Relative Unemployment and Gubernatorial Popularity
Author(s): Jeffrey E. Cohen and James D. King
Source: The Journal of Politics, Vol. 66, No. 4, (Nov., 2004), pp. 1267-1282
Published by: Cambridge University Press on behalf of the Southern Political Science
Association
Stable URL: http://www.jstor.org/stable/3449538
Accessed: 09/06/2008 21:36
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may 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 at
http://www.jstor.org/action/showPublisher?publisherCode=cup.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We enable the
scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that
promotes the discovery and use of these resources. For more information about JSTOR, please contact [email protected].
http://www.jstor.org
Relative Unemploymentand GubernatorialPopularity
Jeffrey E. Cohen
FordhamUniversity
James D. King
Universityof Wyoming
Considerabledebate exists over whether the public holds the governor accountable for the state's
economy. Partof the controversystems from weak design and limitationsin data, but part also from
weakness in theory.We arguethat voters recognize the limitationsof state governmentsto affect the
state economy and that they judge their governorson the state's unemploymentsituationrelative to
the unemploymentsituationof the nationaleconomy.To test this theory we use the Official State Job
ApprovalRatings (JAR) database.Our analysis finds that citizens grade the governor'sjob performance consistent with our theory:no matterthe level of unemploymentin the state, when state unemployment is lower than the national average, governors are rewarded;when it is higher than the
nationalaverage, governorsare punished with lower approvallevels.
Relative Unemploymentand GubernatorialPopularity
Job approvalratingsnow are seemingly ubiquitousin Americanpolitics. Most
researchon job approvalfocuses on the president;we possess comparativelylittle
research on gubernatorialapproval,the topic of this paper.As with presidents,
job approvalis importantto governors. Job approvalratings provide governors
with clues about their reelection prospects;research indicates that job approval
affects gubernatorialelections (Kenney and Rice 1983; King 2001; Svoboda
1995). CaliforniaGovernorGrayDavis's low approvalratings,arguably,emboldened his opponentsin their successful recall efforts in 2003. Approvalalso provides governorswith a resourceto motivatelegislators and bureaucratsto accept
their policy initiatives.Researchfinds that governorswith higher approvallevels
receive more supportfrom the legislature(Crew 1998; Ferguson2003) and from
bureaucrats(Dometrius2002).
Job approval also provides a mechanism for the public to hold governors
accountable.Throughapprovalpolls, the public can voice its supportor opposition to the governorbetween elections, and governorsmay altertheirpolicies and
behavior in office in reaction to their approvallevel. But considerable debate
exists over whether the public actually holds governorsaccountablein approval
THE JOURNAL OF POLITICS, Vol. 66, No. 4, November 2004, Pp. 1267-1282
? 2004 Southern Political Science Association
1268
JeffreyE. Cohen and James D. King
polls for policies and issues over which governorsclaim some responsibility,like
jobs and unemployment.
Specifically, governors take credit for the state economy when it is healthy,
pointingto theiractions andpolicies in office as evidence of their economic stewardshipand publicizing their efforts to improve the economic climate and performance of the state. Factoryopenings or expansions, announcementsof a new
employer coming to the state, and the like are often greeted with great fanfare,
photo opportunities,and media events, all designed to create positive publicity
for the incumbentgovernor'seconomic leadership.And electoral challengersto
sittinggovernorsoften runcampaignsthattargetthe incumbent'seconomic policy
failures. It is, as Grady observes, "good politics for the governor to be seen
aggressively engaged in fostering [economic] growth"(1991, 109).
Despite such an environment,considerable debate exists as to whether the
public holds the governoraccountablefor the state'seconomy.Crew et al. (2002),
Hansen (1999a, 1999b), Howell and Vanderleeuw(1990), and Orth (2001) find
that state unemploymentlowers gubernatorialapproval.In contrast, Crew and
Weiher(1996) findno impact,Adams and Squire(2001) detect only spottyeffects
of state unemploymenton gubernatorialapproval,and MacDonaldand Sigelman
(1999) unearthno relationshipbetween state economic health and gubernatorial
popularity.All told, the literatureoffers a mixed picture of the effect of the state
economy on gubernatorialapproval.
Several factorsmay account for these divergentfindings.First, studies use different samples of states. Designs also differ,with some using time series (Adams
and Squire2001; CrewandWeiher1996; Crewet al. 2002; Hansen 1999b), others
conduct cross-sectionalanalyses with states as the units of analysis (MacDonald
and Sigelman 1999), while still others analyze individual-level survey data
(Cohen 1983; Howell and Vanderleeuw1990; Orth 2001). Studies also differ in
their model specifications. Some control for national economic conditions
(Crew and Weiher 1996; Crew et al. 2002; Hansen 1999a, 1999b; Howell and
Vanderleeuw 1990; Orth 2001), while others do not (Adams and Squire 2001;
MacDonald and Sigelman 1999). Political variables, such as presidential
approval,political predispositions,and time in office are also used inconsistently.
Rarely is presidentialpopularityincluded (but see Crew and Weiher 1996; Crew
et al. 2002; Orth2001), althoughstudies reportthat presidentialapprovalaffects
gubernatorialelections (e.g., Atkeson and Partin 1995; Carseyand Wright 1998;
King 2001; Simon 1989). Moreover,the theoreticallinkage between state unemployment and gubernatorialpopularityis not well specified.
We revisit the issue of the impact of state economic conditions on gubernatorial approval.We use the Official State Job Approval Ratings (JAR) database
(Beyle, Niemi, and Sigelman 2002), which provides approximately2000 gubernatorialapprovalreadings across all 50 states over 20 years. We also employ a
more comprehensiveset of control variables to guard against spurious results.
Plus, we develop a theory of accountabilitythat recognizes that governorsserve
in a complex federal system in which national economic policies and perform-
Relative Unemploymentand GubernatorialPopularity
1269
ance affect state-level economic performance.The public, in our model, grades
the governornot by the absolute level of the economy,but by how well the state
economy performs relative to the national economy. In the next section, we
develop our theoreticalmodel. Then we introducethe data, discuss our variables,
and presentthe data analysis.
