Jacobson 1978

7/30/2019 Jacobson 1978

1/24

The Effects of Campaign Spending in Congressional ElectionsAuthor(s): Gary C. JacobsonSource: The American Political Science Review, Vol. 72, No. 2 (Jun., 1978), pp. 469-491Published by: American Political Science AssociationStable URL: http://www.jstor.org/stable/1954105Accessed: 10/08/2009 14:54

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available athttp://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR's Terms and Conditions of Use provides, in part, that unlessyou have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and youmay use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained athttp://www.jstor.org/action/showPublisher?publisherCode=apsa .

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printedpage of such transmission.

JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with thescholarly community to preserve their work and the materials they rely upon, and to build a common research platform thatpromotes the discovery and use of these resources. For more information about JSTOR, please contact [email protected].

American Political Science Association is collaborating with JSTOR to digitize, preserve and extend access toThe American Political Science Review.

http://www.jstor.org
http://www.jstor.org/stable/1954105?origin=JSTOR-pdfhttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/action/showPublisher?publisherCode=apsahttp://www.jstor.org/action/showPublisher?publisherCode=apsahttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/stable/1954105?origin=JSTOR-pdf

7/30/2019 Jacobson 1978

2/24

TheEffects fCampaign pendingin Congressional lections

GARY C. JACOBSONTrinity College

Incomplete understanding f the connection between campaign pending and election outcomeshas hindered evaluation of enacted and proposed congressional campaign finance reforms.Reanalysis of the 1972 and 1974 House and Senate campaign pending data using both OLS and2SLS regression models shows that spending by challengers has a much greater impact on theoutcome than does spending by incumbents. A similar analysis of the effects of spending on voters'recall of candidates n the 19 72 and 1974 SRC surveys supports the explanation hat campaignexpenditures buy nonincumbents he necessary voter recognition already enjoyed by incumbentsprior to the campaign. The 1974 survey questions on Senate candidates ndicate that, although heinability to remember candidates' names does not preclude having opinions about them, votersrecalling candidates are much more likely to offer evaluative omments, and these more requentlyrefer to candidates personally. Aware voters offer more negative as well as positive evaluations(though positive outnumber negative); amiliarity s not automatically advantageous. And voters'evaluations of candidates trongly influence how they vote. The implications of these indings orcongressional ampaign inance policy are readily apparent.

Legislation extending public funding to con-gressional campaigns was on the agendas ofboth the House and Senate during the firstsession of the 95th Congress. The Senate bill,S 926, won majority support but was killed byfilibuster; it is not likely that this setback hassettled the issue. As with other recent lawsintended to alter the way in which congres-sional campaigns are financed-the FederalElections Campaign Act of 1971 (PL 92-225)and the Federal Election Campaign Act Amend-ments of 1974 (PS 93-443)-consideration ofthis legislation was not informed by any clearunderstanding of its likely consequences. Norwill be future debates, as long as the crucialquestion of how campaign expenditures affect

*1 am grateful to Christopher Achen, Forrest

Nelson, John Ferejohn, Roger Noll, Stephen Rosen-stone, William P. Welch, and Diane Zannoni for theirhelpful suggestions and critical comments on earlierversions of this article. They are of course free of anyresponsibility for its remaining shortcomings. The dataused here were made available in part by the Inter-University Consortium for Political Research; I amsolely responsible for all analyses and interpretations.Some of the material presented here was deliveredunder the title "Campaign Spending and Voter Aware-ness of Congressional Candidates," at the AnnualMeeting of the Public Choice Society, New Orleans,Louisiana, March 11-13, 1977 and in "The ElectoralConsequences of Public Subsidies for CongressionalCampaigns," at the annual meeting of the AmericanPolitical Science Association, Washington, D.C., Sep-tember 1-4, 1977.

the outcomes of congressional elections remainsunanswered.

The work reported in this article is intendedto clarify the structure of the aggregate rela-tionship between spending and congressionalelection results and to indicate how campaignspending is linked to the behavior of voters.Specifically, it will show that spending bychallengers has a substantial impact on electionoutcomes, whereas spending by incumbents hasrelatively little effect; the evidence is particular-ly strong for House elections. The much greaterimpact of the challenger's spending remainswhen simultaneity bias is eliminated by meansof two-stage least squares regression analysis. Insimple terms, the more incumbents spend, theworse they do; the reason is that they raise andspend money in direct proportion to themagnitude of the electoral threat posed by thechallenger, but this reactive spending fails tooffset the progress made by the challenger thatinspires it in the first place.

An explanation for these findings is devel-oped from an analysis of the effects of cam-paign spending on voter recall of incumbentand nonincumbent congressional candidates:campaign expenditures buy nonincumbents thenecessary voter recognition already enjoyed byincumbents prior to the campaign. The analysisshows that awareness has an important effecton voter evaluation of candidates and, conse-quently, on voting behavior in congressionalelections. The article concludes with considera-tion of some salient implications of thesefindings for enacted or proposed changes incampaign finance policy.

469

7/30/2019 Jacobson 1978

3/24

470 The American Political Science Review Vol. 72

Aggregate Effects of Campaign Spending

A number of recent studies have found thatwhat candidates spend in legislative contests isindeed related to how well they do on election

day (Palda, 1973, 1975; Welch, 1974, 1976;Dawson and Zinser, 1976). However, with fewexceptions (Glantz, et al., 1977; Jacobson,1976), these studies have assumed that cam-paign spending has the same consequences forincumbents and challengers alike. In particular,economic models investigating the "productiv-ity" of campaign spending in terms of winningvotes or elections have been grounded on theimplicit premise that the marginal productivityof campaign expenditures is the same for allcandidates (Lott and Warner, 1974; Welch,

1974; Silberman, 1976). For congressional can-didates, a contrary assumption is much moredefensible. The advantages of incumbency arewell known. The list of perquisites and allow-ances senators and representatives have grant-ed themselves is too familiar to require reitera-tion. Incumbents control resources easily worthseveral hundred thousand dollars annually(Cover, 1977; Perdue, 1977); these resourcesare unquestionably used to pursue reelection, ifonly because, for most members of Congress,the campaign never ends (Mayhew, 1974b). In

light of the enormous head start thereforeenjoyed by incumbents, it would be surprisingindeed if campaign spending were not moreimportant to challengers-and to other non-incumbents-than to incumbent candidates.

Evidence from the 1972 and 1974 congres-sional elections, to be presented in this section,holds no such surprises; it strongly supports theconclusion that what the challenger spends is animportant determinant of the outcome, whilespending by incumbents makes relatively littledifference. Incumbents are apparently able toadjust their level of spending to the gravity of aspecific challenge; they spend more when chal-lengers spend more, less when challengers spendless. But he marginal gain in support derivedfrom additional spending does not approachthat of the challenger from an equal spendingincrement; the more both candidates spend, thebetter the challenger does.

The evidence for this interpretation is de-rived from multiple regression equations inwhich challenger and incumbent spending areentered, along with appropriate controls, asseparate variables, so that their differing im-pacts are clearly displayed. An important com-plication is involved, however. Ordinary leastsquares (OLS) regression models presupposeunidirectional causality-in this case, thatspending produces votes. But reciprocal causal-

ity is an equally plausible premise-and failureto take this into account is another commondeficiency in the literature on campaign spend-ing effects (Palda, 1975, being an exception).The expectation that a candidate will do well

may bring campaign contributions. Suppose itis possible to make a rough prediction of theoutcome prior to the election; if campaigncontributors as as "rational investors" who,other things equal, invest more in a campaignthey expect to be successful (since one elementof risk is smaller), contributions to candidatesshould increase with their probability of elec-tion(Ban-Zion and Eytan, 1974; Welch, 1974,1977; Dawson and Zinser, 1976).

Or, from a slightly different perspective,campaign spending may help win popular sup-

port, and thus votes, but characteristics thatalso help to attract votes-personal charm or"charisma," political skill and experience-should also ease the job of fundraising. Candi-dates who are well known and who havepolitical experience (and thus a greater likeli-hood of success) raise money more easily,spend it, thereby further increasing their popu-larity (and chances for victory), acquiring inconsequence even more money, and so on-theultimate payoff coming in the form of addition-al votes on election day.

The ordinary least squares regressions re-ported in most studies are inappropriate forestimating reciprocal relationships; a simul-taneous equation system is required. OLS esti-mates of parameters when the true relationshipis reciprocal are biased and inconsistent becauseendogenous variables (those which have a re-ciprocal effect on one another), when treated asexplanatory variables, are correlated with theerror term (Johnston, 1972, p. 343). The two-stage least squares (2SLS) regression procedure,a standard solution to this problem, is therefore

used later in this section to estimate the effectsof campaign expenditures by challengers andincumbents within a system of simultaneousrelationships.

Despite its potential inadequacies, however,a straightforward OLS regression equation pro-vides a useful starting point for determining theaggregate effects of campaign spending in con-gressional elections. The equation estimated forthe 1972 and 1974 House elections was

CV=a +b1CE+b2IE+b3P+ ('1)b4CPS + e

whereCV is the challenger's percentage of the

7/30/2019 Jacobson 1978

4/24

1978 The Effects of Campaign Spending in Congressional Elections 471

two-party vote1CE is the challenger's campaign expenditures

in thousands of dollars2IE is the incumbent's campaign expenditures

in thousands of dollarsP is the challenger's party (1 if Democrat, 0

if Republican)CPS is the strength of the challenger's party

in the district (approximated here by thepercentage of the vote won by the chal-lenger's party's candidate in the lastelection for this seat)3

a is the intercept, the b's regression coefficients,and e the error or disturbance term. Thechallenger's share of the vote is hypothesized tobe a function of what the challenger and

incumbent spend, the challenger's party, andthe strength of that party in the district. Noticethat an equivalent equation with observationson incumbents would produce estimates of thecoefficients which mirror those derived fromthis model; either one would support the samesubstantive conclusions.

