legislative systems with absolute party discipline: implications for the agency theory approach to...

20
Public Choice 97: 121–140, 1998. 121 c 1998 Kluwer Academic Publishers. Printed in the Netherlands. Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link NEIL LONGLEY Faculty of Administration, University of Regina, Regina, Saskatchewan, S4S OA2, Canada Accepted 22 May 1996 Abstract. This paper examines how the presence of absolute party discipline forces one to re-examine some of the issues surrounding the constituent-legislator link. With absolute party discipline, slack at the individual district level is determined by the policy choices of a political party, rather than by the choices of the individual legislator. This party discipline not only has implications for the representational effectiveness of individual legislators, but also results in the terms “slack” and “shirking” no longer necessarily being synonymous. The empirical work shows that Canadian political parties engaged in wide scale shirking on the 1988 U.S.–Canada Free Trade Agreement. 1. Introduction Previous research examining the application of agency theory to the consti- tuent–legislator relationship has attempted to determine the degree of slack that exists in such relationships. 1 Slack has been seen to be synonymous with legislator “shirking”; a situation where a legislator’s vote on an issue is deter- mined, at least in part, by the legislator’s personal ideological preferences, rather than being exclusively determined by the interests of the legislator’s constituents. This research, however, assumes a legislative system in which the legislator has the power to independently choose his or her voting position on any given issue. While such an assumption is valid with respect to the U.S. Congress, 2 it is not necessarily relevant to other legislative systems. For example, in the Canadian Parliament, party discipline is so strong that individual Members of Parliament (MPs) are almost always forced to vote in accordance with the position adopted by their party. The paper has three objectives. First, it will develop a theory as to the man- ner in which agency theory can be applied to the Canadian Parliament. It will The author would like to thank Rod Fort, Bill Hallagan, and Stuart Thiel for their helpful comments on earlier drafts of this paper. Any errors are mine.

Upload: neil-longley

Post on 29-Jul-2016

213 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

Public Choice97: 121–140, 1998. 121c 1998Kluwer Academic Publishers. Printed in the Netherlands.

Legislative systems with absolute party discipline: Implicationsfor the agency theory approach to the constituent-legislator link�

NEIL LONGLEYFaculty of Administration, University of Regina, Regina, Saskatchewan, S4S OA2, Canada

Accepted 22 May 1996

Abstract. This paper examines how the presence of absolute party discipline forces one tore-examine some of the issues surrounding the constituent-legislator link. With absolute partydiscipline, slack at the individual district level is determined by the policy choices of a politicalparty, rather than by the choices of the individual legislator. This party discipline not only hasimplications for the representational effectiveness of individual legislators, but also results inthe terms “slack” and “shirking” no longer necessarily being synonymous. The empirical workshows that Canadian political parties engaged in wide scale shirking on the 1988 U.S.–CanadaFree Trade Agreement.

1. Introduction

Previous research examining the application of agency theory to the consti-tuent–legislator relationship has attempted to determine the degree of slackthat exists in such relationships.1 Slack has been seen to be synonymous withlegislator “shirking”; a situation where a legislator’s vote on an issue is deter-mined, at least in part, by the legislator’s personal ideological preferences,rather than being exclusively determined by the interests of the legislator’sconstituents.

This research, however, assumes a legislative system in which the legislatorhas the power to independently choose his or her voting position on any givenissue. While such an assumption is valid with respect to the U.S. Congress,2

it is not necessarily relevant to other legislative systems. For example, in theCanadian Parliament, party discipline is so strong that individual Membersof Parliament (MPs) are almost always forced to vote in accordance with theposition adopted by their party.

The paper has three objectives. First, it will develop a theory as to the man-ner in which agency theory can be applied to the Canadian Parliament. It will

� The author would like to thank Rod Fort, Bill Hallagan, and Stuart Thiel for their helpfulcomments on earlier drafts of this paper. Any errors are mine.

Page 2: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

122

show how the Canadian parliamentary tradition of extreme party disciplineforces one to modify how one views the constituent-legislator relationship.Second, it will also present a theory as to why a system of complete partydiscipline implies that the terms “slack” and “shirking” are no longer synony-mous. Third, the paper will then test these theories by empirically modellingthe constituent-legislator links within the context of the Canadian parliamen-tary system.

2. Defining “constituent interests” under systems of complete partydiscipline

Under the Canadian parliamentary system, individual legislators do not havethe power to independently determine their position on an issue. Even wheretheir party’s position on an issue clearly conflicts with the interests of theirconstituents, or conflicts with their own personal ideological preferences, theindividual MP must still vote in accordance with the party position.3

Dyck (1993) attributes this strong party discipline to a number of factors.The most important factor is the Canadian parliamentary tradition that thedefeat of a government bill is interpreted as an indication of non-confidencein the government. When such indications of non-confidence occur, the gov-ernment is expected to resign and call an immediate election. Thus, becausethe fate of the government is at stake in essentially every parliamentary vote,parties are able to exert tremendous pressure on their MPs to vote the partyline.

This natural incentive to vote the party line is further enhanced by the factthat the incentive structure facing Canadian MPs is largely controlled by thepolitical parties to which they belong. Only loyal party members can expectperks (such as attractive travel opportunities) and internal promotions,4 whiledisloyal members face possible expulsion from the party caucus.5

Thus, in Canada, once an individual makes a decision to join a party, thatindividual gives up the right to independently determine his or her policyposition; the individual, in effect, becomes a “policy-taker” from the party.This is much different from the U.S. Congress, where Senators and Represen-tatives do have the power to independently assess each policy initiative, andto adjust their voting behavior in accordance with their constituents’ interests.

