Literature Review, Theory, and Hypotheses

As is well established in the literature on political business cycles, politicians’ willingness to try to use the powers of their office to enhance their chances of re-election is substantial (see, e.g., Nordhaus, Reference Nordhaus1975; Rogoff and Sibert, Reference Rogoff and Sibert1988; Alesina and Roubini, Reference Alesina and Roubini1992; Drazen, Reference Drazen, Bernanke and Rogoff2001; Schultz, Reference Schultz2009; Schneider, Reference Schneider2010; Potrafke, Reference Potrafke2012). Similarly, evidence exists that governments strategically time their legislative activity to maximize their chances of success in future elections (Lagona and Padovano, Reference Lagona and Padovano2008; Brechler and Gersl, Reference Brechler and Gersl2014; Lagona et al., Reference Lagona, Maruotti and Padovano2012, Reference Lagona, Maruotti and Padovano2015). To the extent that voters decide who to vote for based on retrospective concerns and have a recency bias, it is in the interests of politicians to increase their activity levels and, thus, visibility as elections approach. As a result, activity cycles emerge.

At the level of the individual legislator, using passed legislation to appeal to voters is difficult, given that the legislative agenda in parliamentary regimes is dominated by the government (Döring, Reference Döring and Döring1995; Mattson, Reference Mattson and Döring1995). However, introducing bills and amendments allows subsequent credit-claiming with the electorate, making it possible for a legislator to argue in a re-election campaign that she is hard at work, addressing policy issues of interest to her voters (Box-Steffensmeier et al., Reference Box-Steffensmeier, Kimball, Meinke and Tate2003; Crisp et al., Reference Crisp, Escobar-Lemmon, Jones and Taylor-Robinson2004, Reference Crisp, Olivella, Malecki and Sher2013; Bowler, Reference Bowler2010; André et al., Reference André, Depauw, Shugart, Martin, Saalfeld and Strøm2014; Loewen et al., Reference Loewen, Koop, Settle and Fowler2014; Däubler et al., Reference Däubler, Bräuninger and Brunner2016). A similar logic applies to giving speeches (Proksch and Slapin, Reference Proksch and Slapin2012) and asking parliamentary questions (Martin, Reference Martin2011). Given that voters tend to reward the volume of credit-claiming rather than the content as such (Grimmer et al., Reference Grimmer, Messing and Westwood2012), higher activity levels can be an effective re-election strategy.

Turning to the literature on staggered membership renewal, it has tended to focus on its potential for enhancing the stability of institutions and outcomes, achieved by the lengthening of time horizons, as those members of an institution the furthest from re-election are expected to restrain or at least counter-balance the re-election-oriented behaviour of their colleagues nearer the renewal of their terms of office (Waller, Reference Waller1989; Majone, Reference Majone1996). When staggered terms are combined with bicameralism, itself an institution often seen as providing a ‘second thought’ in terms of legislation (Money and Tsebelis, Reference Money and Tsebelis1992; Russell, Reference Russell2001) and generally expected to enhance stability in political systems (Riker, Reference Riker1992a, Reference Rikerb; Mughan and Patterson, Reference Mughan, Patterson, Patterson and Mughan1999), a reduction in short-term election-seeking behaviour might be expected. This theoretical work on staggered terms has, however, focussed on policy outcomes, rather than individual behaviour, and has not explored the possible effects of staggered terms on individual behavioural incentives.

Rather than confirming this expectation of enhanced stability, the few studies that have explored the link between staggered membership and individual activity suggest that staggering leads to additional political cycles, whether financial (Shepsle et al., Reference Shepsle, Van Houweling, Abrams and Hanson2009), as in the case of the American Senate, or behavioural (Fukumoto and Matsuo, Reference Fukumoto and Matsuo2015), as in the case of the Japanese upper chamber, the Sangiin. Recent comparative work analysing the introduction of bills and amendments has similarly found additional political cycles as a consequence of staggering (Willumsen et al., Reference Willumsen, Stecker and Goetz2018). What explains these findings? Briefly, the logic of stability through staggered terms can be turned on its head, in particular when staggering is employed in an institution that affords its members the possibility to affect their chances of re-election. Combining the willingness of politicians to strategically time their actions with the existence of staggered terms can give rise to a multiplication of cycles rather than restraining re-election-seeking behaviour. Since staggered terms of office increase the frequency of elections, they may eventuate in a near-permanent electoral campaign within the staggered legislative chamber, as successive classes of members near re-election and alter their behaviour accordingly.