Since our interest is in the effects ofcampaign spending, variables P and CPS serveprimarily as controls in this equation. The partyvariable accounts for national short-term forcesfavoring one party or another in a particular

election year. District party strength is mea-sured by the vote for the challenger's party inthe most recent prior election for that seat;though far from ideal as an approximation ofthe expected or "normal" vote, it has theadvantage over the other possible index (per-centage of registrants with the challenger'sparty in the district) of being available for amuch larger proportion of the districts. At onestage of this research I used registration percen-tages in place of CPS, and the results wereessentially the same as those reported in thisarticle, though the number of cases was halved.Both the party and party strength variables areexpected to affect a candidate's ability to raisemoney as well as to win votes and so must betaken into account in this preliminary model.

Challenger and incumbent spending are en-tered as separate variables rather than as somecomposite (for example, the challenger's per-

tThe election results are from Scammon (1975).2The 1972 data are from Common Cause (1972);

the 1974 data are from Congressional Quarterly(1975, pp. 789-96).

3The data source is Congressional Quarterly(1974b). Previous vote percentages have been adjustedin these data for changes in district boundaries whereredistricting has occurred.

centage of expenditures by both candidates)because their coefficients are not expected tobe the same. The functional relationship be-tween spending and votes is assumed to belinear. This has the advantage of simplicity butthe drawback that it fails to allow for thediminishing returns that must apply to cam-paign spending; no candidate can get more than100 percent of the vote, no matter how much isspent. An attractive alternative is the semilogform in which spending is entered as the naturallogarithm of actual expenditures (Welch, 1976);it permits diminishing returns but does notallow them to become negative as would, forinstance, a quadratic model (Silberman andYochum, 1977).

Both the linear and semilog forms fit thedata equally well; the R2's are identical. Butthe semilog model has the defect of seriouslyunderestimating the challenger's vote at higherlevels of spending; that is, it provides estimateswhich exaggerate the extent to which returnsdiminish as spending increases. Examination ofthe residuals (a residual is, in this case, thedifference between the percent of votes actual-ly won by the challenger and that predicted bythe regression equation) showed this to be thecase. The problem is illustrated by comparingthe actual number of winning challengers in

both election years with the number predictedby the linear and semilog equations:

Winning Challengers 1972 1974

Actual number4 9 39Number predicted by:

Linear quation 5 29Semilog equation 1 2

The linear equation exaggerates the expectedvote of challengers at higher levels of spending,but

inspectionof the

residualsindicates that

this is not a significant problem until thechallenger's spending exceeds $160,000, whichoccurs in less than 2 percent of the cases ineither election year; at this level of spending theequations are less likely to overpredict thenumber of winning challengers than they are tooverstate the size of the challenger's victory. I

4Actually 13 incumbents ost in 1972, but three ofthem were defeated by other incumbents they were

forced to run against because of reapportionment nda fourth lost a three-way race running as an inde-pendent. Forty incumbents ost in 1974, but one ofthese had just been elected in a special election and noseparate pending igures were available or the secondcontest.

7/30/2019 Jacobson 1978

5/24


therefore chose to present the linear equations,but the reader should be aware that a compar-able analysis with semilog equations woulduphold the substantive conclusions defendedbelow. The regression estimates of equation 1.1

are reported in Table 1.5According to the equations in Table 1, it isclearly the challenger's level of spending thathas the greatest impact on the outcome of theseelections; challengers are expected to gain alittle over 1 percent of the vote for every$10,000 they spend. Incumbent spending ap-parently makes much less difference. The sim-ple correlation between incumbent expendi-tures and the challenger's vote is in fact positive(.39 for 1972, .46 for 1974); ignoring otherfactors, the more incumbents spend, the worse

they do. With the challenger's spending con-trolled, the incumbent's spending has a weaknegative effect on the challenger's vote; itscoefficient is not statistically significant in the1972 equation. This implies that incumbentsare able to expand their financial resources inresponse to a serious challenge (represented bythe challenger's level of spending), but that thisadditional spending either does them little goodor at best does not begin to match the muchgreater benefit challengers derive from anequivalent increase.

The evidence that incumbents are able toadjust their spending to the gravity of thechallenge is convincing. Regression of incum-

5Observations included only those contests inwhich a major party challenger aced a major partyincumbent and for which data on the vote in theprevious election were available. This latter require-ment forced twenty-three 1972 and two 1974 observa-tions to be dropped.

bent expenditures on a variety of explanatoryvariables shows that the challenger's level ofspending has by far the greatest explanatorypower. The equation estimated was:

IE=a+bCE+b2P+b3CPS+ b4IP+b5YRS+b6PO+b7L+e (2.1)

where

IP is 1 if the incumbent ran in a primaryelection, 0 otherwise

YRS is the number of consecutive years theincumbent has been in the House

PO is 1 if the challenger has previously heldelective office, 0 otherwise6

L is 1 if the incumbent is chair or ranking

member of a subcommittee or holds ahigher leadership position, 0 otherwise

and IE, CE, P, CPS, and the coefficients are asdefined for equation 1.1. The results appear inTable 2. Obviously, the challenger's spending isthe most important explanatory variable inthese equations. And the other variables workas expected on the assumption that incumbentsspend in response to the gravity of the electoralchallenge. For example, incumbents spendmore if the challenger has held elective office,or-in 1974-if the challenger was a Democrat;they spend less, other things being equal, thelonger they have been in office. But all of thesevariables would be expected to show oppositesigns if the equations estimated the incumbent'scapacity to raise money according to politicalassets and likelihood of reelection.

6From information in Congressional Quarterly(1972 and 1974a).

Table 1. The Effects of Campaign pending n the 1972 and 1974 House Elections (OLS): Equation 1.1Standardized

Regression RegressionCoefficient t-ratioa Coefficient

1972 (N=296) CV= a 20.7bjCE .112 9.42 .51b2IE -.002 -.14 -.01 R2 =.49b3P -.47 -.61 -.03b4CPS .299 6.94 .33

1974 (N=319) CV= a 15.6b1CE .121 10.45 A8

b2IE -.028 -2.34 -.11 R2 .65b3P 9.78 11.19 A2b4CPS .351 7.75 .28

aGiven the degrees of freedom in these equations, a t-ratio of at least 1.98 is necessary for a .05 level of sig-nificance, 2.58 for .01, and 3.35 for .001.

7/30/2019 Jacobson 1978

6/24


Even more to the point, the difference inspending between 1972 and 1974 by incum-bents who ran and were opposed in bothcampaigns is an almost identical function of thedifference in spending by their opponents inthe two elections. This is apparent from anestimate of

(IE74-IE72) = a + b1(CE74-CE72) +

b2IE72 +b3CE72 +b4IP72 +b5IP74 +

b6P+e (3.1)

where the variables and coefficients are asdefined for equations 1.1 and 2.1 and thesubscripts on the variables indicate the electionyear. The results are shown in Table 3. If thecontrols for primary elections, party, and 1972spending by both candidates are omitted, the

relationship between the change in challengerspending and the change in incumbent spendingscarcely varies. The regression coefficient be-comes .505, its t-ratio 13.23, and its stan-dardized regression coefficient .61.

Incumbents apparently increase or decreasetheir spending in reaction to changes in theamount spent by opponents. Any increase,however, does not counterbalance benefits tothe challenger from the spending that inspiredit in the first place. The difference in spendinglevels by incumbents between 1972 and 1974 isnegatively correlated (-.58) with the dif-ferences in the proportion of votes won by theincumbent in the two elections. A significantnegative relationship remains even whenchanges in the challenger's spending are takeninto account.

Table 2. Determinants f Spending by Incumbents n the 1972 and 1974 House Elections: Equation 2.1

StandardizedRegression RegressionCoefficient t-ratioa Coefficient

1972 (N=296) IE = a 28.08b1CE .522 10.33 .54b2P 3.31 .87 .04

b3CPS .224 1.05 .06 R2 .39b4IP 10.50 2.69 .13b5YRS -.789 -2.69 -.15bPO 3.10 .69 .03b7L .89 .14 .01

1974 (N=319) IE = a 14.11bjCE .495 10.30 .51b2P 12.69 2.88 .14b3CPS .673 3.05 .14 R2 =.47b4!P .10 .02 .00b5YRS -.035 -.11 -.01b6PO 6.79 1.45 .07b7L -12.50 -2.00 -.10

aSee Table 1.

Table 3. Determinants f Changes n Spending by House Incumbents Between 1972 and 1974: Equation 3.1


(N=295) (IE74-IE72) = a 28.92b1(CE74-CE72) .491 11.73 .60b2IE72 -.454 -11.96 -.58

b3CE72 .398 7.73 .43 R2 .63b4!P72 -8.24 -2.35 -.09b5IP74 -1.35 - .39 -.01b6P -13.14 -3.54 -.15

aSee Table 1.