In a parliamentary system with absolute party discipline, no longer is theindividual legislator the “agent” in the principal-agent problem; rather, thelegislator’s party is now the agent.6 It is no longer meaningful to ask whethera legislator is shirking, since a legislator has no control over his own votingbehavior. Instead, the essential question is whetherpartiesare shirking.

Page 3: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

123

To measure party shirking, a party’s “constituent interests” must first bedefined. The group of constituents a party represents is not stable and fixed,but rather changes with each election. If a party wins a particular seat inParliament (i.e., the party’s candidate in the district is elected to Parliament),the individuals in that district become the party’s constituents, at least untilthe next election.

In defining a party’s “constituent interests”, the median voter concept usedin U.S. models continues to be generally employed. In U.S. models, the pref-erences of an individual constituency with respect to a policy issue are takento be the preferences of the median voter in that constituency. This papersimply takes this proposition a step further.

A party in Canada is essentially faced with a number of median voters. Foreach constituency a party represents, that constituency will have a medianvoter; a voter deemed to reflect the interests of the entire constituency. How-ever, what is needed is a singular measure of constituent interests for a party.This is obtained by examining the preferences of the median voter from eachconstituency a party represents, and then finding the median of these val-ues. The constituency this median voter represents is thus deemed the party’s“median constituency”.

3. Differentiating between slack and shirking

The nature of the parliamentary system forces one to adopt separate defini-tions for the terms “slack” and “shirking”; unlike the U.S. system, the termsare no longer interchangeable.

Shirking can be generally defined to exist if the policy positions of politicalagents are not strictly determined by the configuration of their constituents’interests.

As discussed above, in a parliamentary system, it is the political parties,and not the individual legislators, who are the agents. For these parties, theterm “constituent interests” can be defined as the interests of constituents inthe median constituency of the party. Thus, only where the party’s position isinconsistent with the preferences of its median constituency, can the party beviewed as shirking.

Consider a situation where a party develops a non-shirking, national poli-cy on an issue, and then, as usual, imposes this policy position on each of itsMPs. For the median constituency, the policy position taken will be highlyrepresentative of constituent interests. For other constituencies, however, thesame policy position may be largely unrepresentative of constituent interests.Clearly, the greater the heterogeneity in the economic and ideological inter-ests amongst various constituencies with respect to a certain policy issue,

Page 4: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

124

the greater the slack (total constituent preference violation) that results fromthe imposition on the legislators of those constituencies of a common policyposition.

Thus, even if parties do not shirk, constituent-legislator slack will not nec-essarily be eliminated. Only if the interests of all of a party’s constituencieswere identical to the interests of the median constituency, would there be noslack. Slack will be greatest for those constituencies whose economic andideological interests on an issue are furthest away from the interests of themedian constituency of the party.

Thus, two types of constituent preference violations must be identified. Thefirst, common to both parliamentary and republic systems, results from legis-lator shirking. In Congress, legislators shirk by not voting in accordance withthe preferences of the median voter in their constituency. In Canada’s Parlia-ment, parties shirk by not adopting positions in accordance with the pref-erences of their median constituency. However, in a parliamentary system,a second type of constituent preference violations are commonly present.These are the constituent preference violations that result from the systemitself; in other words, from the imposition of party discipline on individuallegislators.

4. The effects of party discipline on empirical modelling

The empirical model of this paper employs a basic methodology similar tothat of Kalt and Zupan (1984). Although the weaknesses of such a method-ology are well-documented (see, for example, Goff and Grier, 1993), validcomparisons between the U.S. and Canadian systems requires the Canadi-an modelling to be generally consistent with previous research on the U.S.Congress.

Following Kalt and Zupan, a two-step process is used. The first step is todetermine what portion of each legislator’s observed voting behavior reflectsconstituent interests, and what portion are constituent preference violations.The variable PROTRADE represents observed voting behavior and is definedas:

PROTRADE= ln r+:5n�r+:5

where r= the number of times the legislator voted a pro-trade positionn = the number of times the legislator voted on trade-related issues

PROTRADE is regressed on a series of variables measuring constituentinterests. The predicted values and residuals from this regression are thenused in the second step in the process. In this step, a logit model is usedto regress each legislator’s vote on the U.S.–Canada Free Trade Agreement

Page 5: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

125

(FTA) on a series of variables relating to constituent interests with respect tothe FTA, and on the residuals and predicted values from the step 1 regression.

In the step 1 regression, the predicted value of PROTRADE (representingconstituent interests) is denoted by PROHAT, while the regression residual isdenoted by RESIDHAT. Equation (1) more formally defines RESIDHAT as:

RESIDHAT= PROTRADE� PROHAT (1)

Just as in the Kalt and Zupan model, RESIDHAT measures the total con-stituent preference violation for that constituency. Positive values of RESID-HAT indicate the legislator’s voting behavior was less protectionist (morepro-trade) than constituents desired, and negative values of RESIDHAT indi-cate the legislator’s voting behavior was more protectionist (less pro-trade)than constituents desired. However, unlike the Kalt and Zupan interpretation,these residuals do not necessarily represent the shirking of individual legis-lators, since individual legislators, being bound by party discipline, have noability to independently choose their own voting position on an issue.

Do, then, the residuals represent shirking by political parties, since the par-ty, and not the individual legislator, is the decision unit in the Canadian Par-liament? To answer this question, recall that a party in a parliamentary systemshirks if it does not set policy in accordance with the preferences of its medi-an constituency. A party’s median constituency, denoted by MEDPROHAT,is found by examining the PROHAT values of each legislator from that party.The median of these PROHAT values is considered to designate the interestsof the party’s median constituency.