Staggered membership renewal introduces a systematic difference between members of the same body. While equal in terms of legislative rights, members belong to one of a finite – usually no more than three – number of sub-groups (‘classes’) that share the same electoral horizon, which differs from the other groups. Studying class effects seeks to understand whether the existence of classes is associated with differential behaviour at individual and aggregate/class levels at any given point in time. The difference in electoral horizons implies different incentive structures for the different classes of senators at any given time.

Most importantly, the extent to which they have incentives to focus on a possible re-election campaign, and so to appeal to their future electors, varies systematically across classes. As such, we would expect behaviour to vary by class, with those senators whose terms are nearest expiry engaging more in activities that may improve their chances of re-election. Thus, different incentives at the individual level can be expected to aggregate across entire classes, with the different focus of each class of senators leading to class-specific behavioural patterns. The standard expectation would be that proximity to an election goes hand in hand with an increase in parliamentary activity (see, e.g., Crisp, Reference Crisp2007).

However, a different logic may be at play. As discussed in detail below, the electoral system used for the French Senate is focussed on a relatively small number of elite electors, rather than on a mass electorate. The most effective way to appeal to such electors is not necessarily to be active in the legislative arena, building a record of accomplishments in the run-up to elections, which can then be referred to in campaigning, but rather to focus on direct interactions with as many electors as possible. Work on the US Congress has found that as elections approach, members significantly decrease their attendance at recorded votes (Rothenberg and Sanders, Reference Rothenberg and Sanders1999, Reference Rothenberg and Sanders2002) and are most likely to spend more time campaigning for re-election in their district. With electorates numbering (at most) in the low thousands, direct personal contact can be a very effective way for French Senators to pursue re-election.Footnote ¹ Such direct interactions are time-consuming but, as previous scholarship on the French Senate has noted (Grangé, Reference Grangé1988: 42; Hérin, Reference Hérin2012: 29; Schnatterer, Reference Schnatterer2014: 95–96), cultivating the ‘grand electors’Footnote ² is both expected of senators and likely to pay electoral dividends.Footnote ³ In this study, we assume that the total amount of effort of senators is constant throughout the term, and can be understood as a combination of work in the legislature and work in the constituency. As such, while we cannot directly observe work in the constituency, we can, as discussed in the fourth section, observe effort in the legislature. A reduction in work in the legislature should thus correspond with an increase in re-election-oriented activities ‘back home’. Given the electoral system used for Senate elections, we thus have:

We expect a different activity pattern by senators ‘not up next’. Comparative work has found that legislators in staggered upper chambers who are not up for re-election nonetheless become more active as elections approach (Willumsen et al., Reference Willumsen, Stecker and Goetz2018). We would expect a similar pattern here, as legislators pursue a legislative footprint prior to potential changes to the composition of the Senate, and possibly pick up the slack from their colleagues who are seeking re-election. We thus have:

As the Senate has employed staggered membership renewal for the entirety of the Fifth Republic, we cannot compare behavioural patterns under staggering with a situation where all senators share the same time horizon. However, since the 2003 Loi Poncelet shortened the term length from nine to six years – with the change phased in between the 2004 and 2014 Senate elections – we can observe behaviour under different staggering formulas. With the length of senators’ terms of office shortened, a mechanically greater share of their term can be expected to be used for election-seeking activities, and the amount of time available to achieve their aims in the legislature reduced. Further, senators will have to seek re-election a greater number of times to achieve the same overall length of Senate service. Accordingly, we would expect senators elected for 6-year terms to be more focussed on re-election overall than those elected for 9-year terms, due to their shorter time horizon and the greater share of their mandate that is proximate to the next election. Thus, we have:

As discussed below, some senators are elected by proportional representation (PR), and some are elected by a majority ballot, depending on the number of senators to be elected from their department. These two electoral systems provide differing incentive structures in terms of how to appeal to the electorate (Carey and Shugart, Reference Carey and Shugart1995). Senators elected by majority vote have incentives similar to deputies in the National Assembly, who are also elected by a two-round majority vote system, and who are generally very active in their constituencies (Costa et al., Reference Costa, Lefébure, Rozenberg, Schnatterer and Kerrouche2012; Brouard et al., Reference Brouard, Costa, Kerrouche and Schnatterer2013). As such, we expect senators elected under majority ballot to be more oriented towards their electors. By contrast, senators elected under PR have stronger incentives to engage in work valued by their party, so as to gain a place near the top of the list. Being active in the legislative arena is a key way of doing this. Thus, for senators elected in majority ballot seats, the optimal electoral strategy is to be more present in their department, personally cultivating their electors,Footnote ⁴ whereas senators elected by PR are more dependent on their party’s support which, in turn, also affects representational roles (Schnatterer, Reference Schnatterer2014). This gives us:

Having developed our hypotheses, we now turn our attention to our case, the French Senate, a staggered chamber with both substantive parliamentary powers and incentives and opportunities for individual-level re-election-seeking behaviour.

