Term Length and Political Performance
Ernesto Dal Bó
Martín A. Rossi
June 10, 2010
Abstract
We evaluate the e¤ects of a fundamental lever of constitutional design: the duration of o¢ ce terms. We present a simple model grounded in interviews with legislators
highlighting opposing forces. We exploit two natural experiments in the Argentine
congress (where term lengths were assigned randomly) to ascertain which forces are
empirically dominant. Results for an index as well as separate measures of legislative performance show that longer terms enhance performance. In addition, shorter
terms appear to worsen performance not due to campaign distractions but due to an
investment logic: when returns to e¤ort are not immediate, longer terms yield a higher
chance of capturing those returns. Job stability may dilute accountability (as in classic
theories of electoral discipline–Barro 1973, Ferejohn 1986) but when e¤ort embodies
an investment some stability may strengthen incentives.
JEL Classi…cation: H1
Keywords: Term length, political incentives, legislatures.
Ernesto Dal Bó, University of California at Berkeley and NBER, [email protected];
Martín A. Rossi, Universidad de San Andrés, [email protected]. Rossi acknowledges …nancial
support through grant PICT 2007-787. We thank Pedro Dal Bó, Erik Eyster, Claudio Ferraz,
Je¤ry Frieden, Sebastián Galiani, Martín Gonzalez-Eiras, Neil Malhotra, Ted Miguel, Ernesto
Schargrodsky, Ken Shepsle, Mariano Tommasi, and participants at various seminars and conferences
for valuable comments and suggestions. We are deeply grateful to legislators Patricia Bullrich,
Héctor Maya, Mario Negri, Aldo Neri, Jesús Rodríguez, and Jorge Young who provided expert
advice on the functioning of the Argentine Congress. Juan Escobar, Esteban Petruzzello, Orie
Shelef, and Gabriel Zourak provided excellent research assistance.
1
1
Introduction
A fundamental question in constitutional design is how long o¢ cials should serve before they
can be replaced. One advantage to keeping terms short is that more frequent accountability
should allow tighter control over political agents. This notion is formalized in the classic
papers by Barro (1973) and Ferejohn (1986) on electoral discipline. In their models, shorter
o¢ ce terms lead to less shirking (see Barro 1973, p. 30, for the explicit argument). But a
high frequency of elections may discourage e¤ort when rewards are not immediate and may
also distract o¢ cials from their work. What force dominates the incentives facing real life
politicians?
We present a simple model highlighting how longer terms have ambiguous e¤ects on legislative e¤ort, and try to identify the forces that are empirically dominant by exploiting two
natural experiments in the Argentine legislature. The …rst instance involves the Argentine
House of Representatives in 1983, when 254 House members where randomly assigned to
two and four year terms; the second instance involves the Argentine Senate in 2001, where
a constitutional reform prompted the random allocation of 71 senators to terms lasting two,
four, and six years.
The empirical investigation of the e¤ects of term length faces several identi…cation challenges. Consider for example the substantial cross-country variation in the length of legislative terms (see Table 1). This variation, one may think, could help identify the e¤ects
of term length on legislative performance. However, the length of terms might have been
endogenously selected in response to di¤erent incentive trade-o¤s, hindering identi…cation.
An alternative approach would be to focus on a single legislature with staggered terms
and compare the behavior of legislators facing reelection at di¤erent points in the future.
Amacher and Boyes (1978) followed this approach focusing on the 93rd Congress of the
United States. They found that senators closer to reelection voted more in line with House
representatives (who presumably proxy for constituency interests). Kalt and Zupan (1990)
studied the e¤ects of election proximity for senators in the 95th Congress without …nding
a strong e¤ect. Thomas (1985) tracked the voting pattern of senators in their third versus
their sixth year between 1959 and 1976, …nding a moderating tendency of election proximity.
2
Lott and Davis (1992) provide further references in this area. Their discussion emphasizes
how most of the papers attempting to identify the e¤ects of electoral proximity focused on
voting patterns and su¤ered from measurement and speci…cation problems.1 Closer to our
focus on shirking, Lott (1987) estimated a negative association between retirement (itself an
endogenous variable) and the frequency of voting.
Some of the main problems in the existing literature relate to endogeneity as well as
confounding the e¤ect of a shrinking term length with time e¤ects, tenure e¤ects, or cohort
e¤ects.2 The ideal setting would allow to observe legislators elected at the exact same
time, who di¤er only in the term length they are assigned. Our design o¤ers precisely that
possibility and avoids the aforementioned identi…cation problems.
We …rst analyze the case of the House of Representatives in 1983, for which we have more
observations and more measures of performance–in our view, the measures we use capture
primarily an e¤ort/shirking dimension of performance. The allocation of two and four year
terms was done through a well documented random assignment. We compare the level of
legislative performance of the two year legislators to that of their four year counterparts.
We do this for the …rst two years of legislative activity, while both groups worked side
by side. We follow Kling, Liebman and Katz (2007) in that we …rst study overall e¤ects
by aggregating various performance measures into an index of legislative performance. We
next study the e¤ect of term lengths on the six individual measures of performance that we
obtained: attendance to ‡oor sessions, participation in ‡oor debates (measured by number
of speeches), committee activity (attendance to committee sessions and participation in the
production of committee bills), the number of bills each member introduced, and how many
1
Lott and Davis reexamine data for United States senators and …nd that the proximity of elections does
not signi…cantly a¤ect voting behavior in the United States Senate. The general consensus in the profession
since then has converged on the idea that voting patterns in the Congress of the United States are largely
independent from electoral pressure (see Poole and Rosenthal 1997 and Lee, Moretti, and Butler 2004).
2
In an interesting paper Crain and Tollison (1977) consider governorships as investment projects and
compare campaign expenditures across races for two versus four year positions. They …nd that expenditures
are larger when campaigning for a four year governorship than along two consecutive campaigns for two
year governorships. The main potential confounding factors here are state e¤ects, and the possibility that
di¤erent term lengths may attract di¤erent candidates and di¤erent campaign contributions.
3
of these were approved.3 We …nd a signi…cant e¤ect of a longer term on the performance
index as well as on four of the six metrics of performance. Our study of the Senate in 2001
shows a similar pattern.
According to our theoretical model incentives to perform get stronger in a shorter term
as the “carrot”of reelection lies closer in time (an accountability e¤ect). But the model also
highlights two forces operating in the opposite direction. First, if campaigning commitments
clash with legislative duties, a shorter term may lower legislative e¤ort (a campaigning effect). Second, if legislative e¤ort yields rewards that accrue over time, a shorter term lowers
expectations of reaping those rewards, again discouraging e¤ort (a payback horizon e¤ect).
The empirical …ndings suggest that the accountability e¤ect is overriden by the campaigning
or the payback horizon e¤ects (or both). The size of the e¤ects is substantial. For example,
longer terms increase bill submissions by senators by almost 50%, and they almost double
the number of bills by representatives that get approved. Longer terms might be a coste¤ective way to promote legislative productivity: Ferraz and Finan (2008) estimate that a
20% increase in wages to local Brazilian legislators increases the number of bills submitted
by 25% and the number of bills approved by 22%.
Our next step is to determine whether the data can further substantiate the presence
of the campaigning and/or the payback horizon e¤ects. We …rst compare the performance
of senators who were dealt four versus six year terms, while restricting attention to the
…rst two years of their terms, when campaigning commitments are absent. We …nd that
the senators in the six year track display stronger performance, suggesting that legislators
care about the payback horizon. Could campaigning be a force behind the results in the
House? An implication of our model, and a prediction by legislators we interviewed, is
that if campaigning drives the e¤ects of term length, then these e¤ects should be stronger
among legislators representing geographically remote districts. Those legislators face more
of a con‡ict between campaigning and legislating. We fail to …nd evidence of such a pattern.
3
Some of these measures have been used in the context of American politics. For example, Schiller (1995)
uses the number of bills sponsored by a legislator to measure performance and …nds among other things
that senior senators sponsor more bills than junior ones. Similar results are reported in Hamm, Harmel, and
Thompson (1983) for a few state legislatures in the United States.
4
Can we …nd additional evidence of a payback horizon e¤ect in the House? Our model
predicts that if payback horizon is the relevant channel, legislators who are electorally safer
care less about term length than legislators at risk. The data support this prediction, suggesting that payback horizon e¤ects play a role in the House as well. Our last empirical
exercise shows that the e¤ects of term length do not seem to vary with experience.
Our results underscore the fact that the advantages of more frequent accountability may
be dominated by other forces, such as a need for job stability, at least when terms are short
enough. This would occur whenever returns to e¤ort accrue over time, so that e¤ort resembles
an investment and a longer guaranteed tenure allows further payback. The legislators we
interviewed rated this explanation highlighting an investment logic as highly plausible.