7/30/2019 Jacobson 1978

7/24


Of the 39 incumbents who lost in 1974, 34(87 percent) spent more in losing than they didin winning in 1972. The mean expenditure forall incumbents who ran and were opposed inboth elections was $61,799 in 1972 and

$63,609 in 1974, an increase of 3 percent. Themean expenditure of the 1974 losers was$69,218 in 1972 and $101,645 in 1974, a 47percent increase. Most were of course Repub-licans. But in a year electorally disastrous forRepublicans and financially hopeless for Re-publican challengers (1974 Democratic chal-lengers spent an average of $59,352, Repub-lican challengers $21,463), Republican incum-bents actually outspent Democratic incumbentsby, on the average, about $35,000 ($81,437 to$46,261). Despite the extraordinarily hostile

political environment, Republican incumbentswere able to increase their spending by almost55 percent over the previous election; spendingby Democrats actually decreased.

Since the circumstances of the 1974 electionwere rather unusual, evidence from other elec-tions would be pertinent. Unfortunately, 1972is the first election for which reasonably com-plete data on spending in House election areavailable. However, for both 1970 and 1972data have been published on spending for radioand television time by House candidates in the

general election campaign, so some comparisonsare possible. In 1970 the losing incumbentsspent an average of $9,628 on radio andtelevision; winning incumbents spent an averageof $4,572. The comparable figures for 1972are $16,220 and $5,727, respectively. Meanexpenditures for broadcast time grew from$4,738 to $6,097 for all incumbents between1970 and 1972, an increase of 27 percent; thesame expenditures for incumbents who ran andwere opposed in both years and who lost in1972 went from $8,696 to $16,220, an increase

of 87 percent.7The point is clear-and fundamental tocomprehending the role of money in congres-sional elections. Incumbents are evidently ableto raise and spend money in direct proportionto the perceived necessity to do so, this being afunction of the gravity of the electoral threatposed by the opposition. None of the contribu-tion or demand functions previously estimatedfor models of campaign finance processes havetaken this into account (Bental, Ben-Zion, andMoshel, 1976, 1977; Dawson and Zinser, 1976;

Silberman and Yochum, 1977; Welch, 1977);

7Data upon which these figures are based are fromUnited States Congress (1971 and 1973).

they therefore require respecification.

The argument that reactive spending by

incumbents does not offset the gains accruingto challengers from the spending that inspiredthe reaction is based, in part, on possiblyunreliable OLS estimates. A more completemodel of the money-vote relationship postu-lates a reciprocal connection between thesevariables:

CE = f(P, CPS,PO, YRS, CP,E V) (4.1)

IE =f(PCPSPOYRSIPEVCE) (4.2)

CV = f(CEIE,PO, CPS) (4.3)

EV_ CV (4.4)

The variables are as defined for the previousequations, with the addition of CP (1 if thechallenger ran in a primary election, 0 if not)and EV, the challenger's expected vote.

Equation 4.1 hypothesizes that the chal-lenger's ability to attract contributions is, tobegin with, a function of the challenger's party(consider 1974) and district party strength(measured, remember, as the proportion ofvotes won by the party's candidate for the seatthe last time around; it may also be interpretedhere as an indicator of the vulnerability of theincumbent). The variable PO measures theeffects of prior electoral success, and theexposure, experience, and contacts that comewith holding elective office, on the ability toraise campaign funds. The number of years theincumbents has held office is another indicatorof vulnerability, this affecting the attractivenessof the challenger as an "investment."

Since the data do not include separatefigures for primary and general elections, CP(and IP in the second equation) is included topick up the differences in spending broughtabout by the demands of a primary contest.There is no way to determine accurately howmuch of a candidate's money was spent in theprimary. This is not so troublesome as it mayseem. Challengers spend as much as they canraise anyway. Their principal problem is tomake themselves known to voters, and this canbe done as effectively in a primary as in ageneral election campaign (Jacobson, 1976);primary election spending should thereforehave general election payoffs. And incumbentsspend according to what the challenger does.We do not, therefore, expect particularly dra-matic changes in spending levels if a candidatedoes or does not contest a primary. But

7/30/2019 Jacobson 1978

8/24


coefficients on these variables will give someindication of the degree to which primaries doaffect spending.

Finally, challengers are expected to attractcontributions in proportion to their probability

of being elected, which is here approximated bytheir expected vote, itself of course related tothe actual vote.

Interpretation of the equation for incum-bent spending is quite different. Incumbentspending is not a positive function of thelikelihood of victory at all; rather, the morecertain they are of election, the less incumbentsspend. This does not mean that rational in-vestors deliberately ignore them-who wouldnot want to invest in a sure thing?-or that theycould not raise a great deal more money if they

wished. The explanation is simply that incum-bents sure of victory feel no need for themoney that is available. Soliciting and acceptingcontributions is hardly something politiciansenjoy. Hubert Humphrey called it "a disgusting,degrading, demeaning experience," and othershave echoed his sentiments (Adamany andAgree, 1975, p. 8). Indeed, this may be themajor reason many members of Congress cur-rently favor public funding of congressionalcampaigns.

Incumbents, then, acquire funds only in

proportion to the felt necessity to do so. Andthey can usually get all they need. The variablesthat determine incumbent spending, therefore,indicate how much the candidate is likely toneed. And this in turn, is primarily a functionof the strength of the challenger. Since thechallenger's strength is indicated in good mea-sure by financial resources, CE belongs inequation 4.2 as an explanatory variable. Themeasure of incumbent expenditures, IE, doesnot similarly belong in the challenger's expendi-ture equation; challengers are not able to raisemoney at will to contest an incumbent whomay be spending a great deal. Rather, theyspend all they can (at least up to very largeamounts) independently of what the incumbentis spending.

The other variables in the incumbent'sspending equation can be interpreted in thesame way; they determine how threatenedincumbents are likely to feel and therefore howmuch they find it prudent to raise and spend.

The third equation is of course the oneoriginally estimated by OLS. The other exo-genous variables (those determined outside theequation system) are left out on the theoreticalpremise that they affect CV only indirectly(through their effect on spending); empirically,these variables had no statistically significantconnection with the election outcomes with the

other variables controlled. Since CV is approx-imated with some significant degree of accuracyby EV (equation 4.4), the OLS estimates areliable to bias and inconsistency and thus areunreliable estimates of the structural

parameters (those of the true causal relation-ships in equation 4.3). A standard solution tothis difficulty is the two-stage least squaresprocedure. The first step is to regress challengerand incumbent spending on all of the exo-genous variables in the system. The equationsto be estimated are:

CE*=a +bjP+b2CPS +b3PO +

b4YRS+b5CP+b6IP+e (5.1)

IE*=a +bjP+b2CPS+b3PO+b4YRS+b5CP+b6IP+e. (5.2)

The estimated parameters are then used tocompute CE* and IE* for each observation,and these variables replace CE and IE in thesecond stage equation,

CV=a +bCE*+b2IE*+b3P+b4CPS + e. (5.3)

The 2SLS procedure "purges" the ex-planatory variables CE* and IE* of the com-ponent associated with the error term. Theresulting estimates are still biased estimates ofthe true structural parameters but are nowconsistent; the bias decreases as the sample sizeincreases, approaching zero in the limit (Raoand Miller, 1971, p. 214; Johnston, 1972, pp.380-84). The estimates of equations 5.1, 5.2,and 5.3 appear in Table 4.

The 2SLS results recapitulate the OLS find-ings in one very important respect: a givenamount of

campaign spending does not havethe same consequence for challengers and in-cumbents. Spending by challengers has a muchmore substantial effect on the outcome of theelection even with simultaneity bias purgedfrom the equation. Indeed, the regression coef-ficients on CE * are larger than those for CE.However, the standardized regression coeffi-cients for CE* are smaller than those for CE(.36 compared to .51 for 1972, .38 comparedto .48 for 1974); the steeper slopes are evident-ly an artifact of the much smaller range of theinstrumental variable CE*, less than half that ofCE in both election sets. Even so, OLS does notappear to greatly exaggerate the effects of thechallenger's spending, nor does it substantiallyunderestimate the effects of incumbent spend-ing.

7/30/2019 Jacobson 1978

9/24


Table 4. The Effects of Campaign pending n the 1972 and 1974 House Elections (2SLS):Equations 5.1, 5.2, and 5.3

StandardizedRegression RegressionCoefficient t~ratioa Coefficient

1972 (N=296)First-Stage CE* a -17.56Equations b1P -11.34 -2.60 -.14

b2CPS 1.39 5.92 .34b ?PO 21.20 4.18 .23 R2 =.23b4YRS -.40 -1.36 -.07b5CP 8.28 1.96 .10b61P 1.92 .42 .02

IE* a 18.23b 1P -2.86 -.66 -.04b2CPS .93 3.95 .23bPO 13.40 2.64 .15 R2 =.18b4YRS -.99 -3.36 -.19b5CP 8.85 2.10 .12b JP 10.73 2.34 .13

Second-Stage CV= a 22.3Equations blCE* .163 2.5. 36

k2IE* -.051 -.89 -.10 R2.=46bb3P -.01 -.01 -.00b4CPS .269 3.55 .30

1974 (N=319)First-Stage CE*= a -16.03Equations bIP 28.15 5.40 .31

b2CPS 1.07 4.25 .22bPO 21.00 3.89 .20 R2 =.29b4YRS -.19 -.58 -.03b5CP 3.77 .77 .04bJ6P -3.16 -.66 -.03

IE* a 6.15bIP 22.55 4.40 .25b2CPS 1.23 4.94 .26b3PO 17.00 3.19 .17 R2=.28b4YRS -.42 -1.31 -.07b5CP 7.37 1.53 .08b JP -2.30 -.49 -.02

Second-Stage CV a 17.1

Equation bICE* .179 1.21b 38b2IE* -.022 -.14 -.05 R2 = 63bb3P 7.51 4.20 .32b4CPS .264 2.94 .21

aSee Table 1.bThe R2s and t-ratios for the seco Td-stage equations are adjusted figures; they cannot be computed directly

from the second stage regression runs.