Staggered Terms and the French Senate

Several political and constitutional changes have taken place in France in the last 15 years that make over-time variation in both individual-level parliamentary activity and class-specific legislative behaviour more likely than in the past. First, as already noted, the staggering formula has changed. The 2003 Loi Poncelet reduced the Senate term length from nine to six years, phased in over the 2004, 2008, 2011 and 2014 Senate elections. This reform also added an additional 25 seats to the Senate, bringing the total to 348, and lowered the age of eligibility to become a senator to 30 years of age (Smith, Reference Smith2009: 102–06). Shorter terms restrict the time budgets of senators and may, accordingly, be expected to accentuate the extent of re-election-seeking behaviour in the Senate since, ceteris paribus, a greater share of the senatorial term will be used for activities related to re-election in a six-year term compared to a nine-year term.

Table 1 summarizes the key characteristics of the French Senate, including the staggering formulas used, the rights of individual senators to engage in parliamentary activities, and the electoral rules used for elections to the Senate.

Table 1. Key characteristics of the French Senate

^a Order of oral questions determined by Conference of Presidents.

^b Senators may further also always request the floor for a speech.

^c Two readings if the government declares the bill urgent.

^d For communes with over 30,000 inhabitants.

The latter require some further explanation. Unlike members of the National Assembly, who are elected directly by popular vote, senators are indirectly elected by approximately 150.000 so-called ‘grand electors’ grouped into electoral colleges by département; the size of the colleges depend on the size of the populations of the respective département. Grand electors include deputies of the National Assembly, regional councillors, general councillors, and municipal delegates. The latter category is by far the largest, at around 96 per cent; roughly one in four municipal councillors serve as a grand elector. The size of the electoral college varies widely. The largest is Nord with 5697 members, Lozère the smallest with 341 members; 31 colleges have fewer than 1000 electors. The average number of electors is 1500, and 78 of 96 mainland départements have fewer than 2000. The average elector represents approximately 400 voters, although some represent only around half that number, while others represent more than twice as many (Smith, Reference Smith2009: 110–113).

The number of senators elected per département ranges from 12, in the case of Paris, to a solitary senator in the case of seven départements. Importantly, senators are elected using two different electoral systems, depending on the number of senators elected in the département. For larger départements, closed-list PR is used, whereas for smaller ones, majority vote is applied. The Loi Poncelet, in addition to the reforms mentioned above, also altered the threshold for the use of PR for elections to the Senate to four seats, having been five seats until 1998, and three seats in the 2001 election,Footnote ⁵ and made small changes to the seat distribution across the départements. For seats smaller than the PR threshold, senators are elected by majority vote, where each elector has one vote per Senate seat, and cannot cast more than one vote per candidate (Kerrouche et al., Reference Kerrouche, Brouard, Deiss-Helbig and Schnatterer2011). As discussed above, the electoral rules under which senators are elected matter for their pattern of parliamentary behaviour.

The 2008 constitutional reform meant a substantial transfer of power to parliament, giving it more control, particularly over its timetabling (Constitution of October 4, 1958 [France], 2008, art. 48), which previously was almost entirely dominated by the government (Huber, Reference Huber1996: 30–36; Brouard, Reference Brouard, Tsebelis and Erik Rasch2011). It also restricted the use of the so-called ‘guillotine’ procedure (Article 49(3) of the Constitution; see Huber, Reference Huber1992)Footnote ⁶ to finance bills, bills dealing with the financing of the social security system, and one additional bill per year, as chosen by the government. Previously, the government was able to freely use this procedure, meaning that it was able to disregard the will of the Parliament relatively easily if it was willing to invoke Article 49(3).

The 2000 constitutional referendum reduced the term of office of the President from seven to five years, the same as the National Assembly. This had the effect of aligning the political time budgets of the two institutions, leading to a synchronization of their respective legislative business cycles (Padovano and Gavoille, Reference Padovano and Gavoille2013). It also virtually eliminated the risk of co-habitation, whereby the presidency is held by a party different from the majority in the National Assembly. The alignment of the political time budgets of the President and the National Assembly can also be expected to make any senatorial cycles stand out clearer, in particular, when the controlling majorities in the two chambers differ.