State EconomicConditionsand Gubernatorial
Approval
A large literaturedemonstratesthat the American public holds its political
leaders accountablefor economic performance.Most of these studies focus on
the president and demonstratethat presidential approval cycles with the performance of the macroeconomy and with people's attitudes about the national
economy (Erikson, MacKuen, and Stimson 2002; Gronke and Newman 2003).
We assume the public wants to hold the governoraccountablefor state economic
performance,but the federal context complicates mattersfor the public.
The first complication is that state economies are to some degree integrated
into the national economy. While some evidence suggests that state economies
are becoming more independent of national economic forces (Brace 1993;
Hendrickand Garand1991), there is little doubt that national economic conditions and policies affect a state's economy. Furthermore,state policy makers do
not possess the necessary tools to influence all types of economic conditions.
Most notably,state policy makershave little ability to manage a state's inflation
rate or money supply. Consequently,state policy makers possess only limited
means of affecting their state's economic performance.
We assume that the public takes into account these two limitations-the lack
of state economic autonomy and state policy makers' limited economic policy
toolkit. Still, governorsare the most visible political leadersin their states (Squire
and Fastnow1994), and they often focus people's attentionon theirability to generatejobs for the state's economy and to maintaina healthyjob climate. Herzik
(1991) found two-fifths of governorsmentioningeconomic developmentin their
1988 state-of-the-stateaddresses;only educationwas mentionedmore frequently.
Many states have establishedoffices to attractjobs and help employers,and state
industrialpolicy initiativesoften focus on attractingnew employers(Brace 1993;
Grady 1991). When it comes to job retentionor expansion, states often compete
with each other. During economic expansion, employers must make decisions,
not only abouthow many new employees to hire, but where they will work. Such
locational decisions fosterjobs competitionamong states.
This context leads us to hypothesize that state publics will hold governors
accountablefor the employment situation in the state, but the absolute level of
unemploymentwill be less relevantto state voters than the level of state employment comparedto the national unemploymentlevel, the relative level of unemployment. Thus, in good times, when the absolute unemploymentlevel is low,
public support for the governor will erode if the state's unemployment level
supersedesthe nationallevel. And in badtimes, when stateunemploymentis high,
1270
Jeffrey E. Cohen and James D. King
governors should receive an approval boost if the state level falls below the
national level.1
This hypothesis assumes a somewhat sophisticatedstate electorate, which is
contraryto much research on public opinion. However, research suggests that
state publics reasonablydiscriminatebetween the state and national economies
(Niemi, Bremer,and Heel 1999) and thatthey arereasonablyknowledgeablewith
regardto unemploymentrates (Holbrookand Garand1996). Limited evidence of
voters distinguishingbetween nationaland state economic conditions is found in
King's (1999) analysis showing Wyoming voters evaluatingstate economic performancemore negativelyat a time when the state'seconomy was mired in recession during the national economic boom of the 1990s. Furthermore,voters in
Wyoming acted on this distinction,as evaluationof state economic performance
significantlyinfluencedvote choice for governorwhile evaluationof nationaleconomic performancedid not. Finally,some researcharguesthatthe public is more
responsiveto changes in unemploymentrates than to changes in other economic
indicators(Conover,Feldman,and Knight 1986; Haller and Norpoth 1997).2
Two studies employ variables that resemble our relative unemployment
concept. Hansen (1999b) uses a variable that measures the difference between
national and state unemployment,but does not find it to be statisticallysignificant.3Crew et al. (2002) set the problemup slightly differentlythanwe do, using
the level of state unemploymentand the difference between national and state
unemployment.As we detail below, we too use the differencebetween state and
national unemployment,but we control for national unemploymentratherthan
state unemployment.While Crew et al. (2002) find that the level of state unemployment is consistently significant,the difference variableis not. It is not clear
why state publics would punish or reward the governor when the national
economy is performingbetter or worse than the state economy.
'Another hypothesis is that the state electoratecomparescurrentunemploymentlevels today with
past levels. Unfortunately,we cannottest this hypothesiswith the JARdatabecause we do not possess
enough contiguous data points for gubernatorialpopularity.Thus, we were forced to keep gubernatorial approvalin levels form. We experimentedwith change indexes for the unemploymentdata,but
results were far from satisfying because of the cyclical natureof economic data. More often than not,
when we use change economic variables in estimation, we get contrary signs (increases in unemployment leads to higher approval).The reason for this is the cyclical natureof the economic data.
When the economy is at the top of a cycle, with low unemployment,gubernatorialapprovalis likely
to be at its peak. But at the same time, because the economy is at the top of a cycle, there is a greater
likelihood of economic decline, which will producehigherunemployment.Thus, we tend to find that
change in unemploymentis often associated with higher levels of gubernatorialapproval.If we had
enough gubernatorialapprovaldata, we could correct for such cyclical effects by creating change
scores for the approvaldata. Consequently,we retainthe levels analysis, which producesresults for
the economic variablesthat make sense-high unemploymentis associated with low approval,etc.
2
However,a large literaturefinds that aggregateinflation,ratherthanunemployment,affects trends
in presidentialapproval(see Gronkeand Newman 2003).
3This conclusion is based on an
analysis that pools data from eight states. In a separateanalysis
focusing only on California,Hansen (1999a) finds the relativeunemploymentvariableto be significant. However,Hansen spends little time discussing the theoreticalbasis for using this variable.
