Anchors Away A New Approach for Estimating Ideal Points Comparable

across Time and Chambers

Nicole Asmussen and Jinhee Jo

University of Rochester

Work in progress. Ask permission before citing.

Existing methods for estimating ideal points of legislators that are comparable across time

and chambers make restrictive assumptions regarding how legislators’ ideal points can

move over time, either by fixing some legislators’ ideal points or constraining their

movement over time. These assumptions are clearly contradictory to some theories of

congressional responsiveness to election dynamics and changes in constituency. Instead of

using legislators as anchors, our approach relies on matching roll calls in one chamber and

session with roll calls or cosponsorship decisions on identical bills introduced in a different

chamber or session. By using these “bridge decisions” to achieve comparability, we can

remove any assumptions about the movement of legislators’ ideal points. We produce

these estimates for both chambers of the 107th (2001-2002) and 108th (2003-2004)

Congresses, and we show that our estimates provide interesting insights into the nature of

legislative behavior change in response to electoral dynamics.

1

The ideal point estimation techniques developed in the last two decades have added

significantly to our understanding of the ideology of members of Congress, partisan

cleavages, and the dimensionality of the issue space. The microfoundation of these

estimation techniques in the spatial model has also given researchers the confidence that

these estimates, based on explicit and defensible assumptions, are suitable for quantitative

analyses of legislators’ behavior. However, in contrast to the theoretically grounded

assumptions needed to generate ideal points within a single session of Congress, the

assumptions normally made to allow for comparability of ideal points across time and

chambers are not well grounded in theory, and in most cases are clearly contradictory to

theories of congressional responsiveness to electoral dynamics and changes in

constituency. This is particularly troubling because most uses of ideal point estimates for

empirical testing or even simple description involve comparisons over time.

Existing methods for estimating ideal points across time and chambers make

restrictive assumptions regarding how legislators’ ideal points can move over time, either

by fixing some legislators’ ideal points or by constraining their movement, usually by

restricting them to linear movement in one direction.1 These restrictions present two

problems. First, they preclude the finding of the sort of abrupt behavior changes we would

expect from members of Congress responding to electoral dynamics (e.g., a primary

challenge) or changes in constituency (e.g., redistricting). Second, they obscure the effect of

national trends that involve many members moving in the same direction (for example, in

1 One exception is Treier (2010a). Although there are no restrictions on the movement of legislators’ ideal

points, the ADA, the ACU, and the president are treated as if they are members of Congress, and their ideal

points are assumed to be constant.

2

response to a national crisis), because the shift in ideal points coincides with a shift in the

scale on which the ideal points are measured. As a consequence, existing ideal point

estimation techniques are not adequate for testing theories of legislative behavior change

because the findings (or lack thereof) are an artifact of the estimation technique.

Moreover, existing estimates muddle our understanding of Congress. By depending on

legislators as anchors, who may themselves be drifting, they present us with a description

of Congress that has too many moving parts. Neither the members’ ideal points, nor the

meaning of the scales can be said to be truly stationary.

The approach presented in this paper does not depend on the use of legislators or

any other actors as anchors, as all existing methods do. Instead, we make the agendas

comparable across time and chambers by matching roll calls in one chamber and session

with roll calls or cosponsorship decisions on identical bills in a different chamber or

session, making use of the fact that many bills are introduced in identical form in both

chambers and over multiple sessions, a subset of which receive a floor vote. By identifying

the bills common to more than one chamber or session to use as anchors, we can remove

any assumptions about the movement of legislators’ ideal points. Furthermore, the

meaning of these ideal points is easier to interpret because the scale on which they lie is

stationary.

This paper demonstrates the utility of this approach by presenting ideal point

estimates for both chambers of the 107th (2001-2002) and 108th (2003-2004) Congresses.

We show that these estimates provide interesting insights into the nature of legislative

behavior change in response to electoral dynamics. Specifically, we find that following the

Republican Party’s surprisingly strong showing in the 2002 midterm elections, Republicans

3

moved to the right while the behavior of Democrats was more mixed, with moderate

Democrats less likely to shift to the left than extremists. Also, senators who were reelected

in 2002 had more extreme ideal points in the session following their reelection than in the

preceding two years, suggesting that elections have a moderating effect on candidates.

Before presenting these results, we examine the assumptions underlying existing methods.