In addition to these altered constitutional conditions under which staggering takes place, the 2011 Senate elections produced, for the first time, a Centre–Left majority in that chamber. The Left’s success was short-lived, with the Centre–Right winning back the majority at the next renouvellement in 2014. However, the fact that the Centre–Right lost the majority in the Senate meant that its position changed within the French political system. Prior to 2011, the Senate had always been controlled by Right parties. As such, there was little reason for the Senate as a whole to focus on ‘getting things done’ before Senate elections, since the parties of the Right could be confident of returning as the majority. Also, adding to the sense of stability was a combination of 9-year terms and a relatively low rate of turnover with, on an average, more than half of senators being re-elected at each election between 1968 and 2008 (Smith, Reference Smith2009: 129). Illustrating the stability of Senate membership, Alain Poher could be elected as President of the Senate on eight occasions, serving from 1968 to 1992, facing opposition only twice.

Finally, the introduction in 2000 of compulsory alternation between male and female candidates on the lists for PR elections to the Senate (Loi n°2000-493, 2000) meant that for these seats, the need to include female candidates in winning positions on the ballot led to increased turnover in membership, as the number of winnable positions for male candidates standing for re-election was necessarily reduced.

As can be seen in Table 2, after the conflict between de Gaulle and the Senate during the 1960s (Smith, Reference Smith2009, chapter 2), a long stretch of political alignment between the Presidency, the National Assembly and the Senate under presidents Pompidou and Giscard D’Estaing was followed by substantial variation in control, with political control of the key political offices split between the two main political wings to varying degrees for 23 out of 36 years.

Table 2. Partisan control of the presidency, National Assembly, and Senate (1959–2017)

‘Full alignment’ indicates that President, National Assembly and Senate were all controlled by the same side of the left-right spectrum.

Sources: Smith (Reference Smith2009); Parlgov Database (Döring and Manow, Reference Döring and Manow2018).

These different constellations of partisan control of the presidency, the National Assembly and the Senate matter since politics in France, despite the prominence of the President, is at heart parliamentary, with the government having to rely on the confidence of the National Assembly, irrespective of the political preferences of the President (Shugart, Reference Shugart2005). And despite the possibility of being over-ruled by the National Assembly, the Senate has substantial powers of delay in the legislative process (Tsebelis and Money, Reference Tsebelis and Money1995, Reference Tsebelis and Money1997; Constitution of October 4, 1958 [France], 2008).Footnote ⁷ Thus, it can exert influence through ‘persuasion and amendment’ (Mastias, Reference Mastias, Patterson and Mughan1999: 195; see also Grangé, Reference Grangé1984), fulfilling a ‘classic’ function of a ‘revising’ second chamber that is focused on amending bills (Rogers, Reference Rogers2001; Heller, Reference Heller2007).

In sum, the French political system constitutes a setting in which one would expect substantive effects of staggered legislative terms of office. Specifically, it provides ample potential for class-specific legislative behaviour; but to what extent, under which conditions, and in what manner is this potential realized?

Data and Model Operationalization

To investigate the presence of, and changes in, senatorial parliamentary behaviour over the course of a mandate and across classes, we collected data for five distinct parliamentary activities:Footnote ⁸

• Proposed laws, from October 1977 to September 2015;

• Amendments, from October 2001 to September 2015;

• Questions, from April 1978 to September 2015; and

• Legislative and non-legislative speeches, from January 2003 to September 2015.

These activities were selected for several reasons. First, they represent a spectrum of parliamentary activities that senators can engage in, covering law-making, control functions and expressive representation. Second, senators are fairly unrestricted regarding how intensively they engage in these activities. Individual senators are free to introduce bills (Constitution of October 4, 1958 [France], 2008, Art. 39; Standing Orders of the French Senate, 2014, Art. 24) and can freely introduce amendments in the first reading of a bill. From the second reading on, they can freely introduce amendments on issues where the Senate and the National Assembly have not yet agreed (Standing Orders of the French Senate, 2014, Art. 48). Senators are also unrestricted in their right to ask both written and oral questions. For oral questions, the order is determined by the Conference of Presidents (Standing Orders of the French Senate, 2014, Ch. 12, Art. 74–78). Given the limited amount of time available for speaking in the Senate, limitations apply to floor speeches. Senators are free to speak on the floor, but limited to one speech per bill. However, they are entitled to speak for up to five minutes to explain a vote on any amendment (Standing Orders of the French Senate, 2014, Art. 42, 48, 49); further, they may always request the floor from the presiding officer (Standing Orders of the French Senate, 2014, Art. 36).