Relative Unemploymentand GubernatorialPopularity
1271
We formally define relative unemployment as the one-month lagged state
unemploymentpercentage minus the one-month lagged national unemployment
percentage. By applying a control for national unemployment,we can assess
whetherthe public rewards(or punishes) the governorwhen state unemployment
is lower (or higher) than national unemployment.4We cannot easily enter both
state and national at the same time because of their high correlation(r = .64,
p < .0001). Doing so produces multicollinearityand considerable coefficient
instability.5As our theoreticalmodel focuses on the unemploymentsituation of
the state relative to the nation, theory calls for controls for the national unemployment level, ratherthan for the state level. The national unemploymentlevel
serves as a base against which voters'judge the state'sperformance.We include
national unemploymentto fully specify the voters' decision calculus.6
The OfficialState Job ApprovalData Set
Data limitationshave plagued researchon gubernatorialapproval.The paucity
of data and the lack of comparabilityin data sets and designs may be one source
of the differencesin findingsacross studies.Data limitations,too, may force compromises on estimated models, leading to model misspecification, which may
result in spuriousand/orsuppressoreffects. We use the most comprehensivecollection of data on gubernatorialapprovalyet gathered,the Official State Job
ApprovalRatings (JAR) project.7This project collected state-leveljob approval
ratings for governors from numerous survey organizationsfrom 1947 through
2001. All told, the JAR project supplies us with approximately2,000 readingsof
state-levelgubernatorialpopularity.We focus on the datafrom 1980 through2001
because of difficulty in collecting data on some variablesof interest for earlier
years and because the overwhelmingnumberof cases come from this period.8
4 It makes little differencewhetherwe use
lagged or nonlagged forms of these variables.Analysis,
not shown, produces similar results.
5 State (national) income also
highly correlateswith state (national)unemploymentand thus will
also producemulticollinearity.
6The state-nationalunemployment difference is correlated with state unemployment (r = .76,
p < .0001), but is not relatedto nationalunemployment(r = -.0001, p > .99).
7This datawere collected by Beyle, Niemi, and Sigelman and fundedby the NSF (GrantNo. SES9974176). We thankthem for allowing us to use the data.
8Some 76 cases date
priorto 1980, almost all from Californiaand Iowa. Othercases are lost due
to missing or incomplete data.For instance,we requiremonthly observations,but JARpoll data does
not always provide the month of the poll. We also lose some cases when a presidentialtransition
occurs, because we code the lagged presidentialpopularityvariableas missing data duringthe transition month (January)from one presidentto another.All told, we have complete and reliable data
on 2,016 cases. The data are not evenly distributedacross states, yet no subset of states dominates
the data set either.The top five states, (CA, CT, NY, AL, FL), each with 5% or more of the cases,
account for only 29.5% of cases. California,the leader, accounts for only 7%. The standarddeviation of the percentage of cases per state is also quite tight at 1.6. The JOP Web site (http://www.
journalofpolitics.org)presents more detailed informationon the distributionof cases by state.
1272
JeffreyE. Cohen and James D. King
There are several complicationsin using the JAR data. First,because the data
come from a varietyof datasources andpolling organizations,dataqualityvaries.
For instance, sometimes only positive evaluationsof the governorwere available
(Beyle, Niemi, and Sigelman2002) andthe percentageof"don't know"responses
appears to vary because of the large variety of question wordings and other
factors. In orderto retain as many cases as possible, we need to make the data
as comparableas possible and compensate when comparabilityis at issue. To
correctfor the variabilityin "don'tknow"responses,we use the formula"Percent
Positive / (Percent Positive + Percent Negative)" to measure gubernatorial
approval.9On average, governorsreceived a 55% approvalrating, which varies
widely from a low of 10 to a high of 97, with a standarddeviation of 16. Our
dependentvariableis this adjustedgubernatorialpopularityscore.
Second, this data set combines numerous variations on the general approval
question.10Some questionspresumablyask about gubernatorialjob performance
in a mannersimilarto the GallupPoll's presidentialapprovalquestion;othersask
how much the respondentlikes the governor.Exact question wordings are not
available, but the response categories (e.g., approval versus disapproval,like
versus dislike) are. Beyle, Niemi, and Sigelman (2002) find the likelihood of a
negative ratingrises as the numberof response categories increase.Third,house
effects may be present because the data set combines polls across many firms.
To illustratethe potential impact of these factors, we created dummy variables
when 30 or more observations from a single polling firm exist, a total of 41
dummies, as well as three dummies for response categories (for three, four, or
five response categories). A regression analysis (not shown) of gubernatorial
approvalon three response category dummies and polling firm dummies produced an R2 of .31. Each of the response category dummies is statisticallysignificant as are 25 of the firm dummies.11
ControlVariables
More than relative unemploymentmay affect citizens' approvalof their governor.The literatureson gubernatorialapprovaland elections find that state and
national factors affect gubernatorialapprovaland voting. From these literatures
we identify relevant control variables, which we arrange into two sets, state
factors and national factors.
9Beyle, Niemi, and Sigelman (2002) also recommendthis correction.
10The data set contains
generalapprovalas well as policy specific approvalquestions.We only use
the general approvalquestions.
1In several instances, more than one firm asked a gubernatorialevaluationquestion of a state's
citizens. We decided to keep both readingsin the data set as separatedatapoints because we have no
assuredway to combine them into one reading.
Relative Unemploymentand GubernatorialPopularity
1273
Other State Factors and Gubernatorial Approval
STATE PARTISANAND IDEOLOGICAL
COMPOSITION.The structural political
context of the state may affect the governor'spopularity.Strong evidence exists
thatpresidentialcopartisans'and those whose ideological leanings correspondto
the president'sare more likely to judge the president approvinglythan disapprovingly(Bond and Fleisher2001). Agreementwith the executive'spolicies may
account for this effect. We expect a similar dynamic to occur at the state level.
Everything else being equal, we expect governors to find more support from
membersof theirown partythanfrom oppositionpartymembersand frompeople
who hold similarideological leanings than from people who hold divergentones.
Hence, we expect gubernatorialpopularityto co-vary positively with the partisan or ideological advantagein the mass public.