We analyzed two di¤erent natural experiments involving two di¤erent chambers at two
di¤erent points in time, …nding similar patterns and partly addressing issues of external
validity. But much remains to be done to understand comprehensively the e¤ects of term
length. Longer terms may eventually back…re, or have di¤erent e¤ects in di¤erent countries.
But recent work in a very di¤erent context corroborates our …ndings. Titiunik’s (2008)
analysis of state legislatures in Arkansas and Texas where randomizations took place tends
to …nd stronger performance by legislators receiving longer terms.
The structure of the paper is as follows. In Section 2 we present a simple theoretical
model to frame the empirical investigation of the e¤ects of term length. In Section 3 we
describe the natural experiment in the House and present the data. In Section 4 we lay
out the econometric approach and report the main results from the House. In that section
we also discuss threats to identi…cation and robustness. In Section 5 we present data and
additional evidence from the Senate. In Section 6 we investigate possible mechanisms behind
the e¤ects of term length. Section 7 concludes.
2
The model
As mentioned earlier, the classic works by Barro (1973) and Ferejohn (1986) yield a picture
where more frequent elections induce less shirking (see also Schultz 2008 for a model on term
length where politicians have private information and set policy under electoral pressure).
5
We postulate a very simple model of legislative e¤ort that is better suited to match the
structure of our data and captures other e¤ects.
Legislators choose legislative e¤ort e upfront in their term. They must also deploy campaigning e¤ort. There is a basic campaigning e¤ort that all legislators exert, even when they
are not up for reelection, simply because they must collaborate with the party e¤ort of behalf
of those who face reelection. That basic level is normalized to zero. In addition, legislators
facing reelection at the end of a period may face an extra campaign e¤ort outlay c early
in that period. Here
when
2 f0; 1g ; so when
= 0 campaigning is purely a team e¤ort, and
= 1 those facing reelection meet extra campaign commitments. The commitment
c > 0 is assumed exogenous for simplicity, and (in keeping with the legislative sources we
interviewed) the campaigning e¤ort depends on the distance
0 between the capital and
the legislator’s district.
A legislator’s probability of reelection P (e; ) depends on e¤ort e exerted in the term,
and an electoral safety shifter
on notation. P (e) satis…es
requiring that
d3 P
de3
dP
de
. We will typically obviate this latter argument to save
2
0; ddeP2
0; dP
> 0. We impose a technical condition
d
not be too high to avoid uninteresting e¤ects stemming from changes in
the degree of concavity; also we remain agnostic about the sign of the cross-partial
dP
ded
and
to simplify matters we will assume it is zero (this is not necessary for our results but yields
a cleaner analysis). Short Term legislators (ST) face reelection after one period, and Long
Term legislators (LT) face reelection after two periods. After the …rst reelection both types
of legislator have two period terms (matching our House study) so their prospects contingent
on being reelected are identical and have value V .4 Our exposition abstracts from dynamic
programming aspects to focus on …rst period e¤ort choices.
Each unit of legislative e¤ort yields a stream of unitary returns beginning in the current
period in the form of recognition, policy achievements, or legacy. If H = 0 we say the payback
horizon is short because e¤ort yields only a unit return in the current period. If H = 1;
we say the horizon is long because the stream of returns lasts for two periods (the stream
under this long horizon can be made arbitrarily long at some extra notational burden). ST
4
It is easy to show that c low enough yields V > 0, which should hold for legislators to run.
6
legislators have objective,
e
(e + c )2 + P (e) (eH + V ) ;
where (e + c )2 is the cost of overall e¤ort in period 1, while LT legislators have objective,
e
e2 + eH
( c )2 +
2
P (e) V:
Writing the …rst order conditions and rearranging we get the following expressions implicitly characterizing the …rst term e¤ort levels of ST and LT legislators,5
1 + P eST H + dedPST V + eST H
2
2 dP
1 + H + deLT V
=
:
2
eST =
eLT
(1)
c ;
(2)
An interior solution (if it exists) is given by the intersection of a 45 degree line (the left hand
side) and a function with positive intercept (the right hand side) that cuts the 45 degree line
dP (eST )
from above.6 From (1) and (2), if = 0 and H = 0 e¤ort levels are eST = 21 + 2 de V
2 dP (eLT )
and eLT = 12 + 2 de V from which eST > eLT for any 2 (0; 1), proving,
Proposition 1 If campaigning is a team e¤ort ( = 0), legislative e¤ort yields instantaneous
returns only (H = 0), and the more distant future is discounted more heavily ( 2 (0; 1)), then
eST > eLT , i.e., short term legislators exert more legislative e¤ort than long term legislators.
The intuition for this result is that if legislative e¤ort yields no future policy returns
and campaign commitments are the same for all legislators, a more distant reelection carrot
translates into a lower incentive to exert legislative e¤ort today. This “accountability” e¤ect
of election proximity requires that the future be taken into account ( > 0) and that more
distant dates be discounted more strongly ( < 1). The role of discounting relates this e¤ect,
5
The second order conditions for a maximum are, respectively,
2+
2 d2 P
deLT 2
00
2+
d2 p
deST 2
(V + eH) + 2
dP
deST
H < 0 and
dP
de
not be too
V < 0: Both are satis…ed if H = 0 under P < 0. If H = 1, it is required that
large.
6
Given V > 0, the Inada condition lime!0 dPde(e) = 1 guarantees a solution with positive e¤ort.
7
with di¤erences in the precise mechanism at work, to the accountability pressure highlighted
in the Barro and Ferejohn papers.
But the comparison of e¤ort levels eST and eLT is ambiguous in general if
> 0 and/or
H > 0. Under some conditions the accountability e¤ect can be dominated by a campaigning
and/or a payback horizon e¤ect. In particular,
Proposition 2 a) If
= 1 (short term legislators face more campaigning), and the marginal
e¤ect of legislative e¤ort on reelection chances is low enough or the discount factor is low
enough, then long term legislators exert more e¤ort than short term legislators.
b) If H = 1 (the payback horizon is long), reelection is less than fully certain and the marginal e¤ect of legislative e¤ort on reelection chances is low enough, then long term legislators
exert more e¤ort than short term legislators.
Proof. The e¤ort di¤erential is
marginal reelection return to e¤ort
which is positive if either
a), note that if
dP
de
e=
dP
de
H (1 P (eST ))+
2
dP
deLT
V
dP
deST
(V +eST H )
2
approaches zero, we get
e=
+ c . If the
H (1 P (eST ))
2
+ c ,
> 0 or H > 0 (and P < 1). To complete the proof for part
is not small a discount factor close enough to zero still ensures that the
campaigning e¤ect c dominates.
This proposition tells us that two di¤erent forces may make long term legislators exert
more legislative e¤ort despite a discounted reelection motive. First, short term legislators are
busier campaigning during the …rst period–the campaigning e¤ect. Second, these legislators
have a shorter guaranteed time during which to reap returns to their e¤ort–the payback
horizon e¤ect. Whether any of these e¤ects can overcome the accountability e¤ect is an
empirical question. Both the campaigning and the payback horizon channels were deemed
plausible by the legislators we interviewed. It is worth mentioning that the tradeo¤ characterized by the model has wider applicability. The length of review periods will in most
organizations alter the pressure to perform, but also a¤ect in‡uence activities and investment
in organization-speci…c projects.
8
3
Natural experiment and data
3.1
The natural experiment in the Argentine House of Representatives
Argentina is a federal republic consisting of twenty four legislative districts: twenty three
provinces and an autonomous federal district. The National Congress has two chambers, the
Chamber of Deputies (i.e., the House of Representatives) and the Senate.7 At the time of
the return to democracy in December 1983, all 254 deputies were elected at the same time,
starting their terms on December 10, 1983. The Argentine Constitution stipulates four year
terms, no term limits, and the renewal of half the House every two years. In order to stagger
the renewal half of the representatives elected in 1983 got two year terms. The allocation of
two and four year terms was done through random assignment.
The procedure for the random allocation of terms, set by the Comisión de Labor Parlamentaria (the equivalent of the Rules committee in the United States) involved dividing the
representatives into two groups of equal size, Group 1 and Group 2. Each party-province
delegation apportioned an equal number of its members to each group (see Table 2).8 During
a public session on January 20 of 1984, the Secretario Parlamentario performed a lottery
draw, which gave legislators in Group 1 a four year term and legislators in Group 2 a two
year term.9
The party-province delegations did not know which group would get assigned the long
term when apportioning legislators to groups 1 and 2. One could be concerned that if
delegations systematically assigned the better legislators to one particular group then the
assignment of term lengths would not be e¤ectively exogenous. But behind a veil of ignorance
there would be no reason for risk averse legislators to introduce any imbalance, so we do
7
8
For descriptions of the Argentine Congress see Molinelli, Palanza, and Sin (1999) and Jones et al. (2007).