There is no reason to assume that all five types of individual parliamentary activity will be uniformly affected by an approaching election. All activities have both costs and benefits, and each senator has only a limited number of hours each day to engage in any of them. Empirically exploring how senators choose to spend their time will illuminate which activities they see as most electorally valuable, while overall activity levels will provide information about the trade-off for senators between time spent in the Palais du Luxembourg, the seat of the Senate, and time spent in their constituency. Similarly, the effects will partly depend on the senators’ conception of their roles and on the re-election value that senators place on particular activities (Schnatterer, Reference Schnatterer2014). As such, this is both an empirical question that our data allows us to explore and, as will be discussed in more detail below, in part dependent on the electoral system under which a senator is elected.

For each type of activity, we calculated the number of times a senator engaged in that activity at the individual level, on a monthly basis, for the entire period for which we had data. These observations constitute our dependent variables. As all these variables take a minimum value of zero (i.e., no activity by a senator in that month), and increase in discrete steps of 1, they should be modelled as count data. For all five variables, the variance was much greater than the mean, meaning that a Poisson model would be inappropriate. Instead, a negative binomial model was used.

As we have repeated observations of the same cross-section (senators) over time, panel data analysis is necessary. This, firstly, introduces the possibility of serial correlation of error terms across time due to unobserved unit-specific characteristics. Hausman tests on the data indicate that fixed effects are necessary to address this issue for all five dependent variables.Footnote ⁹ As such, we estimated all the models using unit fixed effects. This also allows us to focus on within-unit changes in behaviour, rather than between-unit changes.

To capture time-dependence across individual cross-sectional units, we follow Carter and Signorino (Reference Carter and Signorino2010) and include a variable in the model which measures the number of months since a senator was last active for a given activity, along with the squared and cubed versions of this variable.Footnote ¹⁰ These variables will also capture any serial correlation of error terms across the panels caused by common exogenous shocks.

Second, as illustrated by the plots of activity levels in the online appendix (Figures 5 to 9), while the levels of activity of different classes tend to follow each other, indicating that senators belonging to different classes are responding to the same stimuli, it is also clear that the magnitude of the response varies by class. There is also some visual evidence of a general over-time increase in the activity levels of members of the Senate (see Figures 10 to 14 in the online appendix).

Conducting Hadri-Lagrange tests for the presence of a unit root indicates that there is evidence for non-stationarity in the data used for all five dependent variables. The null hypothesis of no unit root can be rejected at the 0.01% level in all five cases. However, introducing fixed time effects to control this is problematic, for two reasons. Firstly, as some of the data covers more than 35 years, measured at the level of a single month, this would require over 400 months of dummy variables, which is computationally problematic. Secondly, introducing year dummies to correct for non-stationarity is also problematic, since, if there has been a general increase over time in the tendency of senators to use their power of initiative within their chamber to appeal to voters, then including time dummies would capture such an effect as well, underestimating the effect of electoral cycles on behaviour. Therefore, we include a year counter variable to capture behavioural pattern trends over time.

We operationalized Hypothesis 1a and 1b – time to election – by including a variable in the model indicating the number of months until the next Senate election. To differentiate between those senators whose terms were up for renewal and those where this was not the case, we interacted this with a dummy variable indicating whether a senator is ‘up next’, that is, whether her term expires at the next election. This allows us to distinguish between the effects of a senator’s term being up for renewal (Hypothesis 1a) and an approaching election as such (Hypothesis 1b).

To capture the effect of 6-year terms – Hypothesis 2 – we included a variable equal to one for senators elected to six-year terms, and equal to zero for those elected to nine-year terms. As the effect of approaching elections is expected to vary depending on the value of this variable, both for those senators ‘up next’ and those not, we created a three-way interaction between this variable, the ‘up next’ variable, and the variable measuring the number of months until the next Senate election.

For Hypothesis 3 – electoral system – we included a variable equal to one for senators elected by PR, and equal to zero for those elected by majority vote. As for Hypothesis 2, given that the effect of approaching elections is expected to vary depending on the value of this variable, both for those senators ‘up next’ and those not, we created a three-way interaction between this variable, the ‘up next’ variable, and the variable measuring the number of months until the next Senate election. The models thus include two three-way interactions, as well as the constituent terms of these.

