiviet1995), standard errors are bootstrapped with 1,000 repetitions, standard errors are in brackets.

LSDVc = Bias corrected least-squares dummy variable.

*** p < 0.01, ** p < 0.05, * p < 0.1.

The last step in our identification strategy is to alter the dependent variable by looking at police officers with public employee status, rather than civil servant status as we have done so far. We expect the electoral benefit of manipulating public employees in police service to be lower than for civil servants (bobbies on the beat), since the electoral signal that stems from an increase in public employee police servicemen is less visible to voters. Re-estimating the model specifications presented in Tables 2 and 4 using public police employees as the alternative dependent variable confirms that apart from the role of years left in term, which is associated with a short-run reduction of employees by about 0.5 percentage points per year left, none of these operationalisations resulted in significant findings (Online Appendix Tables 4 and 5). As expected, these findings provide very little support for the electoral and partisan hypotheses.

We can thus summarise the overall estimation results as follows. Firstly, partisanship and electioneering have a robust but substantively small impact on the number of civil servant police officers. Secondly, structural social and economic controls are more influential determinants of police staffing than political determinants. Thirdly, after controlling for these essential confounders there is robust evidence that the ideological composition of government as well as party membership of the Interior Minister are associated with distinct changes in police staffing. Finally, we find no signs of interaction between electioneering and partisanship on other levels of government: police staffing remains a policy field where state governments enjoy a large degree of autonomy from the federal government – which they may use to reap electoral benefits from the strategic manipulation of police employment.