The two representatives from Tierra del Fuego (the smallest district) were allocated to the same group.
In the case that a party had an odd number of representatives from one province the imbalance was corrected
with the analogous surplus from another province where the party also had an odd number of representatives.
9
Since the return to democracy in 1983 the two dominant political parties in Argentina have been the
Unión Cívica Radical (UCR, or Radical party) and the Partido Justicialista (PJ, or Peronist party). In the
period under analysis the majority party was the Unión Cívica Radical.
9
not see this aspect of the design as problematic. In fact, and as shown in Table 3, there
are no statistically signi…cant di¤erences in observables across the two groups of legislators
according to a di¤erence in means test, suggesting that the randomization was successful.10
We perform two additional balancing tests: (i) A regression of the probability of being
assigned to Group 1 or Group 2 on the set of pre-assignment characteristics reveals these
characteristics are not signi…cant predictors of assignment (the F statistic p-value is 0.54);
(ii) We run a regression of the index of legislative performance on the set of observable
characteristics, compute the predicted performance, and then regress predicted performance
on a dummy for drawing the long term. We …nd no signi…cant link between term assignment
and the part of performance driven by observable characteristics (the coe¢ cient on the long
term dummy is not signi…cant; p-value = 0.14) The results of these tests provide additional
assurance that the allocation of terms was random.
Moreover, the majority of party-province delegations (75%) did the assignment in a way
that was essentially random. They assigned legislators occupying an odd-numbered position
in the 1983 electoral party list to one group, and those occupying an even-numbered position
to the other. Positions in the ticket (i.e., whether a legislator is close to the top or not) depend
largely on the demographics of the province area to which the legislator belongs. A second
factor a¤ecting list positions is the perceived popularity of the candidate in her area. Thus,
whether a legislator is …rst or tenth in her party list is not random, but whether she falls
in an even- or odd-numbered position is. The remaining 25% of the delegations did not
follow the odd vs. even slot procedure to assign legislators to groups, but by all observable
measures their assignment appears random as well. As will be shown, our …ndings are robust
to eliminating these delegations from the sample and to clustering standard errors at the
party-province level.
10
One may worry that the aggregate data masks systematic unbalancing within parties that cancels out
in the aggregate, and that perhaps one imbalanced party spuriously drives results. We ran balancing tests
for the two main parties separately and found no pattern of systematic di¤erence. Only one observable is
unbalanced for one party (fraction of lawyers for the Partido Justicialista). As we show later, the e¤ect of
term length holds for each separate major party.
10
3.2
Data and measures of performance
Our dataset contains yearly information on individual performance and legislator characteristics for the period February 1984 - December 1985. Of the 254 legislators that started
their term in December 1983, three resigned and …ve died before December 1985. Thus the
sample includes 492 observations corresponding to 246 legislators for two years.
The data we got from the Argentine Congress includes six objective measures of legislative
performance by each legislator: ‡oor attendance (as percentage of legislative ‡oor sessions),
committee attendance (as percentage of committee sessions), number of committee bills in
which the legislator participated (as re‡ected on the committee bills bearing the legislator’s
signature), number of times the legislator spoke on the ‡oor on a legislative topic, number
of bills introduced by the legislator, and number of those bills that were approved.
The legislators we interviewed see these metrics as capturing di¤erent types of legislative e¤ort and as connected with the shirking dimension of legislative performance. They
also valued the diversity of measures because legislators cultivate di¤erent pro…les. Some
legislators may seek to capture the attention of constituents by introducing bills. Others
may care more strongly about policy, so they may introduce fewer bills but focus more on
approval or on committee work. Floor attendance will re‡ect general involvement with the
daily legislative business.11 Table 4 shows the correlation matrix of the measures we use.
Most correlations are weak and in some cases negative. In light of this and the feedback from
legislators, we believe the metrics we use, while noisy, do proxy for di¤erent and relevant
dimensions of legislative performance.12
In order to draw general conclusions in a context of multiple outcomes, we construct
an index of legislative performance that aggregates the six measures described above. To
construct the index we follow the procedure used by Kling, Liebman, and Katz (2007). The
index is the equally weighted average of z-scores of its components. The z-scores are levels
11
To illustrate how attendance may capture forms of involvement, consider the case of César Jaroslavsky
(UCR, Province of Entre Ríos), who was not involved in speci…c committee work nor introduced many bills
of his own, but who played a central political role as majority leader. He was present in over 94% of all ‡oor
sessions, placing second in the attendance ranking.
12
One indication of relevance may be that these metrics are signi…cant predictors of reelection.
11
standardized by using the mean and standard deviation for the two year legislators. In all
cases higher performance measures have higher z-scores.
The variable of interest is Four-year term, an indicator variable equal to 1 for legislators
assigned to an initial four year term and zero otherwise. The database also includes various
legislator characteristics, such as age (as of November 1983), the distance in kilometers from
the capital of the legislator’s province to Buenos Aires, and a series of dummy variables
equal to 1 when the legislator: is male, is a lawyer, is a …rst time national legislator, holds
a university degree, occupies a leadership position (i.e., president of the chamber, chair of a
committee, and majority or minority leader), belongs to the majority party, and belongs to
a small block (i.e., less than four legislators).
Representatives in Argentina are elected through a closed party list at the provincial
level and not through a special race for a uninominal legislative district, as in the United
States. Under the party list system, the degree of electoral safety depends on how high up in
the party ticket a legislator …nds herself. For example, in 1983 the UCR had 19 candidates
running for the seats corresponding to the province of Santa Fe. But given the party’s vote
share and the proportional representation system, only the top 10 members were seated.
Those legislators close to the 10th position in the ticket faced risk going forward, given
that the party’s electoral strength might erode, and that the legislator’s ranking in a future
party list depends largely on relatively permanent factors, such as the demographics of the
legislator’s home area. We develop a dummy variable (Slackness) to capture electoral safety.
We say a legislator is safe if she entered Congress within the top half of her party-province
delegation (in our example, in the top 5 slots), in which case Slackness is equal to 1. We say
she is relatively at risk otherwise, in which case Slackness is equal to zero.
4
Econometric model and results
Given random assignment, the impact of serving an initial four year term relative to serving
an initial two year term can be estimated by using the following regression model:
Yit =
+ Four year termi + Xi +
12
t
+ "it
(3)
where Yit is any of the performance measures under study for legislator i in period t (where
t = 1984; 1985, the two years following the assignment),
a matrix of legislator characteristics,
Table 5 reports estimates of
t
is the parameter of interest, Xi is
is a time e¤ect, and "it is the error term.
when the dependent variable is the index of legislative
performance.13 Results with and without controls indicate that legislators serving a four
year term display better performance than those serving a two year term and that the
di¤erence is statistically signi…cant. The size of the estimated e¤ect is a third of a standard
deviation of the performance distribution. The size of the e¤ect appears considerable relative
to the e¤ects of observable characteristics. For instance, the performance di¤erence between
the long and short term legislators is more than one and a half times the e¤ect of a university
degree, almost one and a half times the impact of being a legislative leader, and roughly the
same as the e¤ect of being in the majority party. This underscores the strength of the term
length e¤ect. For example, belonging to the majority party has been estimated to play a
sizeable role on legislative e¤ectiveness by Padro-i-Miquel and Snyder (2006).
To determine whether the e¤ects are wide-ranging or concentrated on just one or two
outcomes, we estimate and report in Table 6 the e¤ects on each separate performance metric.
The e¤ect of term length appears quite general: the point estimates are positive for all six
metrics and statistically signi…cant for four of the six.14
15
Figure 1 shows box-and-whiskers
plots comparing the two groups on each measure of performance, which gives a view of the
whole distribution of e¤ects.
13
A typical concern when conducting inference for the estimated parameters of Equation 3 is that the
errors for the same legislator might not be independent. To address this concern we cluster standard errors
at the legislator level, and we later report a speci…cation using data collapsed by legislator.
14
The variables committee bills, ‡oor speeches, bills introduced, and bills rati…ed take discrete values and
are strongly skewed to the right with many observations at zero; thus, ordinary least squares estimation
would be inappropriate. For all of these variables we were able to reject the hypothesis that the dispersion
parameter is equal to zero according to a likelihood-ratio test, which suggests that the correct speci…cation
is a negative binomial model for count data as adopted here.
15
One may conjecture that the years 3 and 4 of four year legislators are similar to the years 1 and 2 of the
two year legislators, so the former should display a lower performance in their last two years. This conjecture
is true (results available upon request), although it could possibly obey to time e¤ects. Thus, the comparison
made in the paper provides the best identi…cation of term length e¤ects.