We also include four control variables. First, to capture cross-chamber partisan links, we control for whether a senator’s party was in government at the time of any given activity. Second, we control for the number of years a senator has been a member of the Senate. Third, in order to control for behavioural changes stemming from the first-ever Left-of-Centre control of the Senate from 2011 onwards, and the subsequent real possibility of change in the partisan control of the Senate, we include a variable equal to one for months after the first-ever switch in partisan control in the Senate in 2011, and otherwise equal to zero. Fourth, to control for the effects of a split parliament, which can negatively influence the probability of Senate bills and amendments becoming law (Maus, Reference Maus1988; Tardan, Reference Tardan1988), we included a variable equal to one whenever the Senate was controlled by a partisan majority different to that in the National Assembly, as shown in Table 2, and otherwise equal to zero.

Analysis and Results

Given that the effects of the interaction terms cannot be understood directly from the model outputs (Brambor et al., Reference Brambor, Clark and Golder2006),Footnote ¹¹ we present plots of predicted levels of activity for each type of behaviour, at the level of the individual senator, split into subgroups as indicated by the hypotheses. This allows us to observe the effect of time to election, by looking at the slope of the lines, the difference, if any, between those senators facing the electorate at the next election and those whose re-election date is further in the future, and to observe the difference in these effects between the sub-groups defined by the hypotheses. These plots are shown in Figures 1–4. We also show the overall marginal effects of time to election (i.e., the slopes on the figures), broken down into the various sub-categories in Table 3. Please note that negative values imply higher activity levels as elections approach, and positive values imply lower levels of activity.

Figure 1. Effect of electoral system on amendment introduction.

Figure 2. Effect of electoral system on activity in the French Senate as elections approach.

Figure 3. Effect of term length on legislative speeches.

Figure 4. Effect of term length on activities in the French Senate.

Table 3. Overall marginal effect of approaching elections

+ P < 0.10, *P < 0.05, **P < 0.01, ***P < 0.001.

Negative values imply higher activity levels as elections approach, positive values lower activity levels.

Starting with the effect of approaching elections on the activity levels of senators who are ‘up next’ (Hypothesis 1a), we find support for the hypothesis. In 13 out of 20 cases, senators whose terms are up for renewal become less active as elections approach, as indicated by the negative slopes. In four cases (bills–6-year term; questions–PR ballot; questions–9-year term; legislative speeches–9-year term), the effect of approaching elections is significant at least at the 5% level. In five further cases, the effect is significant at the 10% level. In nine cases, the slope is essentially flat. The only cases where senators become more active as elections approach are bill introduction for senators elected by PR, and bill introduction for those senators elected for 9-year terms.

We similarly find support for senators ‘not up next’ becoming more active as elections approach (Hypothesis 1b). In 14 of 20 cases, approaching elections lead to significantly more active senators, with 12 cases being significant at least at the 5% level, and two at the 10% level. In six cases, we find a null effect. For no activity do senators ‘not up next’ become significantly less active as elections approach. Thus, as expected, approaching elections lead to senators whose term is up for renewal to become less active in the parliamentary arena, whereas those senators not up for re-election become more active as elections approach.

Turning to the effect of 6-year terms relative to 9-year terms (Hypothesis 2),Footnote ¹² we find that senators ‘up next’ serving 9-year terms become less active as elections approach, with the significant exception of the introduction of bills. For those senators serving 6-year terms, the effect of approaching elections is in all cases either significantly negative or null. The contrast is particularly clear in the case of the introduction of bills, where senators serving 6-year terms become significantly less active as elections approach. For those senators ‘not up next’, those serving 6-year terms become significantly more active in all activities except for bill introduction, whereas those serving nine-year terms only become more active in terms of introducing bills and amendments, and asking questions. We thus find support for Hypothesis 2.

Turning to the effect of electoral system (Hypothesis 3), we find that senators ‘up next’ who have been elected by majoritarian ballot only partly alter their behaviour, introducing significantly (at the 10% level) fewer bills. For those elected by PR, the effect is mixed, with senators becoming significantly less active in four activities, but significantly more active in terms of the introduction of bills. For senators ‘up next’, there is thus partial support for Hypothesis 3.

For those senators ‘not up next’, there is little difference in terms of behaviour between senators elected by different electoral systems. Those elected by majoritarian ballot become significantly less active as elections approach for four out of five activities, whereas for those elected by PR, this is only the case for three of the five activities. For both types of senators, there is no effect of approaching elections on non-legislative speeches. Again, there is thus only partial support for Hypothesis 3 for those senators ‘not up next’.