Existing Methods

The techniques used to make estimates comparable across time and chamber can be

separated into three categories: those which hold all legislators’ ideal points fixed over

their careers, those which hold some fixed and allow others to float, and those which allow

legislators’ ideal points to move along a trendline. We discuss methods falling into each of

these categories as well as their susceptibility to two major problems: the inability to

account for abrupt behavior changes and the obscuring of shifts in ideal points that move

many members in the same direction.

Fixed Career Method

One assumption that ensures comparability is that legislators maintain the same

ideal point for their entire career; even members who serve in both chambers are assumed

to have the same ideal point in the House and the Senate. This is the assumption that

underlies Poole and Rosenthal’s Common Space scores.2 Although the most restrictive

2 These are the only ideal point estimates in the NOMINATE family that are comparable across chambers. For

Poole’s explanation of the differences between D-NOMINATE, DW-NOMINATE, W-NOMINATE and Common

Space scores, see http://voteview.com/page2a.htm.

4

assumption, these estimates are commonly used to compare levels of polarization in the

House and Senate and to calculate gridlock intervals. Obviously the assumption that ideal

points are fixed means that these estimates are useless for studying changes in individual

behavior, since the estimated ideal point of each legislator is simply an average of their

voting behavior over their entire career. Neither can these ideal points capture shifts in

ideology that move many members in the same direction since they can only measure a

member’s ideology relative to the other members serving at the same time. As Bailey

(2007) points out, this deficiency leads to the substantively questionable result that pro-

segregationist members of Congress from the 1950’s have ideal point estimates in the same

neighborhood as moderate Democrats of the 1990’s.

Related to the “fixed career” estimation strategy, the “shift and stretch” method of

Groseclose, Levitt and Snyder (1999) assumes that each legislator has an unobserved

constant ideal point throughout their career, but that this ideal point is measured with

error. The task of making session- and chamber-specific ideology scores comparable is

accomplished by minimizing these errors.3 Although allowing for movement across time,

the fundamental assumption of time-invariant ideal points militates against uncovering

changes in legislators’ voting behavior between sessions, and the authors readily admit

that their method cannot pick up on universal shifts in members’ preferences because this

would coincide with a shift in the ideological scale (47).

3 Groseclose, Levitt and Snyder (1999) are concerned with generating comparable ADA scores. Ansolabehere,

Snyder and Stewart (2001) apply the “shift and stretch” technique to ideal point estimates constructed with

the Heckman and Snyder (1997) method, and Ensley, Tofias and de Marchi (2010) apply the technique to

Poole and Rosenthal’s (1997) static W-NOMINATE scores.

5

Fix and Float Methods

A less restrictive approach to making estimates comparable is to fix the ideal points

of some legislators, while allowing all other legislators to “float.” A classic example of this

involves fixing the ideal points of two extremists, say Senators Ted Kennedy and Robert

Byrd, to -1 and 1, and estimating a separate ideal point for each legislator for each session

relative to the career ideal points of these two legislators. The “fix and float” method is also

used by Poole and Rosenthal in computing NOMINATE scores for legislators who switch

parties within a session: a pre-switch and post-switch ideal point is estimated for the party-

switcher, while all other members’ ideal points are held constant. Treier’s (2010b)

overlapping constraints method makes use of this technique to make senators’ ideal points

comparable across time, by fixing members’ ideal points in the first four years of the term,

but estimating a separate ideal point for the last two years. Because of the staggering of

elections in the Senate, it is always the case that one-third of Senators and held fixed while

the other two-thirds are allowed the change. While “fix and float” methods usually are only

applicable to over-time comparisons, cross-chamber comparability can be achieved by

fixing the ideal points of actors who take positions on roll calls in both chambers, such as

the president and interest groups, as in Treier (2010a).

Estimates produced by the “fix and float” method seem well suited for capturing

abrupt changes in legislative voting behavior since no restrictions are placed on the

movement of the ideal points of the floating legislators. However, two caveats are in order.