13
The di¤erences in performance tend to be important in size. Focusing on mean e¤ects
in the controlled regressions, we see from Table 6 that getting a longer term appears to
signi…cantly increase performance on ‡oor attendance by 3% (relative to the mean of the
two year legislators). This is a nontrivial e¤ect; as a point of comparison, Lott (1987)
estimated a 6% decrease in voting frequency for retiring legislators. Committee attendance
is about 12% higher for long term legislators, and the number of committee bills bearing the
legislator’s signature goes up by 19%. Lastly, ‡oor speaking appears to respond by 13% to
a longer term although this result is not signi…cant.
The idea that longer terms increase performance also appears to be backed by the measures of “bill production.” The point estimate in column (10) in Table 6 indicates that the
number of bills introduced goes up by 20%. This estimate, however, is short of signi…cant
with clustered standard errors. When we switch attention from the “volume” measure of
bill production to the “legislative e¢ cacy” measure, namely the number of bills that pass,
the estimates become strongly signi…cant. The point estimate (see column (12) in Table 6)
indicates that moving from a two to a four year term almost doubles the number of bills
passed.16 Ferraz and Finan (2008) …nd a 22% increase in bills approved when Brazilian local
legislators receive wages 20% higher. Thus, the productivity e¤ect of a two year longer term
is commensurate with the e¤ect of nontrivial amounts of money. Overall, our results indicate
a strong tendency for longer terms to increase legislative performance.
Estimates of the control variables (not shown but available upon request) corroborate
the perceptions of the legislators we interviewed and lends credence to their statements.
In their view, ‡oor speaking is the metric less directly connected to e¤ort, and should be
associated …rst and foremost with belonging to a small block. Indeed, this appears to be the
strongest determinant of the frequency of ‡oor speaking. They also indicated that occupying
a leadership position and being a good orator would be associated with participating in ‡oor
16
We explore an alternative de…nition of bill production that considers not only the bills a legislator intro-
duced but also those that she endorsed. Under the alternative de…nition the coe¢ cient for bills introduced
is similar to the one obtained previously. The coe¢ cient for bills rati…ed is smaller, but the magnitude of
the e¤ect is still important (moving to a fouryear term increases the number of bills approved by around
45%). The signi…cance levels remain unchanged for both variables.
14
debates. They were right on this count as well. A leadership position is a strong predictor
of ‡oor speaking. We do not have a direct measure of oratory skills, but one would imagine
that holding a university degree or being a lawyer is correlated with debating abilities. These
variables are indeed strongly related to speech giving.
4.1
Robustness checks
To further address the issue that observations are not independently drawn, we collapse the
observations by legislator and then cluster the standard errors at the provincial level. As
reported in Table 7, our conclusions remain unchanged.
In Table 8 we report additional estimations under a wide range of alternative speci…cations and samples. First, the signi…cance of the term length variable is not a¤ected when we
cluster standard errors at the district level, or according to party-district combinations. Second, our conclusions remain unchanged when we restrict the sample to those party-province
delegations that used the even/odd rule to assign legislators into the two groups, and when,
in addition, we also exclude the …rst two legislators in the party list (we take this extra step
because the di¤erence between occupying an even vs. odd position is generally random but
one could argue this may not apply to the top 2 positions in the party list). Third, we run
separate regressions for the two main parties in order to explore possible heterogeneity in the
e¤ect of term lengths according to political party. Despite the smaller sample size, we still
…nd a positive and signi…cant association between term length and legislative performance
for legislators of both parties. While the point estimate is larger for Peronists, the di¤erence
between the two estimates is not statistically signi…cant. Finally, the value and signi…cance
of the coe¢ cient of interest remains unchanged when we exclude from the sample legislators
that were leaders or those few who changed leader status during the sample period.17
4.2
Potential concerns
Even when our study relies on a well documented randomization, one can still harbor some
potential concerns regarding the exogeneity and nature of the treatment. First, it could
17
We experimented with di¤erent de…nitions of leaderhip and always found similar results.
15
be the case that re-optimizations took place after the random assignment that might have
a¤ected performance for reasons other than the change in term length. For example, legislators given a four year term could have obtained better committee assignments. Thus, in the
presence of hierarchical re-optimization the conclusion that lengthening terms is a good idea
would not follow if such extension were to bene…t all legislators. Our experiment is quite
unique in that it is well documented that all committee assignments, leadership positions,
and placement along the internal hierarchy of the chamber were decided before the assignment of terms was done. Very few re-allocations are observed after the random assignment,
and they appear unrelated to term length.18 The results remain unchanged when we exclude
from the sample those legislators that changed status as chamber leaders or moved in or out
of the most important committees (see the last column in Table 8).
A second story is one where parties may direct resources and responsibilities to the long
term legislators, discriminating against the others. While our data cannot directly reject
such possibility, three observations are due. One, we will show below that the data matches
further predictions of our model, in which term length has e¤ects independent of the share of
legislators obtaining the longer term. Two, the main way in which such centralized process
could work is by means of a¤ecting positions in the internal hierarchy of the House, which as
discussed above, was set before the randomization took place. Three, our interviews failed to
substantiate the notion that such centralized processes play a role in the Argentine Congress.
The view of legislators is that e¤ort choices are essentially an individual decision.19
A third, and perhaps more important, threat to the interpretation of the treatment is this:
18
Only seven legislators left the most important committees after the random allocation of terms (four
two year and three four year legislators). Of the seven substitutes, three legislators ended their term in 1985
and four in 1987. Of all legislators who are considered leaders, only two left their position before December
1985 (one two year and one four year legislator). One of the substitutes ended his term in 1985 and the
other in 1987.
19
Our working paper version discusses in detail an additional story where, in an overlapping generations
context, senior legislators shirk while newly elected ones work hard (so those in the …rst half of their “long”
term outperform others). The bottom line is that such schemes involve a genuine treatment e¤ect that
occurs linked to a form of individual investment. As we discuss below, an investment logic appears to play
a role according to the data.
16
the outcome of the lottery may directly a¤ect the morale of legislators, boosting the spirits of
those who got a four year term, and depressing the rest. In this case, the instrument would
not be a¤ecting behavior through its e¤ect on term length, but directly through its “win”
or “loss”meaning. According to the literature in experimental psychology (see for instance
Amsel, 1992), an implication of the “altered morale” hypothesis is that we should observe
that the relative performance of legislators given two year terms lags in the beginning, and
quickly recovers as spirits revert to normal. As shown in our working paper, an examination
of the performance di¤erential across groups on a monthly basis fails to support the idea that
four year legislators do better only in the …rst few months. This suggests an e¤ect linked to
hard incentives rather than to the lottery having a¤ected morale.
5
Data and additional evidence from the Senate
As a result of a constitutional reform in 1994, the whole Senate needed to be renewed in
2001, when the body’s 71 senators were elected at the same time to start their terms on
December 10.20 The modi…cation of term lengths and renewal rates required that some
senators be assigned two year terms, others four year terms, and others six year terms.
The allocation was done through a well documented random assignment during a public
legislative session on December 12 of 2001. The random assignment was performed at the
district (i.e., province) level. All three senators from each province were jointly and randomly
placed on a two, four, or six year track. One implication of this design is that we cannot use
province dummies, although we can control for distance.
Examining the Senate episode is important for external validity reasons. It allows us to
focus on a di¤erent chamber at a di¤erent historial juncture. Moreover, senators are more
notorious and represent the entire province. This dilutes some of the idiosyncracies of the
party-list cum proportional representation system.
In keeping with the House experiment where the short term legislators were on a two
20
There were 71 senators (instead of 72) because one of the three seats belonging to the federal district
was left vacant until 2003. Like House representatives, senators are eligible for reelection and face no term
limits.
17
year track, we will begin by comparing senators in the two year track against the rest;
accordingly, we will de…ne the Long term variable as a dummy equal to 1 if the senator
got a four or six year term, and zero otherwise. Table 9 exhibits the summary statistics,
showing that the predetermined observables are largely balanced. The only one observable
that shows a signi…cant di¤erence between the short and long term legislators is distance,
which is not an individual but a delegation characteristic. This is not incompatible with
chance when measuring nearly ten observables over two di¤erent experiments. Moreover,
we ran further balancing tests like with did with the House data. We found no ability
of observable characteristics to predict assignment or a component of performance that
signi…cantly correlates with assignment.
Our dataset for the Senate contains yearly information on legislative performance and
legislator characteristics for the two year period starting in December 2001 and ending in
December 2003. We could obtain only three objective measures of legislative performance:
‡oor attendance, the number of bills introduced by each legislator, and the number of bills
approved. Of the 71 legislators that started their term in December 2001, six resigned before
December 2003. Thus the sample includes 130 observations corresponding to 65 legislators
for two years.