To measure partisan advantage,we use the updated Erikson-Wright-McIver
(EWM) state partisanshipdata.12Following EWM, we calculate the percentage
of Democratic and Republican identifiers for subsets of years (1977-1984,
1985-1992, and 1993-2001). The subsets are long enough to protect against
short-term variation that may result from sampling and other error, but still
remain relatively close in time to the measurementof the governor'spopularity.
We calculate partisan advantageas: (Percentageof GubernatorialParty Identifiers) - (Percentageof OppositionPartyIdentifiers).
We use ideological identificationto indicate agreementor disagreementwith
the governor'spolicies, assuming that liberal identifiersare more likely to agree
with the policies of Democrats than Republicans,while the opposite will hold
true for conservativeidentifiers.Again, we use the updatedEWM dataand define
net ideological advantagefor Democratic governors as: (Percentageof Liberal
Identifiers)- (Percentageof ConservativeIdentifiers)and the reverse for Republican governors.For independentgovernors,the partisanand ideological scores
are set to "0."
The partisanadvantageargumentalso implies that independentgovernorsmay
have a hardertime rallying public supportthan partisangovernors,who possess
a ready-madecore of supporters.13Thus, we add a dummy variablefor whether
the governorwas an independent,expecting independentsto have lower approval
levels. Lastly,we include a Republicandummyvariableto distinguishany other
unmeasuredpartisaneffects.
PERFORMANCE.Governors may also possess a personal support coaliELECTION
tion, which may cross party and ideological lines. Gubernatorialelection campaigns provide opportunities for governors to build such a personal base of
12
These data are from McIver's web page: http://socsci.colorado.edu/~mciverj/wip.html.See
Erikson,Wright,and Mclver 1993.
13 Moreover,the maverickbehaviorof
many independents(e.g., Jesse Ventura)may alienatevoters.
Many independents came to office in three-way contests, and thus, rarely enjoyed majority vote
supportin winning the election.
1274
JeffreyE. Cohen and James D. King
supportthat can translateinto votes at reelection time (Adams and Kenny 1989)
and likely into positive assessments in opinion polls. To measure this personal
supportbase we use the percentage of the total vote that the governorreceived
in the last election.
GOVERNING
DIFFICULTY:
STATEDIVERSITY
ANDSIZE. Some states may be harder
to governthan others.Policies are likely to be more popularwhen large segments
of the populationare like-mindedand homogeneous.As the populationbecomes
more dissimilar,some segment of the population is likely to dislike the policy.
Although severepolitical divisions may appearin stateswith few but deep socioeconomic distinctions,in general,more heterogeneousstates are likely to possess
greaterlevels of dissimilaritythan homogeneous ones. We measureheterogeneity-homogeneitywith two variables,an index of socioculturaldiversity(Morgan
and Wilson 1990) and populationsize. As diversityand populationsize increase,
we expect gubernatorialpopularityto fall.'4
OFTENURE.Several studies detect a temporalpatternto gubernatorial
LENGTH
approval (Beyle, Niemi, and Sigelman 2002; Crew and Weiher 1996; Hansen
1999b), as do many studies of presidentialpopularity(Brace and Hinckley 1991;
Mueller 1970; Stimson 1976).15Several differenttheoreticalaccounts have been
offered for approval'stemporal path, including Mueller's (1970) coalition of
minorities'notion, Stimson's(1976) cycle of expectationand disillusionment,and
Brace and Hinckley's (1991) cycle of deflatingexpectations.We cannot resolve
these issues here but we need to control for the possible effect of time in office
on approval.Accordingly,we test both the linear and cyclical models on gubernatorial popularity,measuring length of tenure as months in office. The linear
model predicts a negative impact of tenure on approval.The cyclical model predicts the same, but furtherpredictsthat as tenureextends, approvalwill increase.
To test this cyclical model, we use tenure in months squared,which we hypothesize to have a positive sign.
UNITED
GOVERNMENT.
Lastly,divided governmentmay affect public approvalof
the governor.When the opposition party holds at least one house of the legislature, tensions between the governor and the legislature are likely to rise. Open
conflict between the executive and legislaturemay underminepublic confidence
in gubernatorialleadership.But Nicholson, Segura,and Woods (2002) arguethat
presidents (executives) will garnerhigher popularityunder divided than unified
government.Under divided government,the executive and legislaturecan point
the fingerof blame at each other,makingit difficultfor the public to assess blame.
As a result, some people will blame the legislatureand others the executive. But
'4Gronke(2001) notes that population size and diversity are conceptuallydistinguishable,but in
this data set the two are highly correlated(r = .50, p = .000).
5Kernell
(1978) argues that time is atheoreticand tries to model it substantively.
Relative Unemploymentand GubernatorialPopularity
1275
when governmentis united,the executivewill bear the full bruntof public outcry
and criticism,being unableto deflectblame onto a legislatureof his party.Lowry,
Alt, and Ferree(1998) find a similarresultin state elections, with a greaterimpact
of economic factors under united government.Unified governmentis a dummy
variable defined as the governorshipand both chambersof the state legislature
controlledby one political party equal to one, and zero otherwise.
National Factors
PRESIDENTIAL
EFFECTS.Despite the fact thatthe public seems easily able to recognize or recall their governor'sname (Squireand Fastnow 1994), we should not
infer that knowledge about the governor or state politics and policy runs very
deep. Studies indicate that the public possesses only a limited amount of attention and interest in political affairs (e.g., Delli Carpiniand Keeter 1996). Consequently,people may use heuristics, cues, or other short-cutsto organize their
thinkingabout state-levelpolitics. The presidentand the parties often are used as
referentsfor people in organizingtheir political world. Voterswho rely on such
short-cutdevices may associate all politicians of a particularparty as part of one
team, rewardingor blaming the entire set, rather than distinguishing between
national and state politicians (Carsey and Wright 1998). People also may be
inclinedto see partiesin termsof the presidentand generalizetheirattitudesabout
the presidentto all politicians of his party.Throughsuch a mechanism (Simon,
Ostrom and Marra 1991), the president'sapprovalmay affect the approvallevel
of governors of the president'sparty, much as it has been found to influence
gubernatorialvoting decisions and elections (e.g., Atkeson and Partin1995; King
2001; Simon 1989).