First, much hinges on the assumption that the fixed legislators’ ideal points are actually

constant. Treier (2010b) finds that even such firm ideologues as Kennedy and Byrd have

shifted their ideological positions in response to electoral cycles—a cautionary example

6

that should make researchers wary of establishing comparability from a small number of

fixed legislators. A second difficulty is establishing a baseline with which to compare

changes in legislators’ ideal points. For example, if only legislators suspected of changing

ideal points (e.g., party-switchers) are allowed to float, while all others are fixed, then the

finding of a significant change in the ideal point of the floaters lacks a relevant reference

point because the change in all other members’ ideal points is by assumption required to be

zero.4

Furthermore, the “fix and float” method cannot properly account for universal shifts

in members’ ideal points. Since these shifts in preference can only be registered as changes

in the ideal points of the floating members, the cause of the movement of ideal points may

be erroneously attributed to the characteristic that separates the floating legislators from

the fixed legislators. Consider the following example in which we wish to determine the

effect of redistricting on legislators’ voting behavior. We estimate legislators’ ideal points

pre- and post-redistricting, and to ensure comparability across time, we fix the ideal points

of the legislators whose district boundaries remained the same. Suppose that in the true

data generating process, redistricting has no effect, but there is a national trend that pulls

all legislators two units to the right. We would estimate the non-redistricted members as

having an ideal point at the average of their pre- and post-redistricting ideal point (that is,

4 This is the shortcoming that Clinton, Jackman and Rivers (2004) identify in McCarty, Poole and Rosenthal’s

(2001) analysis of party-switchers. While their criticism is conceptually right, their reanalysis of the effect of

Jeffords’s party-switching sidesteps the issue of cross-time comparability altogether by simply assuming that

the distribution of pre-switch and post-switch ideal points both have mean zero and unit variance—in

essence implying that no anchors are needed to make the estimates comparable.

7

one unit to the right of their initial ideal point), while redistricted members’ second period

ideal points are estimated to be two units to the right of their first period ideal points.

Therefore, what appears to be an effect of redistricting is simply an artifact of the

estimation technique.

Trendline Methods

A final category of estimation techniques restricts legislators’ movements across

time to a specific functional form. The most common restriction, used in Poole and

Rosenthal’s D-NOMINATE and DW-NOMINATE scores, allows for only a linear time trend in

legislators’ ideal points. Poole and Rosenthal dismiss estimations with higher order

polynomials as not worthwhile given the low gains in classification success at the cost of

computational time (1997, 28-29). Bailey (2007) also makes use of time trends, allowing

members serving 7-14 years to move with a linear time trend, those serving 15-20 years to

move with a quadratic trend, and those serving more than 20 years to move with a

polynomial trend of degree five.

One might think that the inclusion of polynomial trends would allow these estimates

to capture abrupt changes in legislative behavior. However, as Treier (2010b) points out,

neither a linear nor a quadratic trend can imitate the kind of oscillating behavior we might

associate with Senate election cycles. Figures 1 and 2 illustrate this point further by

depicting situations in which legislators are commonly hypothesized to change their voting

behavior, such as redistricting, moving from the House to the Senate, a primary challenge,

retirement or senatorial election cycles. In Figure 1, the colored lines represent the true

ideal points of these hypothesized members, while the black lines are the best-fitting linear

trendlines, which are obviously inadequate to capture these types of movements. In Figure

8

2, we fit trendlines based on polynomials of degree five. Surprisingly, even such a complex

trendline is not able to mimic these hypothesized movements. Except in the case of the

retiring member, the change in estimated ideal points based on the trendline greatly

understates the effect of the “treatment.” It is no wonder, then, that Poole and Rosenthal

(1997) find minimal benefits to upping the degree of the polynomial. Furthermore, even

though polynomial trends are more suited to picking up on gradual rather than sudden

changes in ideal points, trendline methods, like fixed career methods, still fail to capture

universal shifts in legislators’ ideal points because they can only account for relative

changes.

Although Martin and Quinn’s (2002) dynamic ideal point estimation technique does

not rely on trendlines, we address it here because it has a flavor similar to trendline

methods. Specifically, in Martin and Quinn’s estimation, the current period’s ideal point is

modeled as a random walk from the previous period’s ideal point, which means that the

expectation of one’s ideal point in the current period is one’s ideal point in the previous

period. However, the amount of movement detected over time depends entirely on the

choice of the prior distribution of the evolution variance parameters which determine the

amount of smoothing, as the authors readily admit (2002, 147). Thus, while we can use

this method to see whether there is some solid evidence showing that people really change

their ideological positions over time, by assuming very small evolution variance parameter,

it is not appropriate to measure how big such movements are.