Again, in order to draw general conclusions in a context of multiple outcomes we use an
index of legislative performance. In columns (1) and (2) in Table 10 we show results with
and without controls showing e¤ects that are about a third larger than those in the House,
although in a context of fewer observations the result becomes insigni…cant in the controlled
clustered regression.21 Delving into the individual metrics of performance (columns (3)-(5)),
we …nd that the change of receiving a longer term over the mean performance of the two
year senators is of 2% for ‡oor attendance (an e¤ect similar to that in the House), 49% for
the number of bills introduced, and 27% for the number of bills passed. As shown in Table
10, only one of theses di¤erences is signi…cant. However, the di¤erences in all three metrics
go in favor of the long term legislators, which is informative. The e¤ect on bills introduced is
21
When we consider three treatments categories (two, four, and six year terms) we …nd that the point
estimate of being assigned to a four year term is positive but smaller than the one associated to being
assigned to a six year term. In other words, e¤ects appear to get stronger the longer the term assigned.
18
statistically signi…cant and its magnitude is large. To provide an idea of the potential money
value of the e¤ect, note that Ferraz and Finan (2008) estimate that a 20% wage raise for
local legislators in Brazil increases bills submitted by 25%. Thus e¤ect of extending terms
for senators in Argentina appears to be twice that of a 20% increase in wages in Brazil.
Overall, the picture from the Senate corroborates the one we obtained from the House.
6
Investigating mechanisms
6.1
Campaigning and payback horizon
Legislators with longer terms appear in average to exert more e¤ort both in the House and
the Senate. According to our model in Section 2 this could be due to campaigning keeping
short term legislators busier (in terms of the model,
= 1) or to legislative e¤orts yielding
returns that accrue in the future (in terms of the model, H = 1). To investigate whether any
of these forces could be at play, in this section we rely further on our model and our data.
Test 1 - Campaigning vs. Payback Horizon - Senate
A straightforward test is possible in the context of the Senate. We can compare four vs
six year senators during their …rst two years in o¢ ce, when none of them are campaigning.
In terms of the model, that amounts to
= 0. It follows from Propositions 1 and 2 that for
eLT > eST it is necessary that the payback horizon channel be present, i.e., that H = 1.
Columns (1) and (2) of Table 11 report estimates obtained by excluding from the sample
all senators in the two year track, keeping only those in the four and six year tracks. The Long
term variable is now equal to 1 for six year senators and zero for the four year senators. The
results favor the view that six year legislators perform better than four year legislators during
the …rst two years of their tenures. This suggests that it is not campaigning distractions but
a concern with the payback horizon that induces long term legislators to work harder.
This test is not feasible in the House given the lack of six year terms. But another test
is feasible and we present it immediately below.
Test 2 - Campaigning - House
The legislators we interviewed indicated that campaigning commitments arrive as the
19
election nears; they also told us that campaigning, which requires presence in the home district, poses a larger con‡ict with legislative performance for legislators from geographically
remote districts. Thus, according to the legislators we interviewed, if campaigning distractions drive the e¤ect of term length this e¤ect should be stronger among representatives from
distant districts.
In our model, the e¤ort di¤erential is eLT eST
e=
H (1 P (eST ))+
c . The second term of this di¤erential clearly increases with
dP
deLT
V
dP
deST
(V +eST H )
2
and captures the direct
e¤ect described by legislators. Under our assumptions on the function P (e), we can state,
Proposition 3 If campaigning drives term length e¤ects (i.e., if
= 1 causes
e > 0) then
these e¤ects must increase with geographic distance .
Proof. See Appendix.
This proposition says that if campaigning drives term length e¤ects an interaction between the Long term dummy and distance should be positive. As reported in column (3) of
Table 11, the coe¢ cient of that interaction has the wrong sign and is insigni…cant (p-value:
0:57). We fail to …nd support for the campaigning hypothesis in the House.
It seems prima facie counter-intuitive that campaigning, being a time demanding activity,
would not appear to di¤erentially damage the short term legislators. One possibility is that
campaigning does not really pose a stronger con‡ict with legislation for those representing
more remote districts–so the distance-based test is misguided. But legislators appear to
be right that campaigning is costlier for representatives from remote districts.22 Another
possibility is that, as pointed out by the legislators we interviewed, campaigning in Argentina
is to a great extent a team e¤ort at the party level. Legislators that are not running for
o¢ ce often campaign alongside those who are, which in the language of our model would
yield
22
= 0.
For example, ‡oor attendance, to take a metric directly related to physical location, diminishes more in
the second year for legislators from remote districts (e.g., 1,500 kilometers away), indicating campaigning
takes a statistically detectable toll. But belonging to those more remote districts does not signi…cantly a¤ect
the size of the term length e¤ect on ‡oor attendance, suggesting that the term length e¤ect is not driven by
campaigning. We thank a referee for suggesting this check.
20
+
As the House data show that long term legislators exert more e¤ort, and we cannot
say that campaigning e¤ects drive that result, it would be desirable to have additional
con…rmation of the presence of a payback horizon e¤ect in the House.
Test 3 - Payback Horizon - House
Not being able to reject the null that
= 0, we now investigate the presence of the
payback horizon e¤ect by looking at variation in a measure of “electoral safety” . Under
the null that
= 0 and our assumptions on the function P (e), we can state,
Proposition 4 If the extension of the payback horizon drives term length e¤ects (i.e., if
= 0 and H = 1 causes
e > 0) then these e¤ects must decrease with electoral safety .
Proof. See Appendix.
This proposition captures the intuition that if the payback horizon hypothesis is true
legislators who are electorally safer should care less about the term length they get than
those at risk. In one extreme, term length does not a¤ect the expected time in o¢ ce of
someone whose reelection is guaranteed. In the other extreme, a term length extension of
one day increases by one full day the expected tenure of someone who is certain not to
be reelected. (Formally, notice the e¤ort di¤erential
e contains a term
H
2
(1
P ) so the
payback horizon matters in inverse proportion to electoral certainty.) Thus, if the payback
horizon channel drives our main …ndings, an interaction variable between being safer and
getting the longer term should be negative.
The variable capturing electoral safety, “Slackness,” was introduced in Section 3. In
column (4) of Table 11 we report results indicating that the interaction of the Four year
term variable and the electoral safety measure is indeed negative and statistically signi…cant.
In other words, the e¤ects of term length are stronger among “at risk” legislators. Moving
from the bottom half to the top half of one’s party-province delegation undoes two-thirds of
the e¤ect of being dealt a longer term. These results suggest that the payback horizon plays
a role in the term length e¤ects found in the House, and in turn imply that legislative e¤ort
contains an investment aspect.
We believe the investment logic is plausible given the time structure of legislative activity.
The mean time lag since the introduction of a bill to its approval was, in our sample period,
21
of 327 days, even abstracting from the time spent preparing the bill. Also, a legislator will
often have to decide whether to spend time absorbing information that will be useful while
a policy issue remains current, which may be a few years.23 Alternatively, a legislator may
buy an apartment close to the legislature in order to lower her future costs of attending
meetings, or shut down her private law …rm to more fully focus her time and attention on
legislation. These costs are slow to amortize and legislators with shorter terms may decide
not to incur them. In addition, a change from a two to a four year term should signi…cantly
a¤ect the e¤ective payback horizon facing Argentine legislators: reelection rates in Argentina
have been traditionally low (around 25% for our sample).24
6.2
Does experience matter?
Lastly, we enquire about the nature of the potential investments involved in legislative activity. This is useful from the perspective of the optimal design of terms. One possibility is that
investments, once accumulated, render legislators unresponsive to term lengths. This would
be the case for instance if longer terms a¤ected performance because they foster learning by
doing about general legislative aspects that do not depreciate. Another possibility is that
investments depreciate and new investment must be made. If the …rst possibility were true,
the investment logic would be relevant early in a legislative career. Then it could be optimal
to allow inexperienced legislators a long …rst term in order to incentivize initial investments,
while having more senior legislators face shorter terms in order to bene…t from stronger accountability. If the second possibility were true, term lengths could be determined as they
are now, without regard to seniority.
In order to explore what type of investment predominates, we ask whether the e¤ects
of term lengths are stronger for inexperienced senators. We rely on the 2001 Senate data
23
As one legislator put it to us (our translation from the Spanish), “The library of Congress is vast...you
can do a ‘doctorate’in here, as some of the committees deal with really complex issues. Obviously, if you
are going to be around for longer, you dig in more...”