We measure this presidential effect with the Gallup presidential approval
measure,lagged one monthand correctedfor partyof the governor.Thatis, when
the president and governor share party affiliation, we use the Gallup approval
reading.When they come from differentparties, we subtractapprovalfrom 100
(100 - approval).This suggests a process in which opposition party governors
are helped when the president is unpopular,but harmed when he is popular.
Lastly,we enter a dummyvariable,indicatingwhetherthe governorcomes from
the president'sparty to pick up any aspect of presidential-gubernatorial
partisanship that the approvalvariableleaves unmeasured.
NATIONAL
ECONOMY.
Our theoretical model implies that people distinguish
between the state and national economy and that they understandthe limited
ability of state policy makers to affect the state economy (unemployment).
Despite the fact that people seem to distinguish between the two economies in
reasonableways (Niemi, Bremer,and Heel 1999), they may still targetthe governor based on the condition of the national economy. They may do so in part
because the governor is a handy target, being a relatively well-known public
figure, and people may feel a need to vent their discontent. Moreover,despite
1276
Jeffrey E. Cohen and James D. King
being awareof differencesbetween the state and nationaleconomies, people may
hold all public officials accountable for the health of the economy. With fixed
terms of office, often withholding approvalis the only way that the public can
signal displeasurewith the way things are going duringthe interimbetween elections. To nationalunemployment,which we discussed above, we add the monthly
nationalinflationrate, lagged one month, definedas the percentagechange in the
consumerprice index.16
Results
Table 1 presentsresultsof threeestimations:OLS, robustregression,and fixedeffects regression (fixed on state). We use robust regression because the gubernatorialapprovalratings are not necessarily randomlydistributedover time and
state. The state fixed effects control for other factors associated with the states
that are unmeasuredand may affect levels of gubernatorialapprovalacross the
states.17Both of these estimationsessentiallyduplicatethe OLS results,with only
minor changes in statisticalsignificance and regressioncoefficient magnitudes.18
Unless otherwise noted, our discussion focuses on the OLS results.
Results indicate that gubernatorialapprovalis highly responsive to relative
state unemploymentand the effect is strongly significant,with p values of .000
or better in all estimations.The OLS and robustregression estimationsfind that
each one-percentage-pointspread between state and national unemployment
leads to a 3.14% point shift in approval.The fixed-effects estimation slightly
dampens the effect of relative unemployment to 2.54. Based upon the OLS
results, when state and nationalunemploymentdivergeby one standarddeviation
16Nationalincome cannot be added because of its high correlationwith national unemployment,
which will producemulticollinearity.Since our theoreticalfocus is on the importanceof unemployment for gubernatorialapproval,we retain the unemploymentvariables.We also experimentedwith
economic perceptionsdataby enteringmonthly(lagged) consumereconomic confidenceand monthly
(lagged) consumereconomic expectationsfrom the University of Michigan'sSurvey of Consumers.
Results (not shown) indicate that both variablesare statisticallysignificantand signed properlyand
do not detractfrom othervariablesin the estimation.Given the theoreticalimportancethat economic
perceptionshas taken in recent studies of presidentialapproval,it would be of interestto incorporate
economic perceptions data into studies of gubernatorialapproval.Howell and Vanderleeuw(1990)
and Orth (2001) use perceptualeconomic data in their studies from their surveys of Louisiana and
Michigan voters, respectively.
7 The
xtreg procedurein Stata 8.0 was used to estimate the fixed-effects model.
"8Abovewe also noted complications in using the JAR data. Thus, we also ran numerous other
estimations to ensure that our results are robust and that these identified data set complications do
not affect our findings. Hence, we add controls for polling firm, numberof response categories, the
response scale used, and governorinto the OLS, robustregression,and state fixed effects estimations.
Although overallmodel fit usually improveswith these additionalcontrolvariables,in each case, relative unemploymentand national unemploymentretain their significant effects, although in some
instances the impactof some controlvariableschanges. Details of these analyses can be found in the
appendixon the JOP Web page.
Relative Unemploymentand GubernatorialPopularity
1277
Table 1. Impact of Relative Unemployment on Gubernatorial Popularity
OLS
b
Se
Variable
.24
Relative Unem.
-3.14
Inflation
-.07
.02
Nat. Unem.
-4.53
.36
.11
Pres. Pop.
.03
Gov-Pres. Same
-6.13
.78
St. PID
-.13
.04
St. Ideol.
-.11
.04
Months
-.38
.09
.001
.004
Months-sq.
4.83
1.35
Rep. Gov.
-2.26
1.79
Indep. Gov.
-59.50
9.02
Diversity
.00004 .00005
Population
Unified Govt.
-.89
.78
Gov. Elect. %
.33
.04
Constant
104.08
7.34
N
2016
.27
.26
R2/Adj. R2
F/Prob. F
49.11
.0000
Wald chi2/ Prob. chi2
Root MSE
13.88
sigma_u
sigma_e
Rho
* R2 within= .22, R2 between = .46
Fixed (State) Effects
Regression*
Robust Regression
p
.000
.000
.000
.001
.000
.001
.007
.000
.001
.000
.207
.000
.438
.259
.000
.000
b
-3.14
-.07
-4.53
.11
-6.13
-.13
-.11
-.38
.004
4.83
-2.26
-59.50
.00004
-.89
.33
104.08
2016
.27
61.03
SE
.24
.02
.36
.03
.78
.04
.04
.12
.002
1.46
1.68
8.97
.00006
.77
.05
7.65
p
.000
.000
.000
.001
.000
.002
.012
.001
.006
.001
.180
.000
.482
.250
.000
.000
b
-2.54
-.12
-4.68
.10
-5.75
-.09
-.10
-.48
.005
5.78
.33
-72.03
.0002
-.37
.24
124.93
2016
.26
SE
573.30
.0000
.31
.02
.34
.03
.77
.04
.04
.08
.001
1.36
1.98
28.12
.0002
.78
.04
14.09
p
.000
.000
.000
.001
.000
.025
.013
.000
.000
.000
.867
.010
.310
.638
.000
.000
.0000
13.88
6.34
12.52
0.20
(1.4), we can expect a shift of about 4.4 points in gubernatorialapproval.Governors in states with the best relative unemploymentperformance (-4.5) can
expect to see a whopping 14.2 points added to their approvallevels. In contrast,
those serving in states lagging behind the nation at maximum levels (6) can
expect an approvalloss approaching19.1%.