Our Approach

The common denominator of all existing methods for making ideal points

9

comparable across time and chambers is the focus on the actors as anchors. The problems

of accounting for abrupt changes in legislative behavior and identifying universal shifts in

members’ ideal points cannot simply be solved by more computational power, higher order

polynomials, or cleverer fix and float arrangements. Instead, the focus must shift from

fixing the actors to fixing the agendas. Recent analyses have incorporated bridge votes,

usually votes on identical versions of conference reports, as a means to make House and

Senate estimates comparable (Bailey 2007, Treier 2010a), yet the lack of votes that bridge

across time means that the sole use of roll call votes as anchors is not enough to allow for

the dropping of all restrictions on the movement of legislators’ ideal points.

Our approach connects the agendas over time by including legislators’ decisions to

cosponsor legislation. We make use of the fact that many bills are introduced in identical

form in both chambers and in multiple sessions of congress, a subset of which receive a

floor vote in at least one session of a chamber. By matching the roll call votes on bills in a

particular chamber and session with cosponsorship decisions on identical bills introduced

in another session or chamber, we can achieve comparability of agendas and do away with

using legislators as anchors.

Although Mayhew (1974) once claimed that cosponsorship is a costless and

therefore inherently symbolic activity, more recent studies have considered cosponsorship

to be indicative of legislators preferences and have employed cosponsorship data to

characterize social networks in Congress and identify influential members (Fowler 2006),

to estimate status quo positions of roll call votes (Peress 2009), and test theories of

position-taking and party effects (Kessler and Krehbiel 1996, Krehbiel 1995). Moreover,

Aleman, Calvo, Jones and Kaplan (2009) and Talbert and Potoski (2002) both use

10

cosponsorship data to estimate ideal points of members of Congress and find that

cosponsorship-based estimates correlate roughly with roll-call-based measures, although

neither of these studies exploits identical, multiply introduced bills as a way to make

comparisons across time and institutions as we do in this analysis, nor do they attempt to

scale cosponsorship and roll-call decisions together.

The question of whether cosponsorship data can be treated in the same way as roll

call data is a challenging one. Bernhard and Sulkin (2010) give support for the notion that

cosponsorship decisions ought to be taken seriously—members of Congress rarely renege

(only about 1.5% of cosponsors vote no on final passage), and reneging members are

subject to punishment. On the other hand, Howard and Moffett (2010) point out the

paradox that roll call support on final passage usually far exceeds number of cosponsors,

and Desposato, Kearney and Crisp (2008) caution against using cosponsorship data to

estimate ideal points because of the difficulty of determining the meaning of non-

cosponsorship.

We deal with this challenge by interpreting the decision to cosponsor (or sponsor) a

bill as equivalent to voting “yes,” but we treat non-cosponsorship as missing information.

This means that the cosponsorship decisions on a bill in a particular chamber and session

are not by themselves informative because the estimation procedure would treat the

decisions the same as a unanimous vote. But if an identical bill receives a non-unanimous

floor vote in another chamber or session, then the locations of the bill’s cosponsors also

help determine where the cutline falls (and vice versa) and hence how the ideal point

estimates of each chamber-sessions fit on the same scale. By matching roll calls with

cosponsorship decisions we can estimate comparable ideal points even while making

11

minimal assumptions about what non-cosponsorship means.

Data

To demonstrate the utility of our method, we restricted our data-collection efforts to

the 107th (2001-2002) and 108th Congresses (2003-2004), with plans to expand the

timeframe in future research. We identified identical bills5 introduced in the House and

Senate of the same congress by browsing the Bill Summary & Status reports on the Library

of Congress’s THOMAS website (http://thomas.loc.gov), and collecting the “related bills”

information. We identified identical bills introduced over multiple sessions in the same

chamber by comparing bill descriptions in the Congressional Bills Project files (Adler and

Wilkerson 2001-2004). To make data collection more manageable, we excluded: 1)

resolutions and amendments, 2) bills that received fewer than 10 (co)sponsors in the

Senate and 20 (co)sponsors in the House, and 3) bills that had no identical companion in

another chamber or congress that also met the above thresholds.

Of the 17,414 bills introduced in the four “chambers” (the 107th House, the 107th

Senate, the 108th House and the 108th Senate), we identify 38 bills identically introduced in

all four chambers, 48 introduced in three of the four chambers, and 526 introduced in two

of the four chambers. Of these 612 bills, 60 received a recorded floor vote in at least one

chamber. Eliminating unanimous votes, 39 “bridge decisions” (i.e., combinations of roll call

and cosponsorship decisions that span more than one chamber) remain. In addition to

these, we also include 57 “bridge votes” (i.e., roll call votes on identical bills, usually

5 In practice, we only checked if bill descriptions were identical, and in the case of small discrepancies in bill

descriptions, the bill texts were compared.