24
The average reelection rate for the 1983-2001 period was 20% (see Jones, Saiegh, Spiller and Tommasi
2002). The reelection rates for the years 1985 and 1987 were 30% and 22% respectively, and are not
statistically di¤erent.
22
because, given the dictatorship in the period 1976-1983, there were only a handful of House
members with previous legislative experience in 1983. In columns (5) and (6) of Table 11 we
estimate speci…cations including an interaction between the Long term variable (four and six
year terms) and Freshman for senators. If once-and-for-all investments are what drive the
investment logic, we would expect experienced senators to care less about what term length
they get. In other words, we would expect the interaction between Freshman and the Long
term variable to be positive and signi…cant. We …nd that interaction to be not signi…cant
and to have the incorrect sign in the model with controls. We conclude that investments
either depreciate after a few years, or are related to varied and continuing opportunities that
are also valuable for experienced legislators. As a result, nothing indicates that term lengths
should be determined with regard to seniority, since the bene…ts of longer terms appear to
accrue to experienced legislators too.
7
Conclusion
Term length is a fundamental aspect of constitutional design. To the best of our knowledge,
this is the …rst investigation that relies on natural experiments to study how term length
a¤ects political performance. We frame our empirical investigation with a theoretical model
highlighting competing forces. On the one hand, longer terms push the reward of reelection
farther into the future (a weakened accountability e¤ect). On the other hand, longer terms
may free legislators from campaigning, as well as increase the chance that they will be
around to bene…t from their past legislative involvement (the campaigning and payback
horizon e¤ects, respectively).
Results from two natural experiments in the Argentine House and Senate, where the
length of terms was randomized, suggest that the accountability e¤ect is dominated by one
or both of the campaigning and payback horizon e¤ects.
In this paper we take steps not just to investigate whether term lengths matter and in
which direction, but also to gain insight on why they matter. Our data does not allow us
to say that campaigning drives results, while supporting the idea that the payback horizon
matters to legislators. Our results suggest that even though accountability may be important,
23
in the context we study incentives seem to be strengthened by job stability. While forms
of job stability may weaken accountability, some stability in the form of a longer term may
strengthen incentives when e¤ort embodies an investment.
The issue of term lengths and more generally the bene…ts of job stability should be important not only to politics but to the private sector, where the accumulation of …rm-speci…c
human capital is deemed relevant. There is evidence that longer tenures are associated with
more productivity (Auer, Berg and Coulibaly 2005) and that lower expected turnover induces forms of e¤ort with an investment component, such as take up of job-speci…c training
(Beeson Royalty 1996). Surprisingly, there is a dearth of empirical work at the micro level
that can neatly identify the e¤ects of a guaranteed longer tenure on employee incentives,
presumably due to the identi…cation di¢ culties that have hindered analogous studies in politics to date. Hopefully our approach can provide a blueprint for studies extending attention
to other political settings and labor relations more generally.
8
Appendix
Proof of Proposition 3: Di¤erentiating the …rst order conditions with respect to
deST
d
=
dP dV
de d
2+
d2 P
de2
+2 c
(V +eH)+2
dP
de
H
and
deLT
d
=
dP dV
de d
2
2
+ d P
V
de2
. It is easy to show that
dV
d
yields,
< 0, so numer-
ators are both positive, while denominators are negative from the second order conditions.
The numerator in
deST
d
is larger by 2 c, implying that if the denominators are close then an
increase in distance reduces e¤ort more strongly among ST than LT legislators, yielding the
result. What is needed is that the denominator for
deST
d
not be a lot higher. It is easy to
show that this is ruled out by our assumption on P 000 (e).
Proof of Proposition 4: Di¤erentiating the …rst order conditions with respect to
dP (eST )
dP dV
dP dV
+
deST
deLT
de d
d
de d
yields, d =
and
=
. It is easy to show that dV
> 0,
2
d P
dP
2
d2 P
d
d
2+
de2
(V +eH)+2
de
H
+
de2
V
so both comparative statics are positive and the one for ST legislators has an extra positive
dP (eST )
. Under the same conditions guaranteeing that denominators are close
term involving d
in the proof to Proposition 3 we obtain that
deST
d
24
>
deLT
d
implying
d e
d
< 0.
9
References
Amacher, Ryan and William J. Boyes (1978). Cycles in senatorial voting behavior: implications for the optimal frequency of elections, Public Choice 33 (1), 5-13.
Amsel, Abram (1992). Frustration theory: An analysis of dispositional learning and
memory. Cambridge, UK: Cambridge University Press.
Auer, Peter, Berg, Janine, and Ibrahim Coulibaly (2005). Is a Stable Workforce Good
for Productivity?, International Labour Review 144(3), 319-343.
Barro, Robert (1973). The control of politicians, an economic model, Public Choice
14(1), 19-42.
Beeson Royalty, Anne (1996). The E¤ects of Job Turnover on the Training of Men and
Women, Industrial and Labor Relations Review 49(3), 506-521.
Crain, Mark and Robert D. Tollison (1977). Attenuated property rights and the market
for governors, Journal of Law and Economics 20 (1), 205-211.
Ferejohn, John (1986). Incumbent performance and electoral control, Public Choice 50,
5-26.
Ferraz, Claudio and Frederico Finan (2008). Motivating Politicians: The Impacts of
Monetary Incentives on Quality and Performance, IZA Discussion Paper No. 3411
Hamm, Keith, Robert Harmel, and Robert Thompson (1983). Ethnic and partisan minorities in two southern state legislatures, Legislative Studies Quarterly 8, 177-189.
Jones, Mark, Sebastian Saiegh, Pablo Spiller, and Mariano Tommasi (2002). Amateur
Legislators –Professional Politicians: The Consequences of Party-Centered Electoral Rules
in a Federal System, American Journal of Political Science 46(3), 656-669.
Jones, Mark, Sebastian Saiegh, Pablo Spiller, and Mariano Tommasi (2007). Congress,
political careers, and the provincial connection. In Pablo Spiller and Mariano Tommasi (eds.),
The institutional foundations of public policy in Argentina, Cambridge University Press.
Kalt, Joseph and Mark Zupan (1990). The apparent ideological behavior of legislators:
testing for principal-agent slack in political institutions, Journal of Law and Economics 33
(1), 103-131.
Kling, Je¤rey, Je¤rey Liebman, and Lawrence Katz (2007). Experimental Analysis of
25
Neigborhood E¤ects, Econometrica 75 (1), 83-119.
Lee, David, Enrico Moretti, and Matthew Butler (2004). Do voters a¤ect or elect policies?
Evidence from the U.S. House, Quarterly Journal of Economics 119 (3), 807-859.
Lott, John R. Jr. (1987). Political cheating, Public Choice 52, 169-186.
Lott, John R. Jr. and Michael Davis (1992). A critical review and an extension of the
political shirking literature, Public Choice 74, 461-484.
Molinelli, Guillermo, Valeria Palanza, and Gisela Sin (1999). Congreso, presidencia y
justicia en Argentina. Materiales para su estudio. Buenos Aires: Temas Grupo Editorial.
Padró i Miquel, Gerard and James Snyder (2006). Legislative E¤ectiveness and Legislative Careers, Legislative Studies Quarterly 31(3), 347-381.
Poole, Keith and Howard Rosenthal (1997). Congress: a political-economic history of
roll call voting, Oxford University Press.
Schiller, Wendy (1995). Senators as political entrepreneurs: using bill sponsorship to
shape legislative agendas, American Journal of Political Science 39, 186-203.
Schultz, Christian (2008). Information, polarization and term length in democracy, Journal of Public Economics 92 (5-6), 1078-1091.
Thomas, Martin (1985). Election proximity and senatorial roll call voting, American
Journal of Political Science 29 (1), 96-111.
Titiunik, Rocío (2008). Drawing your senator from a jar: term length and legislative
behavior, Mimeo UC Berkeley.