National unemploymentalso affects gubernatorialapprovalin expected ways.
Each 1% shift in nationalunemploymentleads to a 4.5 percentagepoint shift in
gubernatorialapproval.19We can compare the impact of relative and national
unemployment.Where a one standarddeviation rise in relative unemployment
depresses gubernatorialapprovalabout 4.4 points, a similar one standarddeviation (1.2) rise in nationalunemploymentwill dampenapprovalabout 5.4 points,
which is only marginallyhigher than the impact of relative unemployment.The
combined impact of both forms of unemploymentsuggests that state publics are
highly sensitive to the state'sjob climate.
Overall, this provides strong confirmationof our hypothesis that voters grade
the governorbased on how well their state'sjob climate fairs comparedto the
nation. Furthermore,relative unemploymentaffects gubernatorialjob approval
controlling for other state and national factors, under varying model specifications, points to which we now turn.
9The
strong correlationbetween nationaland state unemploymentprecludesus from arguingthat
national as opposed to state-level unemploymentis affecting voters.
1278
JeffreyE. Cohen and James D. King
Impact of Control Variables
Other state-level factors affect gubernatorialapproval.Gubernatorialtenure
has curvilineareffects, as hypothesized.The longer a governorstays in office, the
lower his/her approval,up to a point, when approvalbegins to rise with tenure.
These results suggest that each month in office leads to a drop of .39 in gubernatorialapproval,counterbalancedby a rise of .004 from the squaredterm. The
curvilinearmodel anticipatesthat after one year in office a governorwill see a
drop in approvalof 3.95 points just as a function of being in office. At the end
of one term, the net effect of tenure is -8.45 approvalpoints. For a governorto
be reelected,much else must be working to the governor'sadvantage.Only well
into the second term does the effect of service show a positive effect on approval.
At 96 months, two full years, the net effect of tenurestandsat 2.7. The net effect
of tenureonly begins to turnpositive at month 90, when the effect of tenurenets
the governor.2 approvalpoints.
As expected, more diverse states depress gubernatorialapproval,but population size adds little over and above the effects of diversity.Approvalof the governor in the most diverse state will be about 9.3 points less than that of the
governorof the least diverse state and about 4.7 points lower than a governorof
a state of average diversity.
Unified government,however, does not affect gubernatorialapproval.20But
governorswho come to office with strongerelection supportcan expect to see
higher approvalthan those who enter office with less voter support.The results
indicate that each one percentage point in electoral support produces a .33
approvalbase for the governor.Thus, governorselected with a landslide of 60%
will see approvalthat is about three points higher than those elected at a bare
majority of 50.1%. Independent governors do not seem to see a significant
approvalloss. The Republicangovernordummy suggests an additional4.8 point
boost in approval.21
Both state ideological and partisanadvantagework opposite to expectations,
however.Substantively,each 10 percentagepoint shift in partisanadvantageproduces a correspondingshift of 1.3 points in gubernatorialapproval.Governors
with the largest net partisanbases will see approvalabout 4.6 points lower than
a governor serving in a state with perfectly balanced partisanship.Governors
serving in states most deficient in their partisanbase will see approvalthat is 3.1
points higher than governorsin states with equally sized partisanelectoratesand
nine points higher than those in the most partisan advantaged states.22State
20
This variablebecomes significantwith the correct sign when controls for governordummies are
applied. See the appendixon the JOP Web page.
21
Both of these variables are unstable when other variables, such as gubernatorialdummies, are
enteredas controlsbecause of the multicollinearity.Details on these additionalcontrols are provided
on the JOP Web page for this article.
22With the added controls, such as governor(see the JOP Web page), the ideology variabletends
not to maintain statistical significance. But the partisanadvantagevariablekeeps its statistical significance and unexpected sign no matterwhat controls are used.
Relative Unemploymentand GubernatorialPopularity
1279
ideology has comparableimpacts.Each 10 percentagepoint shift in a state'sideological make up will shift gubernatorialapprovalabout 1.1 points. A governor
serving in a state with the greatest ideological advantagecan expect to see an
approvallevel some 3.2 points lower than a governorwith an ideologically balanced statepublic. When the ideological balancetips to the maximumagainstthe
governor,approvalis predictedto rise about 3.2 points above that for governors
in balanced states.
Why this seemingly counterintuitiveeffect for partisanand ideological advantage in statepublics?Perhapsminorityexecutivestry harderto appealto a broader
set of votersbecause they have too few copartisans(or coideologues) to rely upon
them exclusively. In contrast, majority governors do not have to seek public
support from outside of the ranks of their copartisans(or coideologues) in the
mass public. Hence their public style may be more intensely partisanand/orideologically extreme or rigid. As a consequence these governorsmay more easily
(or intentionally?)alienate citizens who are not members of their party or ideological temperament.
Turningto nationalfactors,governorsof the president'spartycan expect to see
a . 1 change in their approvalwith every one percentagepoint change in presidential approval.Although highly statisticallysignificant,the substantiveimplications of presidentialapprovaldo not appearmajor.A one standarddeviation
change in presidentialapproval(11.2) translatesinto only about a 1.2 percentage
point change in gubernatorialapproval.A change from the lowest (11) to the
highest level of presidential(82) approvalproduces a 7.8 point shift in gubernatorial approval,but such great swings in presidentialapprovalare rare. Results
also indicate that governorssee a drop of about 6.13 approvalpoints when they
come from the president'sparty,indicatingthat the presidency affects gubernatorial approvalbeyond the swings in presidentialapproval.If we drop the president-governorsame party dummy from the estimationthe presidentialapproval
variablefails to reach statisticalsignificance,also indicatinga more complex relationship than the simple presidentialapprovalspillover effect.