12

conference reports), 30 occurring in the 107th Congress and 27 in the 108th.

The roll call/cosponsorship matrix is constructed as follows. The rows of the matrix

are the members of Congress, with a separate row for each session that the member

served. The columns of the matrix contain three types of items: the unique roll call votes

for each chamber for each session, and the “bridge votes” which span both chambers of a

single session, and the “bridge decisions” which span multiple chambers. The roll call

entries take the values 1 for yea, 0 for nay, and NA otherwise. The cosponsorship decisions

take the values 1 for sponsorship or cosponsorship, NA for non-cosponsorship, and NA if

the bill was not introduced in the member’s chamber. We also include the president’s

position on roll calls in the House and Senate as recorded by Congressional Quarterly. The

resulting matrix, depicted in Figure 3, has 1099 legislators and 3488 items.

Method

One advantage of our approach is that once this matrix is constructed, ideal points

can be estimated using any of the techniques appropriate for the estimation of static ideal

points. We use the Bayesian simulation approach of Clinton, Jackman, and Rivers (2004),

which fits roll call data with an item response model via Markov Chain Monte Carlo

(MCMC) methods. This is easily implementable by anyone with a basic knowledge of R

using the package pscl and the command ideal. We estimated a one-dimensional

model, running 25,000 iterations, with a burn-in of 10,000, and a thinning interval of 100.

The estimation time was approximately 5 hours.

Results

Figure 4 presents the density plots of the posterior means of the members of each

13

chamber. The dotted line indicates the chamber median. Note that although we present

each chamber on a separate graph for visual clarity, the scales are directly comparable. For

instance, we can see that the median shifted to the right in both chambers in the 108th

Congress and that the House median is to the right of the Senate median in both

Congresses.

Figure 5 plots our ideal point estimates against Poole and Rosenthal’s Common

Space scores, which assume that each member has a fixed ideal point throughout their

entire career. While highly correlated, important differences are apparent. First, our

estimates are able to pick up on members’ changing preference throughout their careers.

Note, for instance, that we estimate Senator Zell Miller (D-GA) and Representative James

Traficant (D-OH) as behaving much more conservatively in their final terms, and President

Bush as less extreme in the first two years of his presidency. Also, our estimates place

Common Space extremists like Russ Feingold and Ron Paul closer to the center of their

parties. Finally, our estimates are able to discriminate more among Democrats than

Common Space scores, as can be seen by the greater vertical spread in Democrats’ ideal

points.

Our estimates are also particularly useful for determining if and in what direction

legislators’ ideal points are moving. Unlike other methods of making ideal points

comparable across time, which can only claim to capture relative changes, our technique of

holding the agenda fixed means that differences in legislators’ ideal points from one period

to another can be interpreted as absolute changes. Figure 6 plots the ideal point of each

member of the 107th Congress against their ideal point in the 108th Congress for those who

served in both sessions. The 45 degree line divides the members who became more

14

conservative (above the line) from those who became more liberal (below the line). Solid

circles indicate that the posterior probability that the member became more conservative

(or more liberal) is greater than 95 percent. Note that nearly all Republicans moved to the

right, but that Democrats exhibit mixed behavior, with moderate Democrats slightly more

likely to have moved to the right, but more liberal Democrats mostly moving to the left.

These patterns are not surprising given the electoral context: The 2002 midterm elections

provided the Republicans with a rare midterm victory for the president’s party, in which

they picked up eight House seats, two Senate seats and regained partisan control of the

Senate. This victory, combined with Bush’s rising popularity and post-9/11 hawkishness

may have emboldened Republicans—both moderates and extremists—to behave more

conservatively. Indeed, Bush himself shows the greatest change between the 107th and

108th Congresses.

It is also clear from Figure 5 that movements in the ideal points of continuing

members between these two sessions are contributing to the long-term phenomenon of

party polarization, with more Republicans above the 45 degree line and more Democrats

below. In this time period, Republicans are the major contributors to polarization—the

mean posterior probability that members serving in both congresses became more

extreme in the second session is higher for Republicans than Democrats (0.91 versus

0.63)—and regardless of party, the effect is more pronounced in the Senate, where the

probability that members became more extreme is 0.97 for Republicans and 0.88 for

Democrats (versus 0.90 and 0.56 for House Republicans and Democrats, respectively).