However, as shown earlier, given that parties represent a series of non-homogeneous constituencies, selecting the non-shirking policy position (i.e.,MEDPROHAT), and imposing it on all legislators from the party, will notreduce RESIDHAT to zero in any constituency but the median constituen-cy. In fact, the further a constituency’s preferences are away from the party’smedian constituency, the less the adopted policy position will reflect that con-stituency’s preferences. Thus, not all of RESIDHAT for any given constituen-cy is due to party shirking; some is due to constituent preference violationsthat inevitably arise under systems of rigid party discipline.

Thus, if the Canadian results are to be meaningful, RESIDHAT must beseparated into two components. The first component, called SHIRK, is definedin equation (2). It is this component of RESIDHAT which measures shirkingin the constituent-legislator relationship.

SHIRK = PROTRADE�MEDPROHAT (2)

SHIRK measures the difference between the legislator’s actual voting be-havior and the median predicted value of PROTRADE for the legislator’s

Page 6: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

126

party. For each legislator, their actual voting behavior, PROTRADE, is con-trolled by their party. From equation (2), it can be seen that if the party adoptspolicy positions such that PROTRADE is set equal to MEDPROHAT, noshirking will exist.

If, however, the party adopts policy positions such that PROTRADE isgreater than MEDPROHAT, SHIRK will be positive, indicating the partyis adopting policy positions that are more pro-trade (less protectionist) thantheir median constituency prefers. Finally, if the party adopts policy positionssuch that PROTRADE is less than MEDPROHAT, SHIRK will be negative,indicating the party is adopting policy positions that are more protectionistthan their median constituency prefers.

Once this shirking portion of the RESIDHAT has been found, the remain-der of the residual must, by definition, result from the rigidities of the partydiscipline system. This remaining portion of the residual, defined as DISCP,is caused by the fact that all members of a party must exhibit identical votingbehavior on all issues. Alternatively, DISCP can be defined as in equation (3).

DISCP= MEDPROHAT� PROHAT (3)

DISCP measures, for each legislator of a party, the difference between thepredicted value of the party’s median constituency, MEDPROHAT, and thepredicted value for a given constituency (as given by the regression), PRO-HAT.

Notice that a political party cannot influence the magnitude of DISCP,since the party cannot influence either MEDPROHAT or PROHAT. Both aredetermined by constituent interests; the former by the interests of the party’smedian constituency, and the latter by the interests of the individual con-stituency in question.

From equation (3), it can be seen that if the interests of a particular con-stituency are identical to the interests of the party’s median constituency,DISCP is zero. This implies that, for the particular constituency in ques-tion, their is no constituent preference violation due to the effects of the partydiscipline system.

If, however, the interests of a particular constituency are more pro-trade(less protectionist) than are the interests of the party’s median constituency,DISCP is negative. Conversely, where the interests of a particular constituen-cy are less pro-trade (more protectionist) than are the interests of the par-ty’s median constituency, DISCP is positive. In general, constituencies whosepreferences are further away from the party’s median constituency will havehigher absolute values of DISCP.

To summarize these methodological complexities associated with the Cana-dian model, notice that the Kalt and Zupan residual, RESIDHAT, is still

Page 7: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

127

Table 1. SHIRK and DISCP scenarios

PROTRADE PROHAT MEDPROHAT RESIDHAT SHIRK DISCP

A 3 1 2 2 1 1

B 3 1 3 2 0 2

C 2 3 1 –1 1 –2

employed, it is just that it is disaggregated into two parts. If one sums SHIRKand DISCP, the result is as given in Equation 4.

SHIRK+ DISCP = (PROTRADE�MEDPROHAT)+

(MEDPROHAT� PROHAT)

= PROTRADE� PROHAT

= RESIDHAT

(4)

To more fully explain the situation, consider the scenarios provided inTable 1. Each cell of the matrix represents a value for the correspondingvariable. The first three columns are given, while the last three are calculatedusing the above formulas. Each scenario is independent of the others.

Under scenario A, the legislator’s actual voting behavior (determined, ofcourse, by the party) is two units more pro-trade than what the legislator’sconstituents desire. One unit of this is due to party shirking. The party hasadopted a more pro-trade position than that of its median constituency. Theother unit is due to the effects of party discipline. The constituency in ques-tion prefers a more protectionist policy than does the party’s median con-stituency.

Under scenario B, the legislator’s actual voting behavior is again two unitsmore pro-trade than what the legislator’s constituents desire. However, sincethe party has adopted a policy position that is identical to the interests ofits median constituency, no shirking is occurring. The entire two units ofconstituent preference violation are due to the effects of party discipline. Theconstituency in question prefers a more protectionist policy than does theparty’s median constituency.

Under scenario C, the legislator, under the party’s direction, has adopteda more protectionist position than what the legislator’s constituents prefer.Hence, RESIDHAT is negative. The party is shirking, but it is shirking byadopting a policy position that is too pro-trade, rather than too protectionist!This is because the party has adopted a policy position that is more pro-tradethan what its median constituency prefers. However, RESIDHAT is still nega-tive for the constituency in question because the constituency has preferencesthat are much more pro-trade than does the median constituency. Thus, thenegative value of DISCP more than offsets the positive value of SHIRK. This

Page 8: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

128

last scenario shows that, in some cases, the total constituent preference vio-lation for a given constituency may be in an opposite direction to the partyshirking.