Lastly, inflation affects gubernatorialapproval. The regression coefficient
points to a -.07 drop in approvalfor each 1%point increase in inflation,which
suggests that gubernatorialapproval substantively is not highly sensitive to
national inflation tides. Moreover,inflation moves very slightly from month to
month, on average 1.4% in these data. Still, over a longer period of time, sustained inflationaryspiralsmay cumulateinto substantivelysignificantimpactson
gubernatorialapproval.
Conclusion
The literature on unemployment and gubernatorial approval is rife with
controversy,with some studies finding state unemploymenteffects (Barth and
Ferguson2002; Crew et al. 2002; Hansen 1999a, 1999b; Orth2001), but others
failing to do so (Adams and Squire2001; Crew and Weiher 1996). Our analysis
1280
JeffreyE. Cohen and James D. King
reformulatesthe issue into one in which voters comparenationaland state unemployment levels.
We argue that the federal structureof the political system, the integrationof
state economies into the nationaleconomy,and the limited economic policy tools
availableto the governorpresentbarriersfor citizens who want to hold their governoraccountablefor the state economy.Votersresolve this complex situationby
holding the governor accountablefor how well or poorly the state's unemployment level fares comparedto the nationalunemploymentlevel.
The actions of governors help the public in rendering such a comparative
assessment. Governors often possess policy tools that may marginally affect
unemployment,making a focus on jobs substantivelyreasonable.They publicize
their efforts to bring jobs to the state, while electoral opponents often chastise
the incumbent administration'sperformancewith regardto jobs. The focus on
relativeunemploymentprovidesa way for citizens to gradetheir governor,while
also recognizing the fact that the national economy and economic policy decisions by national policy makers will affect the performance of the state's
economy. Under different specifications, using different control variables, and
differentestimationassumptions,we findthatrelativeunemploymentlevels affect
gubernatorialapprovalas our model predicts. Our findings are robust to these
alterationsin estimation, and the magnitude of effect is relatively stable across
the variousspecifications.And our results supportour theory,which assumes that
state electorates are sophisticated enough to distinguish between the state and
national economies when evaluatingthe governor'sperformancein office, and
accordinglyhold the governoraccountablefor the state economy.
This study also demonstratesthe benefit of using a more comprehensivedata
collection on gubernatorialapprovalthanused in past research.The JARdata set,
which provides 2,000 cases for analysis, allows us to include a larger array of
control variablesthan past researchhas been able to do. Our theoreticalmodel
and analysis suggests that the public rates governorsbased upon the outcomes
and policies over which they have some control or impact. One direction for
future research is to extend to other areas over which governors possess some
ability to affect policy and performance,and which are often high priorities in
gubernatorialagendas, such as education.
Acknowledgment
We would like to thank Joseph Bafumi, Charles Cameron, James Druckman,
Robert Erikson, Kathleen Knight, Robert Liberman, Robert Shapiro, Gregory
Wawro,participantsof the American Politics seminar at Columbia University,
and severalanonymousreviewersfor their commentson previousversions of this
research.
ManuscriptsubmittedMay 15, 2003
Final manuscriptreceived February 10, 2004
Relative Unemployment and Gubernatorial Popularity
1281
References
Adams, Greg, and Peverill Squire. 2001. "A Note on the Dynamics and Idiosyncrasiesof Gubernatorial Popularity."State Politics & Policy Quarterly1 (Winter):380-93.
Adams, JamesD., and LawrenceW Kenny. 1989. "TheRetentionof StateGovernors."Public Choice
62 (July): 1-13.
Atkeson, Lonna Rae, and RandallW Partin. 1995. "Economic and ReferendumVoting:A Comparison of Gubernatorialand SenatorialElections."AmericanPolitical Science Review 89 (March):
99-107.
Barth, Jay, and MargaretR. Ferguson. 2002. "AmericanGovernors and Their Constituents:The
Relationshipbetween GubernatorialPersonalityand Public Approval."State Politics and Policy
Quarterly2 (Fall): 268-82.
Beyle, Thad,RichardG. Niemi, and Lee Sigelman.2002. "Gubernatorial,Senatorial,and State-Level
PresidentialJob Approval Ratings: The U.S. Officials Job Approval Ratings (JAR) Collection."
State Politics and Policy Quarterly2 (Fall): 215-29.
Bond, Jon R., and Richard Fleisher. 2001. "The Polls: Partisanshipand PresidentialPerformance
Evaluations."Presidential Studies Quarterly31(3): 529-41.
Brace, Paul. 1993. State Governmentand Economic Performance.Baltimore:Johns Hopkins Press.
Brace, Paul, and BarbaraHinckley. 1991. "The Structureof PresidentialApproval:ConstraintsWithin
and Across Presidencies."Journal of Politics 53(4): 993-1017.
Carsey,Thomas M., and Gerald C. Wright. 1998. "State and National Forces in Gubernatorialand
SenatorialElections."AmericanJournal of Political Science 42 (July): 994-1002.
Cohen, Jeffrey E. 1983. "GubernatorialPopularityin Nine States."American Politics Quarterly 11
(April): 219-35.
Conover, Pamela Johnston, Stanley Feldman, and Kathleen Knight. 1986. "Judging Inflation and
Unemployment:The Origins of RetrospectiveEvaluations."Journal of Politics 48(3): 565-88.
Crew,Robert E., Jr. 1998. "GubernatorialLeadership:Testinga PreliminaryModel."Social Science
Journal 36(1): 15-27.