The parameters estimated in the second-stage equations indicate that challengers receive1.63 percent to 1.79 percent of the vote for

8The R2's and the t-ratios for the second stageequations are not taken directly from the statisticsproduced by estimating hese equations. Rather, theyare found by replacing the standard error of theestimate from the equation as computed by a standard

each $10,000 they spend; they are expected tolose between 0.51 percent and 0.22 percent foreach $10,000 the incumbent spends; therefore,

error computed from a combination of the 2SLSparameters with the actual spending variables CE andIE replacing CE* and IE*. I am obliged to JohnFerejohn or explaining his procedure o me.

7/30/2019 Jacobson 1978

10/24


if spending by both candidates increases by thisamount, the net gain for challengers shouldapproximate at least 1 percent of the vote.These equations also show (as did the OLSequations) that it was very advantageous to be a

Democrat in 1974 (it was worth an additional7.5 percent of the vote), whereas in 1972 thecandidate's party made no difference. And theyindicate that the challenger's district partystrength also contributes significantly to theshare of the vote won.

One troubling aspect of these equations isthat in both election sets CE* and IE* are veryhighly correlated. The reason for this is clearfrom the first-stage equations; the variablesseem to have similar effects on the spendinglevels of both candidates. This is of course as it

should be, if incumbent spending is basically areflection of challenger spending, as I have beenarguing. But it raises the problem of multicol-linearity. Multicollinearity destroys the preci-sion of the estimates; notice that the coefficientfor CE in the 1974 equation is not statisticallysignificant by the usual criterion (a t-ratio of1.98 or larger). According to Johnston (1972,p. 163), a very large positive correlation be-tween two explanatory variables is likely toproduce large and opposite errors in the esti-mates of regression coefficients.

Under these circumstances it is useful to runthe second-stage equations excluding one of thecorrelated variables. Estimates of the equationsomitting IE* are found in Table 5. The coef-ficients for CE* decrease somewhat, as anti-cipated. But the precision of the estimatesincreases, especially in the 1974 equation. IfIE* were left in the equations instead, thisvariable would also show a strong positiverelationship to the challenger's share of the

vote; IE* like IE, has a significantly positivesimple correlation with CV (the adjusted figuresare .51 for 1972 and .78 for 1974).

Some further comments on the first-stageequations are in order. Spending by bothchallengers and incumbents changes by approxi-mately $1,000 for a change of 1 percent in thevote won by the challenger's party in the lastelection. Challengers who have won politicaloffice before spend an average of about$21,000 more than those who have not; incum-bents also spend more against these candidates.These two variables make the most difference.The advantages of being the incumbent areclear; notice the intercepts. So are the advan-tages of being a Democrat in 1974. The R2'sfor these equations are not as large as would bedesirable; a number of other explanatory vari-ables were tested in earlier stages of theresearch, but none of them improved any of theequations significantly. The criterion for inclu-sion, other than theoretical plausibility, wasthat a variable had to have a regression coef-ficient at least twice its standard error in atleast one of the equations.

Before proceeding to develop an explanationof these results, it will be instructive to considerthe effects of campaign spending on the elec-toral fortunes of candidates running in contestsin which neither is an incumbent. The OLSregression estimates of the relationship forDemocratic candidates (had Republican candi-dates been chosen instead, the results wouldhave formed a mirror image of those reported)appear in Table 6; the party variable has ofcourse been omitted, the CE, IE, and CPS arereplaced by DE (the Democrat's spending inthousands of dollars), RE (the Republican's

Table 5. The Effects of Challengers' Campaign Spending in the 1972 and 1974 House Elections (2SLS):Equation 5.3 Omitting IE*


1972 (N=296) CV= a 20.8bCE* .117 3.42b .26b3P -.41 -.48 -.02 R2 =.49bb4CPS .289 4.07 .32

1974 (N=319) CV= a 16.7

bCE* .1603.60b .34

b3P 7.53 4.21 .32 R = .62b4CPS .257 3.49 .21

aSee table 1.bSee table 4.

7/30/2019 Jacobson 1978

11/24


spending), and DPS (Democratic partystrength-measured like CPS) in these equa-tions.

The results for 1972 are as we shouldanticipate. The marginal effects of spending are

similar for both candidates; the Democrat'sown spending helps, that of the Republicandoes the opposite. For 1974, however, noticethat the Republican candidate's spending makesa much greater difference in the outcome thandoes that of the Democrat, which is perverselysigned and is not significantly related to theoutcome at all. In other words, Republicancandidates in 1974 were affected by spendingin a way we would expect if they were, instead,challenging incumbents. But this is not sosurprising. Given the post-Watergate political

atmosphere, poisonous for Republicans,Repub-

lican candidates found themselves in the heavilydisadvantaged position usually reserved forchallengers regardless of whether or not theDemocrat was really an incumbent. Campaignspending is evidently most useful to candidatessuffering severe electoral handicaps, no matterwhat the source.

All of the foregoing refers to House contests.Analysis of Senate elections will be muchbriefer. The regression model which clearly fitsthe data best is the semilog form:

CV= a + bl InCEPC + b2lnIEPC +

b3P + e (6.1)

whereInCEPC is the natural log of the challenger's

expenditures in cents per voting-age indi-vidual9

9The voting age populations of the states are fromUnited States Congress (1974, p. 18542 and 1977, p.835).

InIEPC is the natural log of the incumbent'sexpenditures in cents per voting-age in-dividual

and the other variables and coefficients are asdefined for equation 1.1. Actual spending wasdivided by the voting-age population of thestate in recognition of the widely varying statepopulations. No measure of party strength wasincluded in the equation because in no modeltested did this variable (measured by an indexsimilar to Kostroski's base party votel) showany significant connection with Senate electionoutcomes. The results of this analysis are foundin Table 7. They are somewhat more ambiguousthan the results for House elections. Thechallenger's spending does appear to have agreater impact, but for 1972 the difference is

not large, and in both years the incumbent'sspending was negatively related to the chal-lenger's share of the vote. The simple correla-tion between incumbent spending and thechallenger's vote was -.17 for 1972 and .48 for1974, again a mixed outcome.

The 2SLS technique was also attemptedwith Senate data; for 1972 the first-stageequations explained so little of the variancethat even remotely trustworthy 2SLS estimateswere out of the question. Estimates wereobtained for 1974; they were almost identicalto the OLS estimates. The tentative conclusionsuggested by these findings is that the chal-lenger's expenditures are more effective inSenate elections, too, but incumbent senatorsbenefit from their own spending to a greaterdegree than do their counterparts in the House,

1OThe index of state party strength was computedas the smallest proportion of the total statewide Housevote won in aggregate by House candidates of theSenate candidate's party in any election year from1968 to 1974. See Kostroski (1973).

Table 6. The Effects of Campaign Spending in the 1972 and 1974 House Elections for Open Seats (OLS)

StandardizedRegression RegressionCoefficient t-ratioa Coefficinet

1972 (N=52) DV= a 34.7b1DE .045 2.56 .31b2RE -.077 -2.84 -.34 R2 = .46b4DPS .308 4.15 .47

1974 (N=53) DV= a 51.4

b1DE -.002 -.10-.01

b2RE -.130 -5.07 -.54 R2 =59b4DPS .328 3.95 .41


7/30/2019 Jacobson 1978

12/24


Campaign Spending andCandidate Familiarity

An attractive theoretical explanation for thefindings reported n the first section begins with

the observation that incumbents are alreadyfamiliar to voters at the outset of the campaign,whereas nonincumbents probably are not. Theresources of office provide ample means foracquiring voter recognition and, beyond that,building a favorable reputation (Cover, 1977;Abramowitz, 1975). Additional nformation onincumbents disseminated during the campaignshould have less impact on an electorate whichhas already been subjected to a barrage ofmessages from the candidate. Nonincumbentsnormally have much more to gain in the way ofvoter awareness n the course of the campaign,implying that the more extensive-and there-fore expensive-the campaign, he better knownthey will become.

Voter recognition appears to be an im-portant component of electoral success. The

pioneering work of Stokes and Miller (1962)showing the connection between voter aware-ness of House candidates and the frequency ofpartisan defection in 1958 has been reinforcedby studies of more recent House elections

(Arsenau and Wolfinger, 1973). The effects ofdifferential awareness of candidates on partisanvoting patterns in the 1972 and 1974 Houseelections and the 1974 Senate elections (whichwill provide the data base for this section) areshown in Table 8. The candidate familiarityquestion was asked about Senate candidates forthe first time in 1974; notice that partisanvoters respond to differential awareness ofSenate candidates in essentially the same way asto differential awareness of House candi-dates.1 1

ttThe candidate-recognition questions, coded asvariables 2174, 2175, and 2176 of the 1974 SRCSurvey, were "Now let's talk about the campaign forSenator. Do you remember what the candidates'names were? What were they?" (Miller, Miller, andKline, 1975, pp. 103-05).