To further explore the effects of staggered terms, we also calculated the difference in activity levels for those senators ‘up next’, and those whose terms were not up for renewal at the next election. These are shown in Figures 15 to 18 in the online appendix. As can be seen, in all but one case (bill introduction by senators elected by PR), approaching elections lead those senators ‘up next’ to become less active relative to those ‘not up next’. In 17 of the 20 cases,Footnote ¹³ the difference between those ‘up next’ and those ‘not up next’ is statistically significant at least for part of a term. Thirty-six months out, those senators ‘up next’ are more active than those ‘not up next’ in 19 of the 20 cases (statistically significant in 12 cases). Zero months out, the difference between the two types of senators is statistically significant in nine cases, with senators ‘not up next’ being more active in all those cases. In 11 cases, there is no statistically significant difference between senators whose terms are expiring and those where this is not the case immediately before an election.

While senators ‘up next’ may start a term being more active, this difference is quickly caught up with by those senators ‘not up next’. Twelve months out, only in the cases of non-legislative speeches (majoritarian election) and questions asked (6-year term) are ‘up next’ senators more active than those ‘not up next’.

In terms of the control variables, we find that split control of parliament leads to significantly fewer amendments introduced and non-legislative speeches given, but significantly more questions asked. Since the possibility of a switch in partisan control became realistic, senators have become significantly more active in terms of introducing bills and amendments, but there is no significant effect for questions or either type of speeches. Belonging to a government party, on the other hand, leads senators to be statistically significantly less active in all activities other than the introduction of bills, where they are marginally, but not significantly, less active than opposition senators. Longer service as a senator is associated with significantly fewer bills introduced, questions asked, and speeches given (of both types), but with significantly more amendments introduced.

To summarize, it is clear that senators alter their behaviour over the course of their full term. As elections approach in which they are not facing the electorate (Hypothesis 1b), senators become more active, regardless of the type of activity or whether they are serving 6- or 9-year terms, or were elected by PR or majoritarian ballot. At the same time, when they are facing the electorate next, senators display a decrease in their parliamentary activity levels (Hypothesis 1a), suggesting that those senators ‘up next’ are focussing their efforts on their constituency and electors in order to improve their chances of re-election. Further, and again across all activities and institutional settings, approaching elections make senators ‘up next’ less active relative to those ‘not up next’. In other words, regardless of the activity or setting, we observe cyclical behaviour both for senators up for re-election, and for senators whose terms are not up for renewal.

By implication, these variations in behaviour over the course of a senator’s term of office mean that staggering leads to an increase in the number of parliamentary activity cycles. This can also be seen in Figures 5 to 9 in the online appendix. While the trend of activity levels is partially shared due to the Senate’s business schedule, the different classes clearly behave differently at any point in time. Overall, we find evidence that staggered terms clearly influence legislative behaviour, by splitting senators into groups with distinct incentives. It is also worth noting that, in contrast to Shepsle et al. (Reference Shepsle, Van Houweling, Abrams and Hanson2009), we found no evidence of collusion, that is, those senators being nearest re-election being given an increasing share of the spotlight by their colleagues.

What are the implications of this co-existence of distinct classes for the Senate as a whole? Here, we turn from quantitative observations to qualitative judgements based on the views of senators solicited by the authors through a number of interviews. Most clearly indicating destabilizing effects of staggered terms, a Green Party senator noted that they led to an ‘eternal electoral campaign’, as the senators belonging to the cohort nearest re-election shifted their behaviour towards a re-election mode.

The interviews also yielded evidence that the party groups in the Senate only use very limited parliamentary resources to improve the electoral prospects of senators closest to an election. A number of senators remarked that for a subset of questions to the government (questions d’actualité), which deal with issues that are highly salient, those senators up for re-election were given priority so as to maximize their media exposure. However, beyond this, no other such help for re-election-seeking senators was indicated by the senators interviewed. Giving senators seeking re-election increased influence on legislation, such as by allowing them to serve as rapporteur on a piece of legislation to improve their electoral prospects, was explicitly ruled out, as the expertise needed to perform a competent job was seen as more important than the electoral prospects of any individual senator.

The interviewed senators also confirmed that, even while aware that their proposed bills and amendments might have little chance of being passed, individual senators had strong incentives to propose them. Proposing legislation was seen less as about changing policy, and more about senators being able to claim that they had been active, and so improve their electoral prospects and their standing with their party.