5. The empirical model

5.1. Step 1 – Measuring party shirking

As discussed in the previous section, PROTRADE represents each legisla-tor’s cumulative voting record on a wide range of trade-related bills.7 Thevariable was constructed from an analysis of fifteen House of Commonsvotes. All votes occurred during the 1985–1988 time period, and all relateto issues of international trade or foreign investment.

PROTRADE is then regressed on a series of independent variables designedto measure constituent interests. These variables are defined in Appendix A.

The first six variables measure the demographic characteristics of a con-stituency. Their inclusion implicitly assumes that demographic differencesamong constituencies proxy differences in policy preferences among thoseconstituencies. The demographic variables are included as control variables,and, as such, no expectations are placed on the signs of these variables.

The fourth variable, NDP, requires some additional explanation. NDP isintended as a broad-based measure of a constituency’s ideological prefer-ence. The NDP is a social democratic party that is driven by ideology ratherthan pragmatism (Jackson, Jackson, and Baxter-Moore, 1986), and has nevercome close to forming a government at the federal level. Thus, no voter couldrealistically expect any future economic payoff from voting for the NDP. Anyvotes going to the party, therefore, are viewed solely as attempts to registerideological preferences. Because of its broad-based nature, the inclusion ofNDP can also capture the effects of any other demographic variables thatmay be omitted from the regression.

Variables seven through twelve relate more specifically to a constituen-cy’s interests with respect to trade policy. UNION, ESTAB, and LGESTABmeasure the opposing influences of groups that have broad-based interests inthe trade policy area. These groups are defined as broad-based because theirsupport is derived from workers (in the case of UNION) and firms (in thecase of ESTAB and LGESTAB) from a wide variety of industries. UNION isexpected to have a negative sign, while ESTAB and LGESTAB are expectedto have positive signs.

With respect to UNION, labor unions are by far the most influential ofthe broad-based, pro-protectionist groups. Unions, with their concern overthe lost jobs and lower wages that may result from increased foreign com-petition, have consistently taken protectionist positions on almost all recent

Page 9: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

129

trade issues. While there are many other broad-based groups (such as femi-nist groups, anti-poverty groups, and environmentalists) that also have pro-tectionist leanings, these groups have tended to focus their interests on issuesother than trade policy.

The influence of the broad-based anti-protection groups is captured byESTAB and LGESTAB. ESTAB is constructed to proxy the interests of busi-ness groups such as the Chamber of Commerce, whose membership is usual-ly dominated by small to medium-sized firms. LGESTAB, on the other hand,is constructed to proxy the interests of those business groups whose mem-bership is more typically dominated by large firms. LGESTAB is includedas a separate variable because of the particular importance in Canada of bigbusiness in lobbying for freer trade. Since its formation in 1976, the Busi-ness Council on National Issues, which is comprised of the CEOs of the 150largest corporations in Canada, has been at the forefront of the free tradelobby.

EXDEP and IMPSENS measure the opposing influence of groups that havemore specific interests in trade policy issues. The sign on EXDEP is expectedto be positive, while the sign on IMPSENS should be negative. EXDEP cap-tures the influence of firms and workers in industries that are export depen-dent. Industries are classified as export dependent if they meet at least oneof the following three criteria: i) they rank among the top ten in net exports,ii) they rank among the top ten in their ratio of exports to total production,iii) they have been active lobbyists supporting freer trade. In practice, mostindustries that are included in EXDEP meet at least two of the above criteria,and many industries meet all three criteria.

IMPSENS captures the influence of firms and workers in industries that areimport sensitive. Industries are classified as import sensitive if i) during thelast fifteen years, they have been the beneficiaries of any type of product-specific protectionist legislation before Parliament, or, ii) they have tariffrates (including the tariff equivalent of non-tariff barriers) that exceed tenpercent.

The expected sign on IMPORTS is less certain. IMPORTS should be pos-itive to the extent that the constituency’s users of the imports (i.e., industrialusers, retailers, etc.) will oppose any policies which favor higher-cost domes-tic products over these lower-cost imports. However, IMPORTS should benegative to the extent that these imports compete directly with the productsof other firms in the constituency. Thus, the actual sign on IMPORTS willdepend on the relative strengths of these two opposing groups. The results ofthe regression are reported in Table 2. A weighted generalized least squarestechnique is employed.

Page 10: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

130

Table 2. Canadian PROTRADE regressionresults

Independent variable Coefficient t-statistic

1. INCOME .084 2.87�

2. URBAN .516 .69

3. EDUC –.572 –2.52�

4. NDP –9.278 –7.29�

5. AGE .049 1.13

6. WHCOL .401 .15

7. UNION –5.218 –2.93�

8. EXDEP .314 .09

9. IMPSENS –5.438 –1.60

10. IMPORTS .008 E-2 1.31

11. ESTAB –16.935 –.65

12. LGESTAB –16.905 –1.90

R2 = .38 R2 = .35

�Indicates significance at the 5% level.

Table 3. Shirking by party

# of pro-trade votes

Zero

shirking Actual Shirking

Liberals 8 1 –7

NDP 7 0 –7

P.C. 11 15 4

The results indicate that all trade-related variables are of the correct signexcept for ESTAB and LGESTAB. Of the trade-related variables, only UNIONis statistically significant.

The results, in themselves, provide minimal direct information. However,they do provide some insight into the extent and direction of party shirking.Table 3 shows, for each party, the number of votes (out of fifteen) in whichthe party adopted a pro-trade position, and the number of times it should haveadopted such a position if it were seeking to eliminate all shirking.