Table 7. The Effects of Campaign pending n the 1972 and 1974 Senate Elections (OLS): Equation 6.1


1972 (N=25) CV= a 41.8bllnCEPC 5.55 4.15 .78b2JnIEPC -4.81 -2.77 -.49 R =.47b3P 2.49 .81 .14

1974 (N=22) CV= a 35.4bllnCEPC 3.77 3.69 .78b2JnIEPC -1.39 -.98 -.20 R = .81b3P 8.26 5.18 .55


Table 8. The Effects of Awareness f the Candidates n Partisan Defection in House and Senate Elections

Voter Was Aware of:Percent Who Both Own Other NeitherDefected in: Candidates Candidate Candidate Candidate

House Elections1972 23 7 62 21

(I51)a (100) (26) (242)1974 29 1 58 15

(281) (154) (79) (374)

Senate Elections1974 25 2 62 10(317) (138) (87) (205)

aNumber of cases from which percentages were computed. The 1974 sample is weighted.Source: The 1972 and 1974 SRC surveys.

7/30/2019 Jacobson 1978

13/24


The same data provide information on theadvantage in voter recognition typically en-joyed by incumbents over their challengers.Table 9 lists the proportion of voters in thesample able to remember a candidate's name

according to the candidate's incumbency status.The incumbent's advantage here is clear. AndSenate candidates are better known, in general,than House candidates, hardly a counterintui-tive discovery. A much more telling point canbe made if information from both tables iscombined: in those instances in which thecontest is between an incumbent and a chal-lenger and only one of the candidates is known,that candidate is the incumbent 96 percent ofthe time in all three election sets. When theeffects of awareness are most pronounced, the

advantage is almost entirely to the incumbent.Challengers and other nonincumbents clearly

have more to gain by vigorous campaigning. Ingeneral, then, the more nonincumbents spend,the greater should be their saliency. Spendingby incumbents should, by comparison, have aweaker effect on how well they are known.This, if true, would provide an explanatory linkbetween the observed aggregate effects of cam-paign spending by incumbents and challengersand the survey findings on voting behavior incongressional elections.

The relationship between campaign spendingand candidate saliency may, of course, alsoinvolve reciprocal causation. Candidates whoare well known are able to raise more money,which, spent judiciously, increases their renowneven further. If this is the case, a model of theserelationships should consist of two simul-taneous equations with two endogenous vari-ables. Theoretically, candidate saliency is ex-pected to be a function of campaign spending

plus some exogenous variables, while campaignspending is, in turn, a function of candidatesaliency and some exogenous variables. Identifi-cation of the equations (and thus the possibilityof estimating their parameters) depends on the

available exogenous variables and the assump-tions that can be made about them. Consider apreliminary specification:

CR =a+bjE+b2PO +

cl ... 5X1 ... 5 +e (7.1)

E=a +b1CR +b2PO +

b3NPS + b4NI + e (7.2)

where

CR is candidate recall, measured as 1 if therespondent remembers the candidate'sname, 0 otherwise12

E is the nonincumbent candidate's campaignexpenditures (in thousands of dollars)

PO is 1 if the candidate has previously heldelective office, 0 otherwise

NPS is the strength of the nonincumbentcandidate's party in the constituency(measured as was CPS in the first section)

NI is 1 if the nonincumbent is runningagainst another nonincumbent, 0 if op-posing an incumbent

X1 ... 5 are respondent variables: socialclass, education, attentiveness to the massmedia, political interest, and whether ornot the respondent shares the candidate'spartisan affiliation.13

12The data are from the SRC surveys for 1972 and1974; the 1974 sample is weighted.

13The construction of these variables is reported inJacobson (1976).

Table 9. Percent of Voters Aware of Candidates by Incumbency Status

Percent of Voters Recalling Name of Candidates Who Were:Neither

Incumbents Challengers (Open Seat)

House Elections1972 50 27 41

(498)a (498) (220)

1974 57 31 35(856) (856) (374)

Senate Elections1974 73 44 54(595) (595) (428)

aNumber of cases from which percentages were computed. The 1974 sample is weighted.Source: The 1972 and 1974 SRC surveys.

7/30/2019 Jacobson 1978

14/24


The a's are intercepts, the b's and c's regressioncoefficients, and the e's the error terms.

The reasoning underlying this specificationshould be apparent. Observations on all nonin-cumbents, not merely challengers, can be in-cluded because, theoretically, spending shouldaffect popular awareness of all nonincumbentcandidates in about the same way. The smallquantity of empirical evidence available sug-gests that it does (Jacobson, 1976). This alsoaugments the number of observations, par-ticularly at the upper end of the expenditurescale. Whether or not respondents remembernames of candidates depends on how muchcandidates spend, their prior political exposure,and characteristics of the respondents them-selves.

Candidates' spending levels depend on howwell they are known, their prior politicalexperience, which party they belong to (es-pecially important in a year like 1974), thestrength of that party in their districts, andwhether or not they are running against incum-bents. The variables P, NPS, and NI are ex-pected to affect the ability to raise money-primarily because they are closely related topresumed chances of victory-but should not,in theory, affect the likelihood that a voter willrecognize a candidate independently of the

voter's individual partisan orientation. Equation7.2 is similar to equation 5.1 in the firstsection; NI replaces YRS as one measure of theeffects of incumbency (necessarily because notall nonincumbents are challengers), and theprimary election variables are dropped becausethey had no statistically significant effect inany of the regressions examined.

Both equations are identified (overidenti-fied, in fact), but this specification is unsatis-factory. The first equation involves a cate-gorical dependent variable and therefore raises

some difficulties to be addressed shortly. Be-fore we proceed to that, we will find it helpfulto simplify the equation. The first simplifica-tion is to ignore the respondent variablesXI ... 5. Although some of these variables areindeed related to the likelihood that the respon-dent will remember a candidate's name,14 noneis correlated with the other independent vari-ables as high at . 1, so their omission should notaffect the regression coefficients of those vari-

14The only respondent variable that has a con-sistent and significant impact on candidate recall isattentiveness to the mass media; voters who follow theelection regularly in at least one mass medium aresignificantly more likely to remember the candidates'names. See Jacobson (1976).

ables (Kmenta, 1971, pp. 392-95). Further-more, we can assume that b2 = 0 in equation7.1. Although on theoretical grounds we mightexpect that candidates who have held priorpolitical office would have a greater probabilityof being known by voters, empirically this doesnot seem to be true. If spending is taken intoaccount, the relationship between this variableand candidate recognition, weak to begin with,disappears. 1 5

Even though the dependent variable in equa-tion 7.1 actually takes only two values, 1 and 0,the equation can be interpreted as estimatingthe conditional probability that a respondentremembers a candidate's name. A problem isthat least squares estimates of the parametersmay predict values of more than 1.0 or lessthan 0.0 for this probability for some observa-tions. In addition, the error term cannot have azero expectation, invalidating one of the as-sumptions required for unbiased least squaresestimation (Theil, 1971, pp. 632-33). Logitanalysis avoids these difficulties. The odds on avoter's knowing a candidate are defined asPcr(1 -Pcr), where Pcr is the probability that avoter knows the candidate. This term can takeany value from zero to infinity; a logarithmictransformation of the term, 1n(PcrI(1-Pcr))restricts the possible values of Pcr to a range offrom 0 to 1 as the transformed term varies fromminus infinity to plus infinity (Theil, 1971, p.632). The transformed term, designated Lcr,replaces CR in equation 7.1.

In order to get observations for Pcr it isnecessary to group the survey observations andestimate Pcr as Fer, the proportionate frequen-cy with which respondents in each groupremember the candidate's name. The standardprocedure, followed in this instance, is to groupobservations on intervals of the independentvariables. For each group the average expendi-ture (in thousands of dollars) is calculated alongwith the recognition frequency. The equation isnow

Lcr=a+bIE+e. (8.1)

Since the groups providing observations for thisequation are of different sizes, the errors areheteroscedastic, and so weighted least squares

15The simple correlations between the previousoffice variable (PO) and the candidate recall variable(CR) for 1974 Senate and 1972 and 1974 Housevoters are, respectively, .11, .09, and .13; withcampaign spending controlled, the corresponding ar-tial correlations re .01, .03, and -.01.