26
Table 1. Duration of terms in selected legislatures
Term
duration
(years)
Countries, and states in the United States of America
United States House of Representatives
US states: Alaska, Arizona, Arkansas, California, Colorado, Connecticut, Delaware, Florida,
Georgia, Hawaii, Idaho, Illinois, Indiana, Iowa, Kansas, Kentucky, Maine, Massachusetts,
2
Michigan, Minnesota, Missouri, Montana, Nevada, New Hampshire, New Jersey, New
Mexico, New York, North Carolina, Ohio, Oklahoma, Oregon, Pennsylvania, Rhode Island,
South Carolina, South Dakota, Tennessee, Texas, Utah, Vermont, Virginia, Washington, West
Virginia, Wisconsin, and Wyoming
3
Australia, Bhutan, El Salvador, Mexico, Nauru, New Zealand, and Philippines
Albania, Andorra, Angola, Armenia, Argentina, Austria, Belgium, Bosnia and Herzegovina,
Brazil, Bulgaria, Chad, Chile, Colombia, Costa Rica, Croatia, Denmark, Dominican Republic,
Germany, Ghana, Greece, Guatemala, Haiti, Honduras, Hungary, Iran, Iraq, Japan, Jordan,
4
Kazakhstan, Kiribati, Lebanon, Liechtenstein, Lithuania, Macedonia, Madagascar, Mauritius,
Moldova, Mongolia, Montenegro, Netherlands, Nigeria, Poland, Portugal, Romania, Russia,
Slovakia, Solomon Islands, South Korea, Spain, Syria, Tuvalu, and Vanuatu
US states: Alabama, Louisiana, Maryland, Mississippi, Nebraska, and North Dakota
Afghanistan, Antigua and Barbuda, Azerbaijan, Bahamas, Bangladesh, Barbados, Benin,
Bolivia, Botswana, Burkina Faso, Burundi, Cambodia, Cameroon, Canada, Cape Verde,
Central African Republic, China, Comoros, Cuba, Cyprus, Czech Republic, Democratic
Republic of the Congo, Djibouti, Dominica, Egypt, Ethiopia, Fiji, France, Gabon, Gambia,
Grenada, Guinea, Guyana, India, Ivory Coast, Jamaica, Kyrgyzstan, Laos, Lesotho,
5
Luxembourg, Malawi, Malaysia, Mali, Malta, Mauritania, Monaco, Morocco, Mozambique,
Namibia, Nicaragua, Niger, North Korea, Pakistan, Panama, Papua New Guinea, Paraguay,
Peru, Republic of the Congo, Saint Lucia, Samoa, San Marino, Senegal, Seychelles, Sierra
Leone, Singapore, South Africa, Suriname, Tajikistan, Tanzania, Togo, Tunisia, Turkey,
United Kingdom, Uruguay, Uzbekistan, Vietnam, Zambia, and Zimbabwe
6
Liberia, Sri Lanka, Sudan, and Yemen
Note: when the legislature consists of a lower and an upper house, we consider the lower house.
Table 2. Distribution of legislators by province and political party for the random allocation of terms
Group 1 (later assigned a four year term)
Group 2 (later assigned a two year term)
District
Total UCR
PJ
PI
UCD DC AUT
MPJ
MPN PB Total UCR
PJ
PI
UCD LIB MFP MPN PB Total
Capital
25
7
3
1
1
12
7
4
1
1
13
Buenos Aires
70
18
16
1
35
19
15
1
35
Catamarca
5
1
1
2
1
2
3
Córdoba
18
6
3
9
5
4
9
Corrientes
7
2
1
1
4
1
1
1
3
Chaco
7
1
2
3
2
2
4
Chubut
5
2
1
3
1
1
2
Entre Ríos
9
2
2
4
3
2
5
Formosa
5
1
2
3
1
1
2
Jujuy
6
1
1
1
3
1
2
3
La Pampa
5
1
1
2
1
1
1
3
La Rioja
5
1
2
3
1
1
2
Mendoza
10
3
2
5
3
2
5
Misiones
7
2
2
4
2
1
3
Neuquén
5
1
1
2
1
1
1
3
Río Negro
5
2
1
3
1
1
2
Salta
7
2
2
4
1
2
3
San Juan
6
1
1
1
3
1
1
1
3
San Luis
5
1
1
2
2
1
3
Santa Cruz
5
1
1
2
1
2
3
Santa Fe
19
5
5
10
5
4
9
S. del Estero
7
2
2
4
1
2
3
Tucumán
9
2
3
5
2
2
4
T. del Fuego
2
1
1
2
254
65
55
1
1
1
1
1
1
1
127
64
56
2
1
1
1
1
1
127
TOTAL
Notes: UCR is Unión Cívica Radical; PJ is Partido Justicialista; PI is Partido Intransigente; UCD is Unión del Centro Democrático; DC is Democracia Cristiana; AUT is
Partido Autonomista; MPJ is Movimiento Popular Jujeño; MFP is Movimiento Federalista Pampeano; MPN is Movimiento Popular Neuquino; PB is Partido Bloquista de San
Juan; LIB is Partido Liberal.
Table 3. Summary statistics - House
Long track
Short track
Difference of means
82.346
79.833
2.513**
(0.733)
(0.726)
(1.032)
Committee attendance (in %)
56.507
50.872
5.635**
(1.543)
(1.552)
(2.188)
Number of committee bills
47.336
41.397
5.939*
(2.366)
(2.224)
(3.247)
Number of floor speeches
5.616
4.339
1.277
(0.561)
(0.569)
(0.800)
Number of bills introduced
6.224
5.496
0.728
(0.655)
(0.587)
(0.879)
Number of bills ratified
0.276
0.128
0.148***
(0.042)
(0.025)
(0.049)
Index of legislative performance
0.205
0
0.205***
(0.038)
(0.032)
(0.050)
Age
50.168
50.868
-0.700
(0.959)
(0.926)
(1.333)
Male
0.944
0.967
-0.023
(0.021)
(0.016)
(0.026)
Freshman
0.944
0.934
0.010
(0.021)
(0.023)
(0.031)
Lawyer
0.368
0.273
0.095
(0.043)
(0.041)
(0.059)
University degree
0.184
0.157
0.027
(0.035)
(0.033)
(0.048)
Leader
0.136
0.083
0.053
(0.031)
(0.025)
(0.040)
Slackness
0.600
0.521
0.079
(0.044)
(0.046)
(0.063)
Majority party
0.504
0.488
0.016
(0.045)
(0.046)
(0.064)
Small block
0.056
0.058
-0.002
(0.021)
(0.021)
(0.030)
Distance
6.598
6.82
-0.220
(0.515)
(0.572)
(0.769)
Note: Standard errors are in parentheses. The long track corresponds to a four year term. The short track
corresponds to a two year term. Leader is a dummy variable equal to 1 when the legislator is the president of
the chamber, a majority or minority leader, or a committee chair. Freshman is a dummy equal to 1 for
Representatives without any previous legislative experience at the national level. Slackness is a dummy
equal to 1 if, given the party-province list in the 1983 elections, the legislator placed in the top half of the
elected delegation. Small block is a dummy equal to 1 when the legislator belongs to a party holding three
or fewer seats. Distance is the distance (in hundreds of kilometers) from the capital of the legislator’s
district to Buenos Aires (the seat of the national legislature). The number of observations is 492.
*Significant at the 10% level; **Significant at the 5% level; ***Significant at the 1% level, based on a t-test
of equality of means.
Floor attendance (in %)
Table 4. Correlations among measures of legislative performance
Floor
Committee
Committee
Floor
Bills
Bills
Attendance
attendance
bills
speeches
introduced
ratified
Floor attendance
1
Committee attendance
0.39
1
Committee bills
0.27
0.49
1
Floor speeches
0.04
-0.09
-0.08
1
Bills introduced
-0.02
0.05
0.18
0.04
1
Bills ratified
0.16
0.03
0.07
0.15
0.13
1
Note: correlations computed on raw data observed on yearly basis. Collapsing the data by individual
yields similar results.
Table 5. The effects of term length on legislative performance
(1)
Four year term
0.205
{0.061}***
Age
Male
Freshman
Lawyer
University degree
Leader
Slackness
Majority party
Small block
Distance
District dummies
No
Index of legislative performance
(2)
0.188
{0.062}***
0.0004
{0.003}
0.082
{0.165}
0.088
{0.181}
0.031
{0.074}
0.115
{0.071}*
0.201
{0.110}*
-0.021
{0.058}
0.134
{0.064}**
0.132
{0.155}
-0.006
{0.005}
No
(3)
0.186
{0.063}***
0.001
{0.003}
-0.004
{0.137}
0.124
{0.172}
0.055
{0.073}
0.117
{0.076}
0.211
{0.113}*
-0.023
{0.060}
0.128
{0.065}**
0.140
{0.147}
Yes
Notes: Standard errors clustered at the legislator level are in braces. All models include a time dummy and
are estimated by OLS. The number of observations is 492. *Significant at the 10% level; **Significant at
the 5% level; ***Significant at the 1% level.