The results indicate that there was considerable shirking in the voting pat-terns of the three parties. The Liberals and NDP both voted a protectionistposition seven more times than what their median constituency preferred,while the Progressive Conservatives voted a pro-trade position four moretimes than what their median constituency preferred.

Page 11: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

131

The above results address a potential criticism of the median constituencytheory. It may be argued by some that, in a three party system, parties may notbe concerned about the preferences of their median constituency, but ratherwith the preferences of their marginal constituency.

Consider Canada’s three party system. The Progressive Conservatives (P.C.)are on the right of the political spectrum, the Liberals are in the center of thespectrum, and the New Democrats (NDP) are on the left of the spectrum.It may be argued, for example, that the Progressive Conservatives, knowingthat they have assured support of those constituencies on the right, wouldadopt policies to capture constituencies to the left of their party’s medianconstituency. This would presumably take some constituencies away fromthe centrist Liberals. Similarly, the New Democrats may elect to adopt poli-cies to the right of their median constituency. Again, they are assured of thoseconstituencies on the left, and by moving policies to the right, may take sup-port from the Liberals.

However, the results indicate that the Progressive Conservatives and NewDemocrats adopted policies in the opposite direction to what one would pre-dict. The Progressive Conservatives adopted policies that were to the rightof their median constituency, while the New Democrats adopted policies tothe left of their median constituency. Thus, it cannot be argued that theseparties were attempting to capture their marginal constituencies, since theirmarginal constituencies were in the opposite direction to which the party’spolicies were moving. Therefore, if, in fact, the zero-shirking policy positionrequires the party to adopt policies in accordance with the preferences of theparty’s marginal constituency, rather than its median constituency, the aboveanalysis underestimates the presence of shirking, rather than overestimates it.

5.2. Step 2 – Determining the importance of shirking in FTA voting

The second step in the process is to determine whether Canadian politicalparties engaged in shirking behavior on the FTA voting issue.

The dependent variable, FTAVOTE, is set equal to one if the legislator vot-ed in favor of the FTA, and is set equal to zero if the legislator voted againstthe FTA. The independent variables represent four sources of influence: con-stituent ideological interests, party shirking, constituent preference violationsdue to party discipline, and constituent economic interests.

The first three sources of influence are all derived from the step 1 regres-sion. Constituent ideological interests, CFIT, are represented by the predict-ed values of PROTRADE. The next two sources of influence, party shirk-ing (defined as SHIRK) and constituent preference violations due to partydiscipline (defined as DISCP), are simply the disaggregated components of

Page 12: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

132

the residual from step 1. Their exact derivation was discussed earlier. Theremaining source of influence, constituent economic interests, is measuredby the set of variables described in Appendix B.

Organized labor and business groups were at the forefront of the FTAdebate; the former opposed the FTA, while the latter supported it. Laborinterests are captured by UNION and UNCONT, while business interestsare captured by ESTAB, LGESTAB, and BUSCONT. However, lobbying bybroad-based organizations was not restricted to these two groups. In partic-ular, a broad-based coalition of social, environmental, religious, and ethnicgroups arose in Canada to actively oppose the agreement. OPPOSE capturesthe combined influence of these groups, and is expected to have a negativesign.

The remaining variables are more directly trade-related. EXDEPUS andIMPORTUS are control variables designed to measure the overall impor-tance to the constituency of trade with the U.S. EXDEPUS is expected to bepositive, while the sign of IMPORTUS is indeterminate for the same reasonsas the sign on IMPORTS is indeterminate in step 1 above.

The final two variables, FTAGAIN and FTALOSE, measure more precise-ly the lobbying influence of industries directly affected by the FTA. Indus-tries were assigned to FTAGAIN and FTALOSE based on an analysis oftranscripts of hearings before House of Commons committees investigatingthe FTA issue. Industries that lobbied in favor of the FTA are included inFTAGAIN, while industries that lobbied against the FTA are included inFTALOSE. The sign on FTAGAIN is expected to be positive, while the signon FTALOSE is expected to be negative.

The results of the logit estimation are provided in Table 4.SHIRK is the only independent variable that is significant. In fact, it is sig-

nificant at the 1% level, while all other independent variables are insignificantat the 10% level, and most are insignificant at even the 40% level. Further-more, even with SHIRK as the only significant variable, the model has closeto perfect explanatory power, as indicated by the McFadden R2 of .96.

The insignificance of everything but shirking in explaining FTA votingin Canada is particularly revealing when one considers that the FTA issuewas one of the most debated, controversial, and publicized political issues inCanadian history. Even on an issue as important as the FTA, constituent inter-ests had no significant effect on legislative voting. Thus, this is not simplythe case where legislator ideology is but one of many variables significantlyinfluencing FTA voting, as it has been in many studies of U.S. Congressionalvoting, but a case where ideology is the only variable significantly influenc-ing FTA voting.

Page 13: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

133

Table 4. Canadian FTA regression results

Independent variable Coefficient t-statistic

1. UNION –4.96 –.62

2. ESTAB 61.77 .64

3. LGESTAB –1.53 –.04

4. EXDEPUS 2.48 .13

5. IMPORTUS .04 E-3 .14

6. OPPOSE –.51 –.10

7. FTAGAIN 3.62 .31

8. FTALOSE 10.03 .96

9 UNCONT –.75 E-5 –.14

10 BUSCONT .27 E-5 .77 E-2

11. CFIT 2.20 1.36

12. DISCP 1.61 1.13

13. SHIRK 3.77 3.89

CONSTANT –3.35 –.54

Likelihood ratio = 264.0

McFadden R2 = .96

% of right predictions = 1.00.