7/30/2019 Jacobson 1978

15/24


are used to compute the estimates.16Ordinary (weighted) least squares estimates

of equation 8.1 would still be biased andinconsistent because it is part of a simultaneousequation system in which E is assumed to be

endogenous and therefore not independent ofthe error term. The 2SLS procedure is again inorder. First E is regressed on the exogenousvariables,

E* = a + b2PO + b3P + b4NPS +b5NI + e (9.la)

and the results are used to compute expectedvalues, E*, for each observation. Groupingthese observations on intervals of values of E*and taking the recognition frequency and meanvalue of E* (E*) for each group, the secondstage equation,

Lcr=a+biE*+e (9.2)

may be estimated using OLS.This was the procedure followed for the two

House election sets. For the Senate elections,the grouping was done by states, and theobservations on expenditures and the 2SLSinstruments for spending are therefore notaverages but rather figures for the separatecontests in each individual state. Because thestates vary widely in population and hencepresumably in the cost of conducting an equiva-lent campaign, expenditures were divided bythe voting-age population in each state andwere entered as cents per voting-age individual(EPV). But per-voter spending declines as thesize of the population increases (campaignspending enjoys economies to scale); therefore,the equation estimating Senate campaign spend-ing includes as a conditioning variable thenatural logarithm of the voting-age population(in thousands). It replaces NPS, which wasdropped because it had no effect whatever onspending when the other variables were con-

16Grouping the observations on interval values ofthe independent variable minimizes the loss of ef-ficiency engendered by grouping. The groups are ofdifferent sizes because fixed intervals of the explana-tory variable were employed in their formulation. Forthis reason, equations using the grouped observationswere estimated by weighted least squares, the weightsbeing proportional to the reciprocal of the approxi-mate standard deviation of the error term, e, wherethe variance of e is estimated as lI(NFcr(l -Fcr)), Nbeing the number of observations in a group and Fcr

the proportionate frequency a candidate is known forthat group. The procedure is from Theil (1971, p.635). The number of observations (groups) in theseveral logit equations varies from 13 to 29. The meannumber of cases in the Senate groups is 34; that in theHouse groups, 61.

trolled.17 The first-stage equation for Senateelections is thus

EPV* = a + b2PO + b3P +b4lnVAP+b5NI+e (9.lb)

where EPV is spending in cents per voting-ageindividual, In VAP is the natural log of thevoting-age population (in thousands), and theother variables and coefficients are as definedpreviously.

The regression estimates for equations 9.1a,9.1b, 9.2, and, for comparison, 8.1 are listed inTable 10. Equation 8.1 was also used toestimate the equivalent parameters or incum-bents in the three election sets and these alsoappear n the table. Results n this form are noteasy to interpret, so they are displayed

graphically n the three figures. The estimatedcurve for 1972 House candidates s shown forspending up to $160,000, that for 1974 up to$200,000, and the scales are adjusted to assurecomparability; nflation must be acknowledged.

The coefficients of determination R2's) andthe standard errors are not listed for the logitequations because they are rendered meaning-less by the grouping. Estimates based on indivi-dual rather than grouped data (from OLS andreduced form versions of equation 7.1 indicatethat the relationship between spending andcandidate saliency is significant at .001 fornonincumbents in all election sets and forincumbents in the 1974 elections. The relation-ship for 1972 House incumbents is notstatistically significant even when the data aregrouped for the logit analysis.

The hypothesis that campaign spending ismore useful to nonincumbents than to incum-bents because of its greater effect on howfrequently they are remembered by voters is, ingeneral, well supported by these data. Theevidence is strongest for the 1972 Houseelections; the amount spent by incumbents had

17NPSIfor Senate elections was measured as ex-plained in footnote 10. This variable, when included,was not a statistically significant determinant ofspending and had a perverse sign. One extreme casehad to be dropped from both the incumbent andnonincumbent Senate groups. This was South Dakota,

where the challenger spent twice as much per eligiblevoter as any other nonincumbent and the incumbentmore than four times as much as any other incumbentin the states covered by the survey (N=24). Bothcandidates were known by all the voters surveyed inthe state (N=26).

7/30/2019 Jacobson 1978

16/24


little apparent effect on the probability theywould be remembered; the expected gain isonly .02 as spending increases from $0 to$160,000 whereas awareness of nonincumbentsmore than doubles over the same range of

expenditures. Those few nonincumbents for-tunate enough to spend over $150,000 were aslikely to be remembered as incumbents.

In 1974, spending by both incumbents andnonincumbents had a positive effect on theprobability that voters would remember them.Nonincumbents did benefit more than incum-bents from the same amount of spending, butthe difference is not so great as it was in 1972;they gain about .13 more than incumbents asspending increases from $0 to $200,000. We arereminded that 1974 was an unusual year for

Republican Incumbents. Four whose districtswere covered in the survey spent over$200,000; three of them had also spent over

$100,000 in 1972; their collective frequency ofrecall by voters was .93 (N=29). Very highlevels of spending evidently do make a dif-ference, even for incumbents. Parenthetically,three of them lost and the fourth squeaked by

with 51.1 percent of the vote.18 Saliency isobviously not the only important factor deter-mining candidate success (a point ably arguedby Abramowitz, 1975).

18The candidates were Samuel Young, Illinois10th, spending $251,200 in 1974 and $206,166 in1972; William Hudnut III, Indiana 11th, with$201,700 in 1974 and $163,442 in 1972; Joel T.Broyhill, Virginia 10th, with $248,700 in 1974 and

$141,290 in 1972; and Sam Steiger, Arizona 3rd, whospent $203,900 in 1974, but only $37,691 in 1972.Steiger was the only winner.

Table 10. OLS and 2SLS Logit Regression Equations Estimating the Effects of Campaign Spendingon Voter Awareness of House and Senate Candidates

Equation 9.1a. House Elections

1972 k*=2.28+17.11PO+5.19P+.682NPS+60.97NI(4.17)a (3.58) (.128) (4.14)

N=718 R2=.361974 E * = -46.34 + 26.37 PO + 33.27 P + 1.77 NPS + 21.62 NI

(3.22) (2.74) (.139) (3.72)

N = 976 R2 =.47

Equation 9.1b. Senate Elections

1974 EPV*= 205 +41.8PO + 12.6P+ .04NI+24.8 1nVAP(2.89) (2.51) (2.27) (1.15)

N =997 R2 = A6

Equation 9.2: 2SLS

House Elections 1972Lcr

= -1.25 +.0087 E*

1974 Lcr = -1.20 + .0104 E*

Senate Elections 1974 Lcr = -.615 + .0178 EPV*

Equation 8.1: OLS

Nonincumbents

House Llections 1972 Lcr = -1.22 + .0076 E

1974 Lcr = -1.10 + .0094 E

Senate .Jiections 1974 Lcr = -.750 + .0444 EPV

Incumbents

House Llections 1972 Lcr=

-.036+

.0005 E1974 Lcr =-.129 + .0078 E

Senate Ilections 1974 Lcr = .521 + .0209 EPV

aStandard ecror of regression coefficient.

7/30/2019 Jacobson 1978

17/24


Spending was notably more effective inincreasing the awareness of House candidates ofall kinds in 1974 as compared to 1972. Aninescapable inference is that in presidentialelection years, messages from congressional

candidates are crowded out by those comingfrom the presidential campaigns; at midterm,with less competition, congressional campaignsreach the intended audience more consistently.Information on future elections will be neces-sary to test this interpretation.

For both the 1972 and 1974 House elec-tions, the 2SLS and OLS estimates are almostidentical; by this evidence, simultaneity biaswas not a problem in the OLS estimates of therelationship. These findings suggest that thestructure here is actually recursive; spending

affects saliency, but saliency has little effect onspending.The results for Senate elections, however,

indicate that simultaneity bias was present inthe OLS estimate. The slope of the 2SLSregression coefficient for nonincumbents is

much less steep than the OLS slope; comparingthe 2SLS estimate to the OLS estimate forincumbents, the conclusion must be that spend-ing affects both groups similarly. This cor-responds to the finding reported in the first

section that Senate incumbents benefit fromtheir own campaign spending more than doHouse incumbents. No easy explanation for thisdifference comes to mind; it may have to dowith the greater prominence, intensity, ortechnological sophistication of Senate cam-paigns; an answer awaits further research.

A few points dealing specifically with incum-bents and their challengers are in order. Itshould be emphasized that even if spending hasthe same marginal effect on the ability to recallnames of challengers and incumbents, incum-

bents begin with such a great advantage insaliency that an equal increase in spending maystill benefit the challenger. For one thing, it willdecrease the proportionate advantage in aware-ness enjoyed by the incumbent. For example,incumbent senators are remembered 1.8 times

.6

.5 ~~~~~~~~~~~~~IncumbentsLS

.4OL

.30

0

~' .2

0 20 40 60 80 100 120 140 160

Spending in Thousands of Dollars

*Median expendituresI--incumbentsC--challengers

Figure .U.S.HouseOfRepresentatives 972Elections

7/30/2019 Jacobson 1978

18/24


as frequently as nonincumbents (by the 2SLSestimates) if no money is spent; the figuredrops to 1.4 at 70 cents per eligible voter eventhough the absolute gain in saliency is aboutthe same for both groups. In addition, very few

voters know a challenger without also knowingthe opposing incumbent (about 1 percent inthese surveys), while over a quarter of therespondents typically know the incumbentwithout knowing the challenger. Greater spend-ing might therefore help challengers more thanincumbents by increasing the number of in-stances in which both candidates are known.

The points at which the estimated curvesintersect the mean expenditure level for chal-lengers and incumbents are indicated on thefigures. This information reiterates what is

already known about the incumbent's spendingadvantage and displays its connection with thesaliency advantage. Clearly, challengers mustspend much more than they typically do-andmuch more than incumbents-if they hope tomatch the incumbent's saliency.

What's in a Name?

Campaign spending is important to candi-dates who need to make themselves known tovoters; voters are more likely to vote for

candidates whose names they recall. Sinceremembering a candidate's name has itself beengiven no weighty theoretical significance, itmust be interpreted as an indicator of somesort. In earlier work I have argued, withoutsupporting evidence, that candidate recognitionshould be considered a threshold indicator.That is, we should not assume that respondentswho answer positively and correctly knownothing but the name of the candidate (al-though for some this may be true), but ratherthe ability to remember a candidate's name is

best understood as a sign that the respondent"has crossed a minimal threshold essential tothe acquisition of further information and tothe elaboration of opinions about the candi-date" (Jacobson, 1976, p. 17).