Finally, several senators remarked that the shortening of the senatorial term had altered the nature of the Senate. The more frequent elections were seen to lead to shorter careers and younger senators, but also insufficient time to ‘learn the ropes’ and immerse oneself in a legislative topic. While senators had always spent time and effort to ensure their re-election, the share of their term spent on this increased as the length of the term decreased, meaning that senators went from spending a third of a 9-year term on getting re-elected to spending half of a 6-year term on this. The interviews thus confirm the findings of the quantitative analysis, with both effects of electoral horizons and changes in the institutional setting also pointed out by individual senators.

To summarize, the results suggest that legislative behaviour in the French Senate is influenced in a predictable manner by a number of conditions. First, time to election matters, but so does the political setting. Six-year terms and the electoral system were found to influence the behaviour of senators in a number of ways. Consistently important is whether a senator is ‘up next’ or not. As expected, the highly personalized electoral system (due to the small number of electors), made senators nearest re-election less active (with this effect strongest for senators elected by majoritarian ballot), while those senators not facing the electorate become consistently more active in parliament as elections approach. The staggering of terms, and the consequent changes in electoral time horizons of senators, thus matter, but are conditional on the institutional and electoral incentives senators face.

Conclusion: Staggering Matters – But Not as We May Expect

This study has explored the effects staggered terms have on legislative behaviour by analysing class effects in the French Senate. At the same time, the study investigated the effects of term lengths and electoral systems on legislative activity levels. Three key findings, based on both quantitative and qualitative data, emerged. Firstly, there is variation in behaviour over the course of senatorial terms; the class to which a senator belongs has a significant influence on their behaviour. Secondly, both term length and the electoral system were found to significantly influence how active senators chose to be. These findings match what would be expected in terms of behaviour due to differing incentives across time and classes.

However, the third key finding, that those ‘not up next’ become more active as elections approach, whereas those ‘up next’ become less active, is more surprising.Footnote ¹⁴ As we noted above, the unusual nature of the electoral system used for the French Senate helps to account for this otherwise puzzling finding. As senators face very small electorates, it is possible for them to personally interact with a very substantial proportion of them, meaning that a senator seeking re-election is well advised to focus on the electors in her department, even if this comes at a cost to their legislative work.

On the other hand, those senators ‘not up next’ may become more active as elections approach due to a general tendency of legislative work being concentrated near the end of a term. At the same time, the increase in their activity can be driven by the need to compensate for their less present colleagues in the last three years of their term, and to pick up the slack created by their absence. A number of senators interviewed pointed out that around one third of senators are essentially inactive throughout their terms, which means that those senators who are at least somewhat active will have to put in significant effort to compensate for the absence of their re-election-seeking colleagues, and keep the machinery of legislation and executive scrutiny going.

Further, these senators may be preparing for a re-election run several years hence, and seeking to impress the electorate they will be facing, building up a reputation for being hard-working. Being active in terms of legislative work in the first part of their term will allow them to earn credit with both their party leadership and the rank-and-file, regardless of how they perceive their representational role (Schnatterer, Reference Schnatterer2014). This is especially the case with legislation that takes a considerable amount of time to be implemented. By laying the groundwork with their party and their electors and building a reputation, this can then be leveraged into re-nomination and party support for re-election, as well as acceptance of some slackness in terms of legislative work in the run-up to their re-election. Senators can then use the last three years of their terms immediately preceding their re-election to cultivate their personal relationship with their electors.

Such an explanation goes against what most of the literature would expect. Both the political business cycle literature (e.g., Alesina and Roubini, Reference Alesina and Roubini1992; Potrafke, Reference Potrafke2012) and the WHYDFMLFootnote ¹⁵ argument would expect that those senators further away from re-election are not influenced by elections they are not involved in. Similarly, the literature on re-appointment to non-majoritarian institutions would expect only those up for re-appointment to alter their behaviour to suit those deciding on replacement (here: the grand electors) (Majone, Reference Majone1996; Waller, Reference Waller1989).

Combined with our finding that senators further from re-election are more active than those nearest facing the electorate, this suggests that the effects of staggered terms are subtler than both the line of argument expecting them to lead to stability of outcomes and the argument focussing on their potential for leading to a multiplication of behavioural cycles suggest. Rather, the effects of staggered terms depend both on the individual incentives they create and on their interaction with other institutional features of a political system, notably the electoral system. Given the use of staggered terms in over 20 upper chambers, understanding the complexity of how these work is a key challenge for future research on legislatures.