Table 6. The effects of term length on legislative performance by outcome
Four year term
Change
Controls
Method
Floor attendance
(1)
(2)
2.513
2.548
{1.076}**
{0.956}***
3%
3%
No
Yes
OLS
OLS
Committee attendance
(3)
(4)
5.635
6.259
{2.764}**
{2.498}***
11%
12%
No
Yes
OLS
OLS
Committee bills
(5)
(6)
0.133
0.175
{0.099}
{0.108}*
14%
19%
No
Yes
Negbin
Negbin
Floor speeches
Bills introduced
Bills ratified
(7)
(8)
(9)
(10)
(11)
(12)
Four year term
0.263
0.122
0.133
0.184
0.753
0.669
{0.197}
{0.172}
{0.184}
{0.135}
{0.256}*** {0.249}***
Change
30%
13%
14%
20%
112%
95%
Controls
No
Yes
No
Yes
No
Yes
Method
Negbin
Negbin
Negbin
Negbin
Negbin
Negbin
Notes: Standard errors clustered at the legislator level are in braces. For OLS models, Change is calculated
as 100*Estimate/mean of the respective output for legislators in a two year track. For Negbin (Negative
Binomial) models, Change is calculated as 100*[exp(Estimate)-1]. All models include a time dummy.
Controls include Age, Male, Freshman, Lawyer, University degree, Leader, Slackness, Majority party,
Small block, and the set of district dummies. The number of observations is 492. *Significant at the 10%
level; **Significant at the 5% level; ***Significant at the 1% level.
Table 7. Robustness check: data collapsed at the legislator level and standard
errors clustered at the district level
Index
Floor
Committee Committee
Floor
Bills
Bills
attendance attendance
bills
speeches
introduced
ratified
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Four year term
0.242***
2.573***
7.888***
0.176**
0.123
0.182
0.666**
(0.075)
(0.926)
(2.309)
(0.090)
(0.200)
(0.175)
(0.285)
Change
3%
16%
19%
13%
20%
95%
Controls
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Method
OLS
OLS
OLS
Negbin
Negbin
Negbin
Negbin
Notes: Standard errors clustered at the district level are in parentheses. For OLS models, Change is
calculated as 100*Estimate/mean of the respective output for legislators in a two year track. For Negbin
(Negative Binomial) models, Change is calculated as 100*[exp(Estimate)-1]. Change is not calculated for
the Index since this variable is normalized to zero for legislators in a two year track. Controls include Age,
Male, Freshman, Lawyer, University degree, Leader, Slackness, Majority party, Small block, and the set of
district dummies. The number of observations is 246. **Significant at the 5% level; ***Significant at the
1% level.
Table 8. Additional robustness checks
Four year term
Controls
Observations
(1)
0.186***
[0.055]
Yes
492
Index of legislative performance
(2)
(3)
0.186***
0.156***
(0.057)
{0.058}
Yes
Yes
492
316
(4)
0.255***
{0.068}
Yes
168
(5)
(6)
(7)
(8)
0.157**
0.274**
0.207***
0.180***
{0.065}
{0.121}
{0.066}
{0.064}
Controls
Yes
Yes
Yes
Yes
Observations
244
220
438
472
Notes: Standard errors clustered at party/district combinations are shown in brackets in column (1).
Standard errors clustered at the district level are shown in parentheses in column (2). Standard errors
clustered at the legislator level are in braces in columns (3)-(8). Regression (3) includes only those
party-province delegations that used the even/odd rule in order to assign legislators into the two groups,
whereas regression (4) also excludes the first two legislators in the party list. Regression (5) includes
only majority party (Unión Cívica Radical) legislators, whereas regression (6) includes only minority
party (Partido Justicialista) legislators. Regression (7) excludes legislators that are leaders, and
regression (8) excludes those legislators that change leader status during the sample period. All models
include a time dummy and are estimated by OLS. Controls include Age, Male, Freshman, Lawyer,
University degree, Leader, Slackness, Majority party, Small block, and the set of district dummies.
**Significant at the 5% level; ***Significant at the 1% level.
Four year term
Table 9. Summary statistics - Senate
Floor attendance (in %)
Number of bills introduced
Number of bills ratified
Age
Male
Freshman
Lawyer
University degree
Leader
Majority party
Small block
Distance
Long track
83.345
(1.267)
33.591
(3.134)
2.318
(0.361)
50.750
(1.276)
0.591
(0.075)
0.545
(0.076)
0.455
(0.076)
0.273
(0.068)
0.705
(0.070)
0.341
(0.072)
0.091
(0.044)
1284.432
(108.178)
Short track
80.975
(2.090)
23.476
(2.869)
1.667
(0.294)
52.238
(1.801)
0.714
(0.101)
0.571
(0.111)
0.333
(0.105)
0.476
(0.112)
0.619
(0.109)
0.286
(0.101)
0.190
(0.088)
983.715
(71.909)
Difference of means
2.3705
(2.445)
10.115
(4.249)**
0.652
(0.466)
-1.488
(2.207)
-0.123
(0.126)
-0.026
(0.134)
0.121
(0.130)
-0.203
(0.131)
0.085
(0.129)
0.055
(0.124)
-0.010
(0.098)
300.718
(129.896)**
Note: Standard errors are in parentheses. The long track corresponds to four and six year terms. The short
track corresponds to a two year term. Leader is a dummy variable equal to 1 when the legislator is the
president of the chamber, a majority or minority leader, or a committee chair. Freshman is a dummy equal
to 1 for Representatives without any previous legislative experience at the national level. Small block is a
dummy equal to 1 when the legislator belongs to a party holding three or fewer seats. Distance is the
distance (in hundreds of kilometers) from the capital of the legislator’s district to Buenos Aires (the seat of
the national legislature). The number of observations is 130. **Significant at the 5% level, based on a t-test
on equality of means.
Table 10. Evidence from the Senate
Index of legislative
performance
Long term
Change
Method
Controls
Floor
attendance
Bills
introduced
Bills ratified
(1)
(2)
(3)
(4)
(5)
0.353
{0.192}*
0.298
{0.231}
OLS
No
OLS
Yes
1.299
{2.822}
2%
OLS
Yes
0.402
{0.165}***
49%
Negbin
Yes
0.241
{0.221}
27%
Negbin
Yes
Notes: Standard errors clustered at the legislator level are in braces. For OLS models, Change is calculated
as 100*Estimate/mean of the respective output for legislators in a short track. For Negbin (Negative
Binomial) models, Change is calculated as 100*[exp(Estimate)-1]. Change is not calculated for the Index
since this variable is normalized to zero for legislators in a short track. All specifications include a time
dummy. Controls include Age, Male, Freshman, Lawyer, University degree, Leader, Majority party, Small
block, and Distance. The number of observations is 130. *Significant at the 10% level; ***Significant at
the 10% level.
Table 11. Tests for the campaigning and the time-horizon hypotheses
Index of legislative performance
(3)
(4)
(5)
(6)
House
House
Senate
Senate
Long term
0.226
0.300
0.342
0.346
{0.103}** {0.090}***
{0.279}
{0.338}
Long term x Distance
-0.006
{0.010}
Long term x Slackness
-0.207
{0.127}*
Long term x Freshman
0.020
-0.085
{0.387}
{0.462}
Distance
0.00002
-0.004
0.00008
{0.0002}
{0.007}
{0.0002}
Slackness
-0.019
0.080
{0.058}
{0.079}
Controls
No
Yes
Yes
Yes
No
Yes
District dummies
No
No
No
Yes
No
No
Observations
88
88
492
492
130
130
Notes: Standard errors clustered at the legislator level are in braces. In regressions (1) and (2), which
exclude two year Senators, Long term is a dummy variable equal to 1 for senators in a six year track and
zero for senators in a four year track. In regressions (3) and (4) Long term is a dummy equal to 1 for
representatives in a four year track and zero otherwise. In regressions (5) and (6) Long term is a dummy
equal to 1 for senators in either a four year or a six year track and zero for senators in the two year track.
Distance is the distance (in hundreds of kilometers) from the capital of the legislator’s province to Buenos
Aires (the seat of Congress). Slackness is a dummy equal to 1 if, given the party-province list in the 1983
election, the legislator is placed in the top half of the elected delegation. Freshman is a dummy equal to 1
for senators without any previous legislative experience at the national level. All models include a time
dummy and are estimated by OLS. Additional controls include Age, Male, Freshman, Lawyer, University
degree, Leader, Majority party, and Small block. *Significant at the 10% level; **Significant at the 5%
level; ***Significant at the 1% level.
(1)
Senate
0.449
{0.241}*
(2)
Senate
0.451
{0.248}*
Figure 1. Box-and-whiskers plot by type of legislative outcome
(4-year vs. 2-year tracks)
2 years
2 years
4 years
4 years
20
40
60
80
0
100
20
40
2 years
2 years
4 years
4 years
0
50
100
150
200
0
20
Committee bills
2 years
4 years
4 years
20
40
60
80
100
2 years
4 years
0
2
Index
40
60
0
1
2
Bills ratified
Bills introduced
-2
80
100
80
Floor speeches
2 years
0
60
Committee attendance
Floor attendance
4
3
4