6. The source of party shirking

The empirical results show that shirking was the only significant explanatoryvariable in FTA voting in Canada. It was discussed how this shirking behaviorwas attributable to parties, and not to individual legislators, since the lattercannot directly control their voting behavior.

However, while legislators cannot independently choose their voting posi-tions on specific issues, their personal preferences may still be at least par-tially served, albeit in an indirect manner. First, prospective politicians canchoose which party they will join. Presumably, they will join parties whoseexisting ideological beliefs are closest to their own. Second, individual legis-lators may be able to influence the policy choices of the party to which theybelong. Thus, even though party discipline may dictate that they follow theparty line, the individual legislator may have had input into the formation ofthat party line.

This subsection attempts to test the degree to which differences in partyshirking reflects underlying differences in the personal ideologies of the leg-islative members of those parties. To do this, a model is constructed wherethe dependent variable is SHIRK. Recall that SHIRK is found by disaggre-gating the residuals from the step 1 regression. SHIRK measures the effect

Page 14: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

134

Table 5. Legislator ideology and party shirking

Independent variable Coefficient t-statistic

1. AGE –.24 –2.87

2. OCCBUS .62 2.18

3. OCCFARM .40 .91

4. OCCLAW .39 1.08

5. MALE –.17 –.50

6. EDBACH –.65 –2.07

7. EDMAST –1.05 –2.48

8. EDDOC –.86 –1.94

9. MARR .11 .36

10. CHILD .06 .87

11. RELNONE –.57 –2.10

12. CATH –.43 –1.84

13. JEWISH –1.36 –2.37

14. FUND .14 .23

CONSTANT 1.54 4.26

R2 = .22

R2 = .18

F = 4.57

of party shirking on the constituency, but to what degree does it also measureshirking by that constituency’s legislator? To answer this question, SHIRK isregressed on a series of independent variables used to proxy legislator ideol-ogy. The variables are defined in Appendix C.

The independent variables serve the same basic purpose as variables 1–6 instep 1 of the model (i.e., the control variables). Their inclusion assumes thatdifferences amongst legislators in their personal characteristics proxy differ-ences amongst those legislators in their ideological preferences (see Fort etal., 1993). However, due to the nature of the variables, no prior expectationsare placed on their signs. The results of the regression are provided in Table 5.

The results indicate that about 20% of the variation in SHIRK across con-stituencies is explained by differences in legislator ideology. Since SHIRKvaries primarily across legislators from different parties (rather than acrosslegislators from the same party), the results indicate that there is some coin-cidence of ideology between individual legislators and the parties they repre-sent. That is, legislators are receiving some of the benefits of party shirking.Either because of their decision to join a party with similar ideology to them-selves, or because of their ability to influence party policy once they join theparty, the ideological preferences of legislators are, at least in part, represent-ed by the policies of the parties to which they belong.

Page 15: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

135

Table 6. Canadian FTA regression results withSHIRK disaggregated

Independent variable Coefficient t-statistic

1. UNION –4.51 –.57

2. ESTAB 44.69 .45

3. LGESTAB –.82 –.02

4. EXDEPUS .51 .03

5. IMPORTUS .10 E-4 .03

6. OPPOSE 1.03 .18

7. FTAGAIN 3.01 .25

8. FTALOSE 9.95 .94

9. UNCONT .15 E-5 .39 E-2

10. BUSCONT .38 E-6 .71 E-2

11. CFIT 2.15 1.35

12. DISCP 1.70 1.20

13. SHIRKLEG 4.53 3.13

14. SHIRKPART 3.67 3.85

CONSTANT –3.42 –.54

Likelihood ratio = 264.0

McFadden R2 = .96

% of right predictions = 1.00.

For example, a legislator who has pro-trade preferences would be morelikely to join the Progressive Conservatives, rather than the Liberals or NewDemocrats, since the former party’s pro-trade ideology is much stronger thanthat of the latter two. Furthermore, a legislator with a pro-trade ideology ismuch more likely to be able to influence policy in his preferred direction inthe Progressive Conservative party than in the Liberal or New Democraticparties.

The remaining 80% of the variation in SHIRK that is not explained bylegislator ideology must be explained by other forces within the party. Thesources of party ideology are many, and include the party’s leader, the party’sCabinet (or “shadow” Cabinet if the party is in opposition), the party’s mem-bership at large, and the party’s non-elected directors, officials, and policymakers. It is the ideological preferences of these groups that is representedby the unexplained (80%) portion of the above regression.

Given that SHIRK has now been separated into two components, the regres-sion explaining FTA voting was re-run to account for this disaggregation. Thevariable SHIRK in the original regression was replaced with two variables,SHIRKLEG and SHIRKPART. SHIRKLEG measured the fitted componentof the above regression (i.e., it measures legislator-specific ideological inter-

Page 16: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

136

ests), whereas SHIRKPART measures the ideological preferences of all othergroups within the party. The results of the regression are reported in Table 6.

The results indicate that both SHIRKLEG and SHIRKPART are signifi-cant. The significance of SHIRKLEG means that legislator-specific ideologydid influence FTA voting, albeit indirectly. Legislators were not explicitlyvoting in accordance with their personal ideologies, since party disciplineforced them to vote the party line. However, their personal ideologies wereimplicitly influencing FTA voting because these personal ideologies were, atleast in part, consistent with the ideological positions of the party to whichthey belonged.

This latter point raises an interesting issue. The system of tight party disci-pline may actually benefit individual legislators to the extent it allows theselegislators to shirk, and yet not have to suffer the blame for such shirking.If constituents complain to a legislator about his voting record, the legislatorcan sympathise with the constituents, but claim he is not responsible for therecord since his voting patterns are completely dictated by the party. Howev-er, to the extent the legislator’s ideological preferences are congruent with theideological stance of the party, the legislator is capturing some of the benefitsof party shirking.