.8 Incumbents OLS

.7

L~~~~~~~~~~~~~~~~~~~~~~L

.6

I--incumbents ~ ~ ~ ~ ~ ~ ~~Nnicubet

Q)

~6 .5U

w

C

.2

.1

0 20 40 60 80 100 120 140 160 180 200

*Median expenditure Spending in Thousands of DollarsI--incumbentsC--challengers

Figure 2. U.S. House of Representatives 974 Elections

7/30/2019 Jacobson 1978

19/24


More recent studies suggest that the issue ismore complicated. Abramowitz (1975) foundthat many of his respondents were quite readyto offer and amplify opinions on how incum-bent members of Congress had performed their

jobs without being able to recall their names.Ferejohn's investigation (1977) led him toconclude that incumbents are now enjoying anelectoral advantage extending beyond what canbe explained by greater familiarity to voters,since incumbents are favored even by voterswho remember neither candidate's name.19Evidently it is not necessary to rememberpoliticians' names in order to have an opinionabout them. What does name recall indicate,then? The 1974 survey contains some items

19This finding may be an artifact of the way"candidate amiliarity" s measured. Surely voters mayrecognize the incumbent's name when they see it onthe ballot without necessarily being able to recall itwhen asked by an interviewer.

through which the question can be explored.Regarding the Senate races, respondents wereasked, in addition to the name-recall question, aseries of questions that can be summarized as"Was there anything in particular about the

Democratic (Republican) candidate that madeyou want to vote for (against) him (her)? Whatwas that?"20 Responses to these questionswere first recorded simply as positive (if any-thing made the respondent want to vote for thecandidate), negative (if anything made therespondent want to vote against the candidate),or no response (the respondent mentionednothing for or against the candidate) and werecrosstabulated with responses to the recogni-tion question. The results, broken down bypartisanship, appear in Table 11.

20See variables 2177 to 2192 in the 1974 SRCsurvey codebook (Miller, Miller, and Kline, 1975, pp.105-12).

.9

t ~~~~~~IncumbentLS ;.8

.4.

-M .7

U

0 - .5 oicmet

0

.4

.3

.2 . I0 10 20 30 40 50 60 70

Spending in Cents per Voting Age Individual

*Median expenditureI--incumbentsC--challengers

Figue 3. U.S. Senate 1974 Elections

7/30/2019 Jacobson 1978

20/24


Respondents (and only those who reportedvoting are included in the sample) who knowthe names of candidates are also much morelikely to have something to say about them.Nearly two-thirds of the voters who could not

recall the candidates' names had nothing to sayabout them either. Less than one-fifth of thoseaccurately naming a candidate had no furthercomment. Although the effects of partisanshipare quite apparent, aware voters more fre-quently find something good and bad aboutboth their own and the other party's candidate.Familiarity does not invariably produce a fa-vorable evaluation by any means. The relativegain in positive evaluations associated withrecognition occurs primarily among a candi-date's fellow partisans and, to a lesser extent,

independent voters; among the other party'ssupporters the greater proportion of positiveresponses arising from awareness is more nearlymatched by the increase in proportion ofnegative responses.

From another perspective, Table 1 1 makes itclear that ignorance of a candidate's name doesnot preclude expressing an opinion about thatcandidate; a third of the voters in this categorywere willing to do so. Do respondents who donot remember the candidate's name use dif-ferent evaluative criteria than do respondents

who are aware of the candidate? A moredetailed recoding of the answers reported in thesurvey provides a way of finding out. Allpositive and negative responses (and up to threeof each were recorded by the interviewers) wereclassified as personal (those referring specifical-ly to characteristics of the candidates them-

selves), party (references to one of the partiesor to the candidates themselves exclusively aspartisans), or mixed.21 These were cross-tabulated with responses to the recognitionquestion; the results are found in Table 12. Not

surprisingly, voters who do not remembercandidates' names more readily resort to par-tisan criteria to evaluate them; personal com-ments are much more frequent among votersknowing candidates. Still, nearly a third of theunaware group employ purely personal criteriawithout remembering who the person is. No-tice, incidentally, that aware voters are morelikely to have multiple comments about thecandidate; they average 1.7 responses, theunaware group 1.2.

Two tentative conclusions are warranted.

Although the ability to remember a candidate'sname is not a precise threshold-voters areoften able to evaluate candidates without thispiece of information-there is a substantialdifference in both the frequency and characterof evaluative comments between voters who doand do not recall the candidate. And the gainsto be made from campaigning-and therebymaking oneself known to voters-derive fromgathering support among one's own partisansand independent voters rather than from con-

2IResponses coded 00, 01, and 05 in the hundredseries for these questions were considered partyreferences; those coded 02, 03, and 04 were classifiedas personal references; and those coded 06 through 12were considered mixed. See Miller, Miller, and Kline(1975, pp. 400-15) for the complete coding cate-gories.

Table 11. Voter Awareness nd Evaluation f 1974 Senate Candidates

EvaluationPositive and

Percent Evaluating: Positive Negative Negative None

Own Party's CandidateKnown 56.4 12.5 15.4 15.7 (479)aNot Known 38.1 4.6 1.5 55.8 (260)Difference 18.3 7.9 13.9 -40.1

Other Party's CandidateKnown 21.9 40.0 15.7 22.4 (402)Not Known 5.7 27.3 1.0 67.7 (300)Difference 16.2 12.7 14.7 -45.3

Independent Voters, Candidate sKnown 36.4 27.3 18.2 18.2 ( 77)Not Known 10.0 8.3 1.7 80.0 ( 66)Difference 26.4 19.0 16.5 -61.8

aNumber of cases from which percentages were computed. The sample is weighted.Source: The 1974 SRC survey.

7/30/2019 Jacobson 1978

21/24


verting the opposition. This, of course, is partof conventional wisdom.

An overview of the connection between thesimplified candidate evaluation index and thereported vote for senator in 1974 will complete

this part of the analysis. Table 13 displays thisrelationship. Evaluation of both candidates hasa noticeable impact on voting behavior. Andthe effects are as we would anticipate: positiveevaluations increase the likelihood of voting fora candidate, negative evaluations decrease that

likelihood, and the effect in either case dependsalso on the evaluation of the other candidate.Defections are therefore concentrated in theupper right-hand corner of the partisan table,party loyalty is predominant in the lower

left-hand corner. A comparable pattern occursamong independent voters.To sum up briefly, then: our evidence is that

campaign spending helps candidates, most par-ticularly nonincumbents, by bringing them tothe attention of voters. It is not the case that

Table 12. Voter Awareness nd Criteria or Evaluation f 1974 Senate Candidates

Evaluative Criteria Number of Number ofPercent of Evaluations Personal Party Mixed Comments RespondentsPositive

Knowing Candidate 60.7 8.9 30.4 957 538Not Knowing Candidate 37.1 31.1 31.7 167 134

NegativeKnowing Candidate 44.4 14.5 41.1 601 393Not Knowing Candidate 24.6 43.0 32.5 114 108Source: The 1974 SRC survey. The sample is weighted.

Table 13. Evaluation f Senate Candidates nd Voting Behavior n 1974

Percent of Partisan Voters DefectingEvaluation f Own Party's Candidate

Evaluation f Positive and MarginalOther Party's Candidate Positive Negative Negative TotalsPositive 28.6 77.1 77.1 65.9

(21)a (35) (35) (91)Positive and Negative 4.5 31.1 56.0 16.3

(139) (45) (25) (209)Negative 4.2 3.4 40.0 4.8

(167) (59) ( 5) (231)Marginal otals 5.8 30.9 66.2 19.8No Evaluation: 16.5 (115) (327) (139) (65) (531)

Independent Voters: Percent Voting for DemocratEvaluation f Democrat

Evaluation f Positive and MarginalRepublican Positive Negative Negative TotalsPositive 0.0 20.0 6.3

(1) 5) (16)Positive and Negative 88.9 66.7 0.0 58.8

(9) (3) (5) (17)Negative 100.0 83.3 100.0 88.9

( 2) (6) (1) (9)Marginal otals 90.0 35.0 18.2 45.2

(11) (20) (11) (42)No Evaluation: 40.0 (15)

aNumber of cases from which percentages were computed. The sample is weighted.Source: The 1974 SRC survey.

7/30/2019 Jacobson 1978

22/24


well-known candidates simply attract moremoney; rather, money buys attention. Voterswho are aware of candidates are also morelikely to have opinions about them, bothpositive and negative, and the net gain in

positive evaluations for candidates who dosucceed in getting the attention of voters comesprimarily from adherents of their own partyand from independents. And voters' evaluationsof the candidates strongly influence how theycast their votes. Since nonincumbents have themost to gain from campaigning, it is notsurprising that their level of spending has agreater impact on the outcomes of electionsthan does that of incumbents.

Implications forCampaign Finance Policy

The findings reported here have importantimplications for campaign finance policy re-form. For House elections, both the OLS andthe 2SLS models indicate that the marginalgains from a given increase in campaign spend-ing are much greater for challengers than forincumbents. The unmistakable conclusion to bedrawn from this is that, in general, any increasein spending by both candidates will help thechallenger. Public subsidies-or any other policywhich gets more money into the hands ofchallengers-should therefore make House elec-tions more competitive. Incumbents will alsoget more money under such circumstances, butsince for them raising money is not the problemit is for challengers and because their additionalspending does not counterbalance the effects ofgreater spending by challengers, this will notwork to their benefit.