Two noteworthy Senators are highlighted in Figure 5. Senator Zell Miller (D-GA),

already the most conservative Democrat in the 107th Congress, moved even further to the

15

right in the 108th Congress (the same session in which he endorsed Bush at the Republican

Convention). He retired at the end of the 108th Congress. Senator Jim Jeffords (I-VT),

whose party switch from the Republicans caused them to lose control of the Senate during

the 107th Congress, is shown to have moved even further to the left in the 108th Senate,

possibly to position himself for the 2006 campaign where he was contemplating running as

a Democrat or at least a liberal enough Independent to preclude a Democratic challenge.

Our estimates can also tell us about the behavior of members moving from the

House to the Senate. Only three members did so between these two sessions: Saxby

Chambliss (R-GA), Lindsey Graham (R-SC), and John Sununu (R-NH). The posterior

probability that these members became more conservative in the Senate is 0.675, 1.000,

and 0.993, respectively. This is puzzling because we might expect members to behave

more moderately as Senators than as House members, given that states are usually more

politically heterogeneous than congressional districts; however, in the case of these three

members, only Lindsey Graham’s state was significantly more moderate than his district

(57% of South Carolinians voted for Bush in 2000, versus 63% in his district).

Furthermore, electoral dynamics may trump changes in constituency: none of the

candidates faced a tough primary challenge in 2002, but all faced strong general election

opponents, giving incentives to the candidates to moderate their behavior in the 107th

House. Given the small number of chamber-switchers and the limited time series for each

member, it is difficult to draw any broad conclusions.

Our estimates also support the claim that senators moderate their behavior prior to

elections, and upon winning, return to more extreme positions (for additional evidence, see

Treier 2010b). Figure 7 plots the ideal points of senators who were up for reelection in

16

2002 and won. The horizontal axis is their ideal point in the fifth and sixth year of their

pre-reelection term, and the vertical axis is their ideal point in the first and second year of

their new term. Every Republicans’ ideal point moves substantially to the right. The three

smallest shifters were Senators Susan Collins (R-ME), Pat Roberts (R-KS), who faced no

Democratic challenger, and Chuck Hagel (R-NE), whose Democratic opponent was so poor

he could not afford the filing fee (Barone and Cohen 2004). In the latter two cases, the lack

of electoral competition could mean they did not need to moderate their behavior prior to

the election. This could also be the case for Senator Jack Reed (D-RI), one of the two

Democrats who did not move to the left. His general election competition was a casino pit

manager who raised no money (Barone and Cohen 2004). The other stationary Democrat

was Sen. Joe Biden (D-DE). Although senators not up for reelection also moved toward

their parties’ extremes, members reelected in 2002 moved more than those originally

elected in 2000 (0.169 average movement compared with 0.108), and more than those

seeking reelection in 2004 (whose average movement was 0.103 toward their parties’

extremes).6

Finally, Table 1 provides a list of the members who moved the farthest to the right

or to the left, along with possible explanations for their behavior changes, as gleaned from

descriptions of members’ careers and electoral contests in the Almanac of American Politics

(Barone and Cohen 2002, Barone and Cohen 2004). Primary competition, ambitions for

higher office, redistricting concerns, retirements, and first reelection attempts seem to be

6 We exclude retirees in calculating the average movement. Those retiring in 2004 moved 0.089 toward their

parties’ extremes. Excluding Sen. Zell Miller (D-GA), who bucked the trend by moving substantially to the

right, retirees moved an average of 0.153 toward their parties’ extremes.

17

important factors in explaining shifting preferences, but more work remains to be done in

explaining more generally which factors motivate behavior change and when.

Conclusion and Future Research

Many of the ways that political scientists use ideal point estimates involve making

comparisons across time and chambers. To date, the techniques for achieving

comparability have focused on using legislators as anchors, a practice which artificially

limits the movement of legislators’ ideal points and can only give insight into the

movement of ideal points relative to other members. Our approach does away with these

anchors. Instead, we make the agendas comparable by exploiting a feature of

cosponsorship data: bills introduced in identical form in both chambers or in multiple

sessions that receive a floor vote in at least one chamber and session. Using data from the

107th and 108th Congresses, we demonstrate that the assumptions are plausible, the data

collection is doable, the estimation is simple, and the results are believable.

The insights we have gained from observing just one period of change encourages

us to believe that future work on this project will bear much fruit. Extending the scope of

this analysis by identifying identically introduced bills across more sessions of Congress is

a next step which would allow us to observe the movement of legislators’ ideal points over

a longer period of time. Also, by merging our estimates with data on legislators’ retirement

decisions, electoral vulnerability, primary challenges, and redistricting history, we can

make sense of the patterns we observe and even engage in theory testing.