The significance of SHIRKPART means that there are others in the party,besides the individual legislators, that are also benefitting from the shirking.Those that are able to influence party policy, such as the Prime Minister,Cabinet, the party membership, and the party’s non-elected officials, are ableto direct the policies to suit their ideological preferences.

7. Non-ideological shirking by parties

To extend the above discussion, SHIRKPART may not be entirely attribut-able to ideological shirking by those who influence party policy. Some ofSHIRKPART may be caused by the very nature of Canadian parliamentarypolitics.

To explain, parliamentary politics is characterized by an extremely adver-sarial relationship between the parties. If the government brings forth a majorpiece of legislation, the opposition parties usually oppose that bill, regardlessof the bill’s content. Opposition parties do not carefully weigh the benefitsand costs of the bill to their constituents before making a decision on whetheror not they should support the bill; instead, their decision is much more auto-matic. These parties take the term “opposition” literally; they feel it is theirduty to oppose government legislation, even if that legislation may, on net,be beneficial to their constituents.

Page 17: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

137

The econometric implications of such behavior are straightforward. First,consider that, with regard to PROTRADE, all fifteen votes that were usedto derive the variable were pro-trade in nature, and all had identical partyvoting patterns. The government would introduce and support the bills, andthe two opposition parties would automatically oppose the bills (except forone bill where the Liberals voted with the government). The result of thisbehavior, as was shown in an earlier section, is that the Conservatives (i.e.,all legislators from the Conservative party) adopted an overall position thatwas more pro-trade than the preferences of their median constituents, whilethe two opposition parties (i.e., all legislators from those parties) adoptedoverall positions that were more protectionist than the preferences of theirmedian constituents.

However, since voting on the FTA issues followed an identical pattern tovoting on the bills comprising PROTRADE (i.e., all Conservatives votingin favor of the FTA and all opposition members voting against the FTA),SHIRK should be a very effective variable in explaining FTA voting. Thatis, the same group of legislators who shirked in favor of freer trade on thePROTRADE votes (i.e., Conservatives), were also the only ones who votedin favor of the FTA, and vice-versa.

8. Summary and conclusions

Constituent preferences violations in the Canadian Parliament have two sources:those which inevitably arise due to the system of party discipline, and thosewhich arise due to the shirking of political parties. The party discipline forcesindividual legislators to frequently vote against the best interests of their con-stituents. This requirement results in constituent preference violations at theindividual constituency level, but does not, in itself, imply that the legisla-tor (party) is shirking. Parties shirk when they fail to adopt policy positionswhich are in the best interest of their median constituency.

Wide-scale shirking was, in fact, found in the policy positions of the par-ties. It was shown that this shirking element was the only significant variableinfluencing parliamentary voting on the FTA issue. Upon further analysis, itwas discovered that a portion of this party shirking was attributable to under-lying legislator ideology, but a portion was also attributable to the other actorswho may influence party policy, and to the adversarial nature of Canadianparliamentary politics. This adversarial relationship leads opposition partiesto oppose the policy proposals of the government, regardless of the effects ofthose proposals on the constituents of the opposition parties.

The causes of constituent-legislator slack in the Canadian Parliament implythat the major findings of the paper are likely not peculiar to the FTA case,

Page 18: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

138

but rather are common to a whole range of issues, both trade and non-traderelated, which are put before Parliament. Further study is needed to determinewhether this assertion is true.

Notes

1. Kalt and Zupan (1984) were the first to model the constituent-legislator relationship in aprincipal-agent framework. Their ground-breaking article has generated a wealth of sub-sequent research on the issue; research that has tested, extended, and critiqued Kalt andZupan’s original work. A remarkable consistency, however, throughout the literature is analmost exclusive focus on the U.S. Congress as the entity of study.

2. The agency theory literature has generally assumed that party discipline in the U.S. Congressis absent. However, some have challenged this notion that parties are unimportant. Cokerand Crain (1994), for example, contend that a legislator’s votes may be influenced by thepreferences of party leaders. According to Coker and Crain, because party leaders controlcommittee assignments, legislators seek to please these leaders in order to enhance theprobability of gaining a favourable committee assignment.

Thus, Coker and Crain argue that the traditional agency theory approach to the constituent-legislator link must consider a third force influencing legislator voting. Rather than view-ing legislator voting as being influenced only by the legislator’s personal ideology and byconstituent preferences, they suggest that the preferences of party leaders must also beconsidered.

3. “Free” votes (i.e., those votes where party discipline is not strictly enforced) are extremelyrare in the Canadian Parliament. Only on issues that have strong “moral” overtones, suchas abortion and capital punishment, do parties ever allow free votes to occur.

4. The importance of internal promotion rewards to creating strong party discipline is empha-sized by Franks (1987).

Franks notes that party discipline is much stronger in the Canadian Parliament thanin the British Parliament. He attributes this difference, in part, to differences in the waycabinet appointments are valued in each of the two countries.

According to Franks, some backbenchers in Britain, particularly those in one of thatcountry’s many “safe” seats, are content to live the “quiet life” and serve on parliamentarycommittees, rather than to strive to become a cabinet minister. He contends that this moti-vation is much less present in Canada because Canadian MPs face shorter political careers(i.e., there are fewer “safe” seats), and because committee work in Canada is politicallyunrewarding.

He reasons that this greater importance Canadian MPs place on attaining a cabinetposition causes them to display greater loyalty to their parties.