On the other hand, any reform measurewhich decreases spending by the candidates willfavor incumbents. This includes limits on cam-paign contributions from individuals and groupsas well as ceilings on total spending by thecandidates. Even though incumbents raisemoney more easily from all sources, limits oncontributions will not help challengers becausethe problem is not equalizing spending betweencandidates but rather simply getting moremoney to challengers so that they can mountcompetitive races. Anything that makes itharder to raise campaign funds is to theirdetriment.

Ceilings on permissible spending, if theyhave any effect on it at all, can only lessencompetition. The consequences of subsidiescombined with limits-constitutional by thedecision in Buckley v. Valeo ( 1976)-depend onthe size of the subsidy provided and the limit

imposed. For example, the major public financ-ing bill before the House in the 95th Congress,HR 5157, would provide partial public fundingon a matching basis and impose a ceiling of$150,000 on general election expenditures in

House contests. Ignoring the problem that thedata include primary election spending, andadjusting for inflation, the equations can beused to estimate the hypothetical effects ofequivalent subsidies and limits in 1972 and1974. The results of this exercise have beenreported in detail elsewhere (Jacobson, 1977).They suggest that the legislation would havehad very different consequences in the twoelection years. It would have diminished theexpected number of successful challenges in1972 because the limit would have been too

low; but in 1974 the same financing systemwould have increased the number of predictedchallenger victories substantially, with an evengreater increase had the ceiling not been inforce.

The subsidies and spending limits proposedfor Senate elections under S 926, the Senatepublic funding bill, would have had slightlydifferent consequences. They would have madeno difference at all in 1972, but might haveincreased the number of successful challenges in1974. The limits are evidently set high enoughto avoid harming challengers. In 1974, the lawwould have reduced incumbent spending andincreased challenger spending, to the definitebenefit of the latter (Jacobson, 1977).

An explanation of the divergent effects ofthese laws in 1972 and 1974 doubtless lies inthe different political circumstances of the twoelection years. In 1972, despite the decisiveNixon victory, national short-term forces donot seem to have been important in Houseelections; note the coefficients on P for 1972.Quite the opposite is true of 1974. Republicanswere at a serious disadvantage because ofWatergate, the collapse of the Nixon admini-stration, and Ford's unpopular pardon of theformer president. Democrats could thereforeunseat incumbent Republicans with muchgreater ease (and with a good deal less money)than would have been the case in 1972. Almostall of the predicted 1974 winners are Demo-crats under any financing system; the partisanbalance is nearly even for 1972.

Thus public funding and spending limits ofthe kind proposed in HR 5157 and S 926would have different consequences according tovarying political circumstances. If strong short-term forces favor the candidates of one party oranother (or of different parties in differentlocalities), less money will be needed by chal-lengers of the advantaged party for them to

7/30/2019 Jacobson 1978

23/24


win; subsidies will be more effective, even inrelatively small amounts, and ceilings on spend-ing will not be particularly damaging to theirchances. In the absence of strong short-termforces, however, the subsidies may not be high

enough to help significantly, and the ceilingsmay actually diminish the possibility of un-seating an incumbent. Legislation of this kindhas the potential to exaggerate whatever trendsexist at the time of the election. Incumbentsmay be given an added margin of safety inelection years when short-term forces are weak,but when these forces are strong, the turnoverin congressional seats should be greater.

References

Abramowitz, Alan I. (1975). "Name Familiarity,Reputation, and the Incumbency Effect in Con-gressional Elections." Western Political Quarterly28:558-684.

Adamany, David W. and George Agree (1975). Politi-cal Money: A Strategy or Campaign inancing nAmerica. Baltimore and London: The Johns Hop-kins University Press.

Arsenau, Robert B. and Raymond Wolfinger 1973)."Voting Behavior in Congressional Elections."Paper presented to the annual meeting of theAmerican Political Science Association, New Or-leans, Louisiana, September -8, 1973.

Ben-Zion, Uri and Zeev Eytan (1975). "On Money,Votes, and Policy in a Democratic Society." PublicChoice 17:1-10.

Bental, Benjamin, Uri Ben-Zion, and Yair Moshel(1976). "Money n Politics-An Empirical Study."Preliminary draft for presentation at the 1976Econometric Meeting.

(1977). "Money in Politics-An EmpiricalStudy." Revised working draft presented to thePublic Choice Society, New Orleans, Louisiana,March -11, 1977.

Common Cause (1972). 1972 Congressional ampaignFinances. Washington, .C.: Common Cause.

Congressional Quarterly (1972). Congressional Quar-terly WeeklyReport (October 7):2485 -2588.(1974a). Congressional Quarterly Weekly Re-port (October 12): 2715-2815.

(1974b). Congressional istricts in the 1970's.2nd ed. Washington: Congressional Quarterly.

(1975). Congressional Quarterly Weekly Re-port (April 19):789-94.

Cover, Albert D. (1976). "One Good Term DeservesAnother: The Advantage of Incumbency n Con-gressional Elections." Paper presented to the an-nual meeting of the American Political ScienceAssociation, Chicago, September 2-5, 1976.

Dawson, Paul A. and James E. Zinser (1976). "Poli-

tical Finance and Participation n CongressionalElections." American Academy of Political andSocial Science Annals 425:59-73.

Ferejohn, John A. (1977). "On the Decline ofCompetition in Congressional Elections." Ameri-can Political Science Review 71:166-76.

Glantz, Stanton A., Alan I. Abramowitz, and MichaelP. Burkart (1976). "Election Outcomes: WhoseMoney Matters?" Journal of Politics 38:1033-38.

Hinckley, Barbara, Richard Hofstetter, and John H.Kessel (1974). "Information and the Vote: AComparative Election Study." American PoliticsQuarterly 2:131-58.

Jacobson, Gary C. (1976). "Practical Consequences ofCampaign Finance Reform: An Incumbent Protec-tion Act?" Public Policy 24:1-32.

(1977). "The Electoral Consequences of PublicSubsidies for Congressional Campaigns." Paperpresented to the annual meeting of the AmericanPolitical Science Association, Washington, D.C.,September 1-4, 1977.

Johnston, J. (1972). Econometric Methods. 2nd ed.New York: McGraw-Hill.

Kmenta, Jan (1971). Elements of Econometrics. NewYork: MacMillan.

Kostroski, Warren Lee (1973). "Party and Incum-bency in Postwar Senate Elections: Trends, Pat-terns, and Models." American Political ScienceReview 67:1213-34.

Lott, William F. and P. D. Warner III (1974). "TheRelative Importance of Campaign Expenditures:An Application of Production Theory." Qualityand Quantity 8:99-105.

Mayhew, David R. (1974a). "Congressional Elections:The Case of the Vanishing Marginals." Polity6:295-317.

(1974b). Congress: The Electoral Connection.New Haven: Yale University Press.

Miller, Warren E., Arthur H. Miller, and F. GeraldKline, Principal Investigators (1975). The CPS1974 American National Election Study. ICPREdition. Ann Arbor, Michigan: The Center forPolitical Studies, the University of Michigan.

Palda, Kristian S. (1973). "Does Advertising InfluenceVotes? An Analysis of the 1966 and 1970 QuebecElections." Canadian Journal of Political Science6:638-55.

(1975). "The Effect of Expenditure on Poli-tical Success." Journal of Law and Economics18:745-71.

Perdue, Lewis (1977). "The Million-Dollar Advantageof Incumbency." Washington Monthly 9:50-54.

Rao, Potluri and Roger L. Miller (1971). AppliedEconometrics. Belmont, Calif.: Wadsworth.

Ripley, Randall B. (1975). Congress: Process andPolicy. New York: Norton.

Scammon, Richard M. (1975). America Votes 11: AHandbook of Contemporary American ElectionStatistics. Washington: Congressional Quarterly.

Silberman, Jonathan (1976). "A Comment on theEconomics of Campaign Funds." Public Choice25:69-73.

and Gilbert Yochum (1977). "CampaignFunds and the Election Process." Paper presentedto the Public Choice Society, New Orleans, Louisi-ana, March 11-13, 1977.

Stokes, Donald E. and Warren E. Miller (1962). "PartyGovernment and the Saliency of Congress."PublicOpinion Quarterly 26:531-46.

7/30/2019 Jacobson 1978

24/24


Theil, Henri (1971). Principles of Econometrics. NewYork: Wiley.

United States Congress (1971). Federal ElectionsCampaign Act of 1971: Appendix A. Senate,Committee on Commerce, Subcommittee on Com-munications.

(1973). Federal Elections Campaign Act of1973: Appendix A. Senate, Committee on Com-merce, Subcommittee on Communications.

(1974). Congressional Record. 93rd Congress,2nd Session. Vol. 120, Pt. 14 (154):18542.

(1977). Federal Election Reform Proposals of

1977. Senate, Committee on Rules and Admini-stration.

Welch, William P. (1974). "The Economics of Cam-paign Funds." Public Choice 20:83 -9 7.

(1976). "The Effectiveness of Expenditures in

State Legislative Races." American Politics Quar-terly 4:333-56.(1977). "Money and Votes: A Simultaneous

Equation Model." Paper presented to the PublicChoice Society, New Orleans, Louisiana, March11-13, 1977.

Documents

Jacobson 1978