We are looking forward to learning what our approach can tell us about the big

picture of congressional politics. Ideal point estimates have given us a general

18

understanding of how party allegiances and issues have shifted over the course of

American history, but we still view these phenomena through the confusing prism of

shifting scales, relative changes and collapsing dimensions. We eagerly anticipate being

able to view congressional politics played out on a stable stage, rather than floating

between drifting anchors.

19

References

Adler, E. Scott and John Wilkerson. 2001-2004. Congressional Bills Project. NSF 00880066

and 00880061.

Aleman, Eduardo, Ernesto Calvo, Mark P. Jones and Noah Kaplan. 2009. “Comparing

Cosponsorship and Roll-Call Ideal Points.” Legislative Studies Quarterly 34(1):87-

116.

Ansolabehere, Stephen, James M. Snyder, Jr., and Charles Stewart, III. 2001. “Candidate

Positioning in U.S. House Elections.” American Journal of Political Science 45(1):136-

159.

Bailey, Michael A. 2007. “Comparable Preference Estimates across Time and Institutions for

the Court, Congress and Presidency.” American Journal of Political Science 51 (3):

433-448.

Barone, Michael and Richard E. Cohen. 2004. Almanac of American Politics, 2004. National

Journal Group.

Barone, Michael and Richard E. Cohen. 2006. Almanac of American Politics, 2006. National

Journal Group.

Bernhard, William and Tracy Sulkin. 2010. “Commitment and Consequences: Reneging on

Cosponsorship Pledges in the U.S. House.” Paper presented at the annual meeting of

the Midwest Political Science Association, April 2010.

Clinton, Joshua, Simon Jackman, and Douglas Rivers. 2004. “The Statistical Analysis of Roll

Call Data.” American Political Science Review 98(2):355-370.

Desposato, Scott, Matt Kearney, and Brian Crisp. 2008. “Using Cosponsorship to Estimate

20

Ideal Points.” Working paper.

Ensley, Michael J., Michael W. Tofias, and Scott de Marchi. 2010. “Are These Boots Made for

Walking? Ideological Change among U.S. House Members.” Paper presented at the

Annual Meeting of the Midwest Political Science Association 2010.

Groseclose, Tim, Steven D. Levitt and James M. Snyder, Jr. 1999. “Comparing Interest Group

Scores across Time and Chambers: Adjusted ADA Scores for the U.S. Courts”

American Political Science Review 93(1):33-50.

Heckman, James J. and James M. Snyder, Jr. 1997. “Linear Probability Models of the Demand

for Attributes with an Empirical Application to Estimating the Preferences of

Legislators.” Rand Journal of Economics 28:S142-S189.

Harward, Brian M. and Kenneth W. Moffett. 2010. “The Calculus of Cosponsorship in the

U.S. Senate.” Legislative Studies Quarterly 35 (1): 117-143.

Kessler, Daniel and Keith Krehbiel. 1996. “Dynamics of Cosponsorship.” American Political

Science Review 90(3):555-566.

Krehbiel, Keith. 1995. “Cosponsors and Wafflers from A to Z.” American Journal of Political

Science 39(4):906-923.

Martin, Andrew D. and Kevin M. Quinn. 2002. “Dynamic Ideal Point Estimation via Markov

Chain Monte Carlo for the U.S. Supreme Court, 1953-1999.” Political Analysis

10(2):134-153.

McCarty, Nolan, Keith T. Poole and Howard Rosenthal. 2001. “The Hunt for Party Discipline

in Congress.” American Political Science Review 95(3):672-687.

Peress, Michael. 2009. “Estimating Proposal and Status Quo Locations Using Voting and

Cosponsorship Data.” Working paper.

21

Poole, Keith T. and Howard Rosenthal. 1997. Congress: A Political-Economic History of Roll

Call Voting. New York: Oxford University Press.

Talbert, Jeffery C. and Matthew Potoski. 2002. “Setting the Legislative Agenda: The

Dimensional Structure of Bill Cosponsoring and Floor Voting.” Journal of Politics

64(3):864-891.

Treier, Shawn. 2010a. “Comparing Ideal Points across Institutions and Time.” Working

paper.

Treier, Shawn. 2010b. “Ideal Point Estimation Using Overlapping Constraints in the

Senate.” Working paper.