5. During the reign of Brian Mulroney’s Progressive Conservative government, three mav-erick MPs were expelled from the Progressive Conservative caucus.

6. The only principal-agent relationship analyzed in this paper is one where constituents areviewed as the principals and political parties are viewed as the agents.

However, it is possible to identify other principal-agent relationships within the par-liamentary system. For example, individual MPs can be viewed as agents for the politicalparties they represent. Given the complete party discipline that exists in Canada, it wouldseem that Canadian political parties are very effective in controlling the behavior of theiragents. Analysing this relationship, however, is beyond the scope of this study, and is bestleft for a separate paper.

7. Numerous analyses of U.S. Congressional voting have also employed trade-related billsas the focus of study. See, for example, Marks (1993), Nollen and Iglarsh (1990), andTosini and Tower (1987).

Page 19: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

139

References

Dyck, R. (1993).Canadian Politics. Scarborough, Ontario: Nelson Canada.Fort, R., Hallagan, W., Lamoreaux, D. and Stegner, T. (1993). Why Senators cannot vote ide-

ologically on abortion. Unpublished manuscript. Department of Economics, WashingtonState University, Pullman, WA.

Franks, C.E.S. (1987).The Parliament of Canada. Toronto: University of Toronto Press.Goff, B. and Grier, K. (1993). On the (mis)measurement of legislator ideology and shirking.

Public Choice76: 5–20.Jackson, R.J., Jackson, D. and Baxter-Moore, N. (1986).Politics in Canada. Toronto: Prentice

Hall.Kalt, J. and Zupan, M. (1984). Capture and ideology in the economic theory of politics.Amer-

ican Economic Review74: 279–300.Marks, S. (1993). Economic interests and voting on the Omnibus Trade Bill of 1987.Public

Choice75: 21–42.Nollen, S.D. and Iglarsh, H.J. (1990). Explanations of protectionism in international trade

votes.Public Choice66: 137–153.Tosini, S.C. and Tower, E. (1987). The Textile Bill of 1985: The determinants of congressional

voting patterns.Public Choice54: 19–25.

Appendix A: Independent variables for Canadian “PROTRADE”equation

1. INCOME = the median family income in the constituency

2. URBAN = the percentage of the constituency’s population residing in urbanareas

3. EDUC = the median years of education of the constituency’s population

4 NDP = the percentage of the constituency’s vote going to the New Demo-cratic Party, above the national average, in the 1984 and 1988 federalelections combined

5. AGE = the median voting age of the constituency’s population

6. WHCOL = the percentage of the constituency’s workforce employed in whitecollar occupations

7. UNION = the percentage of the constituency’s workforce that belong to a laborunion

8. EXDEP = the percentage of constituency jobs that are export dependent

9. IMPSENS = the percentage of constituents employed in import sensitiveindustries

10. IMPORTS = the per capita imports of the constituency

11. ESTAB = the per capita number of business establishments in the constituency

12. LGESTAB = the per capita number of large business Establishments (defined asthose employing at least 500 people) in the constituency

Note: Further information on data sources and methods is available from the author upon request.

Page 20: Legislative systems with absolute party discipline: Implications for the agency theory approach to the constituent-legislator link

140

Appendix B: Variables measuring constituent economic interests in theCanadian FTA equation

1. UNION = the percentage of the constituency’s workforce that belong to a laborunion

2. ESTAB = the per capita number of business establishments in the constituency

3. LGESTAB = the per capita number of large business establishments (defined asthose employing at least 500 people) in the constituency

4. EXDEPUS = the percentage of constituency jobs that are dependent on exports tothe U.S.

5. IMPORTUS = the constituency’s per capita imports from the U.S.

6. OPPOSE = the percentage of the constituency’s population that belong to atleast one of the coalition of interest groups opposing the FTA

7. FTAGAIN = the percentage of constituents employed in industries that lobbied insupport of the FTA

8. FTALOSE = the percentage of constituents employed in industries that lobbiedagainst the FTA

9. UNCONT = the contributions to the legislator from unions since the last election

10. BUSCONT = the contributions to the legislator from business organizations sincethe last election

Appendix C: Variables measuring legislator ideology

1. AGE = the age of the legislator

2. OCCBUS = a dummy variable equalling one if the legislator had a career ina business organization prior to his parliamentary experience; 0otherwise

3. OCCFARM = a dummy variable equalling one if the legislator had a farming careerprior to his parliamentary experience; 0 otherwise

4. OCCLAW = a dummy variable equalling one if the legislator had a career in lawprior to his parliamentary experience; 0 otherwise

5. MALE = a dummy variable equalling one if the legislator is male; 0 if female

6. EDBACH = a dummy variable equalling one if the legislator’s highest earneddegree is a bachelor’s degree; 0 otherwise

7. EDMAST = a dummy variable equalling one if the legislator’s highest earneddegree is a master’s (or law) degree; 0 otherwise

8. EDDOC = a dummy variable equalling one if the legislator’s highest earneddegree is a doctorate degree; 0 otherwise

9. MARR = a dummy variable equalling one if the legislator is married; 0otherwise

10. CHILD = the legislator’s number of children

11. RELNONE = a dummy variable equalling one if the legislator lists no religiousaffiliation; 0 otherwise

12. CATH = a dummy variable equalling one if the legislator is Catholic; 0otherwise

13. JEWISH = a dummy variable equalling one if the legislator is Jewish; 0otherwise

14. FUND = a dummy variable equalling one if the legislator belongs to a funda-mentalist religion (Baptist or Mormon); 0 otherwise