Crew, Robert E., Jr., David Branham, Gregory R. Weiher, and Ethan Berick. 2002. "Political
Events in a Model of GubernatorialApproval."State Politics and Policy Quarterly2 (Fall): 28397.
Crew, Robert E., and Gregory R. Weiher. 1996. "GubernatorialPopularityin Three States: A Preliminary Model."Social Science Journal 33(1): 39-55.
Delli Carpini, Michael, and Scott Keeter. 1996. WhatAmericans Know About Politics and WhyIt
Matters. New Haven:Yale University Press.
Dometrius,Nelson C. 2002. "Gubernatorial
ApprovalandAdministrativeInfluence."State Politics &
Policy Quarterly2 (Fall): 251-67.
Erikson, Robert S., Michael MacKuen, and James A. Stimson. 2002. The Macropolity.New York:
CambridgeUniversity Press.
Erikson, Robert S., Gerald C. Wright, and John P. Mclver. 1993. Statehouse Democracy: Public
Opinion and Policy in the American States. New York:CambridgeUniversity Press.
Ferguson,MargaretRobertson.2003. "Chief Executive Success in the LegislativeArena."State Politics & Policy Quarterly3 (Summer): 158-82.
Grady, Dennis. 1991. "Managingthe State Economy: The Governor'sRole in Policymaking."In
GubernatorialLeadershipand State Policy, eds. Eric B. Herzik and Brent W Brown. New York:
GreenwoodPress, pp. 105-20.
Gronke,Paul. 2001. TheElectorate,the Campaign,and the Office:A UnifiedApproachto Senate and
House Elections. Ann Arbor:University of Michigan Press.
Gronke,Paul, and Brian Newman. 2003. "FromFDR to Clinton, from Mueller to ?? A Field Essay
on PresidentialApproval."Political Research Quarterly56 (December): 501-12.
Haller, H. Brandon, and Helmut Norpoth. 1997. "Reality Bites: News Exposure and Economic
Opinion."Public Opinion Quarterly61: 555-75.
1282
Jeffrey E. Cohen and James D. King
Hansen,SusanB. 1999a. "Governors'Job PerformanceRatingsand Unemployment:The Case of California."State and Local GovernmentReview 31 (Winter):7-17.
Hansen, Susan B. 1999b. "'Life is not Fair': Governors' Job Performance Ratings and State
Economies."Political Research Quarterly52 (March): 167-88.
Hendrick,Rebecca M., and James C. Garand.1991. "Variationin State Economic Growth:Decomposing State, Regional, and National Effects."Journal of Politics 53(4): 1093-1110.
Herzik, Eric B. 1991. "PolicyAgendas and GubernatorialLeadership."In GubernatorialLeadership
and State Policy, eds. Eric B. Herzikand BrentW.Brown.New York:GreenwoodPress,pp. 25-39.
Holbrook, Thomas, and James C. Garand. 1996. "Homo Economous? Economic Informationand
Economic Voting."Political Research Quarterly49 (June): 351-75.
Howell, Susan E., and James M. Vanderleeuw. 1990. "Economic Effects on State Governors."
AmericanPolitics Quarterly18 (April): 158-68.
Kenney, Patrick J., and Tom W. Rice. 1983. "Popularityand the Vote: The GubernatorialCase."
AmericanPolitics Quarterly11 (April): 237-41.
Kernell, Samuel. 1978. "ExplainingPresidentialPopularity."American Political Science Review 72
(June): 506-22.
King, James D. 1999. "Wyoming'sGubernatorialElections:RetrospectiveVotingin Good Times and
Bad." ComparativeState Politics 20(4): 17-26.
King, James D. 2001. "IncumbentPopularityand Vote Choice in GubernatorialElections."Journal
of Politics 63(2): 585-97.
Lowry,Robert C., James E. Alt, and Karen E. Ferree. 1998. "Fiscal Policy Outcomes and Electoral
Accountabilityin American States."AmericanPolitical Science Review 92 (December):759-74.
MacDonald,JasonA., and Lee Sigelman. 1999. "PublicAssessments of GubernatorialPerformance:
A ComparativeStateAnalysis."AmericanPolitics Quarterly27 (April): 201-15.
Morgan,David R., and L. A. Wilson. 1990. "Diversityin the AmericanStates:Updatingthe Sullivan
Index."Publius 20(1): 71-81.
Mueller,John E. 1970. "PresidentialPopularityfromTrumanto Johnson."AmericanPolitical Science
Review 64 (March): 18-34.
Nicholson, StephenP., Gary M. Segura,and NathanD. Woods. 2002. "PresidentialApprovaland the
Mixed Blessing of Divided Government."Journal of Politics 64(3): 701-20.
Niemi, Richard G., John Bremer, and Michael Heel. 1999. "Determinantsof State Economic
Perceptions."Political Behavior 21 (June): 175-93.
Orth, Deborah A. 2001. "Accountabilityin a Federal System: The Governor,the President, and
Economic Expectations."State Politics and Policy Quarterly1 (Winter):412-32.
Simon, Dennis M. 1989. "Presidents,Governors,and ElectoralAccountability."Journal of Politics
51(2): 286-304.
Simon, Dennis M., CharlesW Ostrom,Jr.,and Robin F Marra. 1991. "The President,Referendum
Voting, and SubnationalElections in the United States."American Political Science Review 85
(December): 1177-92.
Squire, Peverill, and ChristinaFastnow.1994. "ComparingGubernatorialand SenatorialElections."
Political Research Quarterly47 (September):705-20.
Stimson, James A. 1976. "Public Support for American Presidents:A Cyclical Model." Public
Opinion Quarterly40(1): 1-22.
Svoboda, CraigT. 1995. "RetrospectiveVoting in GubernatorialElections: 1982 and 1986."Political
Research Quarterly48 (March): 135-50.
Jeffrey E. Cohen is professor of political science, Fordham University, Bronx,
NY 10458-9993, ([email protected]). James D. King is professor of political
science, University of Wyoming, Laramie, WY 82071-3197, ([email protected]).