22

Figure 1 Changes in Legislators' Ideal Points with a Linear Trend

0 2 4 6 8 10

Idea

l Po

int

Session

Redistriced Memberor MemberSwitching fromHouse to Senate

Member w/ aPrimary Challenge

Retiring Member

Senator FacingPeriodic Reelection

23

Figure 2 Changes in Legislators' Ideal Points with a Polynomial Trend of Degree Five

0 2 4 6 8 10

Idea

l Po

int

Session

Redistriced Memberor MemberSwitching fromHouse to Senate

Member w/ aPrimary Challenge

Retiring Member

Senator FacingPeriodic Reelection

24

Figure 3 Representation of Roll Call/Cosponsorship Matrix

25

Figure 4 Estimated Ideal Points (Posterior Means)

-1.0 -0.5 0.0 0.5 1.0

01

23

4

107th House

Ideal points

Density

-1.0 -0.5 0.0 0.5 1.0

01

23

4

Density

Democrats

Republicans

Median

Democrats

Republicans

Median

-1.0 -0.5 0.0 0.5 1.0

0.6

0.8

1.0

1.2

1.4

01

Democrats

Republicans

Median

-1.0 -0.5 0.0 0.5 1.0

0.6

0.8

1.0

1.2

1.4

0

1

Democrats

Republicans

Median

26

Figure 5 Comparison with Common Space Scores

Sen. Zell Miller 108

Rep. Ron Paul

Rep. James Traficant 107

Sen. Russ Feingold

Pres. Bush 107

27

Figure 6 Movement in Legislators' Ideal Points across Congresses

Pres. George W. Bush

Sen. Zell Miller

Sen. Jim Jeffords (Ind.)

28

Figure 7 Movement of Senators' Ideal Points after Reelection

Pat Roberts

Chuck Hagel

Susan Collins

Joe Biden

Jack Reed

29

Table 1 Members Who Moved the Most Between the 107th and 108th Sessions

Biggest Movers to the Right

Avg. Change

Initial Ideal Point Explanation for Movement

1 Pres. George W. Bush (R) 0.446 0.627 strengthened by Republican midterm victories and post-9/11 popularity

2 Rep. Nick Smith (R-MI) 0.399 0.370 retired in 2004 (announced decision in December 2002)

3 Rep. Jeff Miller (R-FL) 0.389 0.603 2002 was first election after winning seat in a 2001 special election

4 Rep. Bob Filner (D-CA) 0.384 -1.353 redistricting threat and possible Latino primary challenge in maj.-min. district in 2002

5 Sen. Zell Miller (D-GA) 0.362 0.110 retired in 2004 (announced decision in January 2003)

6 Rep. Peter DeFazio (D-OR) 0.341 -0.786 considered running for the Senate in 2002 7 Sen. Jeff Sessions (R-AL) 0.340 0.390 reelected for the first time in 2002 after a close

race in 1996 8 Rep. Nathan Deal (R-GA) 0.340 0.600

9 Rep. Johnny Isakson (R-GA) 0.325 0.448 ran for the Senate in 2004, facing competition from conservatives in the primary

10 Rep. Ernest Istook (R-OK) 0.320 0.368 redistricting threat in 2002; considered running for the Senate in 2004

Biggest Movers to the Left Avg.

Change

Initial Ideal Point Explanation for Movement

1 Rep. Lynn Woolsey (D-CA) 0.287 -0.743 faced a primary challenge from a moderate in 2002 2 Rep. John Lewis (D-GA) 0.287 -0.713

3 Sen. Tom Harkin (D-IA) 0.268 -0.278 faced a strong general election challenger in 2002 4 Sen. John Kerry (D-MA) 0.264 -0.257 ran for president in 2004 (missed a lot of votes in

the 108th) 5 Rep. Jesse Jackson, Jr. (D-IL) 0.256 -0.598

6 Sen. Jim Jeffords (I-VT) 0.247 0.085 post-party-switch positioning to run as a Democrat or liberal independent in 2006

7 Sen. John Edwards (D-NC) 0.243 -0.154 ran for president in 2004 (missed a lot of votes in the 108th)

8 Sen. Dick Durbin (D-IL) 0.223 -0.296 reelected for the first time in 2002 9 Sen. Robert Byrd (D-WV) 0.217 -0.125

10 Rep. Lois Capps (D-CA) 0.209 -0.418 faced a well-financed general election opponent in 2002