Economic Retrospection and the Calculus of Voting

Supplemental Information

Economic Retrospection and the Calculus of Voting

Political Behavior

Brad T. Gomez

Associate Professor of Political Science
Department of Political Science
531 Bellamy Building
Florida State University
Tallahassee, FL 32306-2230

Email:

Thomas G. Hansford
Associate Professor of Political Science
School of Social Sciences, Humanities and Arts
University of California, Merced
5200 North Lake Road
Merced, CA 95343
Email:

Economic Retrospection and the Calculus of Voting

*** Supplemental Information ***

Details of the First-Stage Models

When testing our hypothesis regarding the effect of retrospective economic evaluations on the decision to vote we rely on individual-level subjective assessments of the state of the economy. There are two main advantages to this approach. First, the most appropriate and proximate economic variables are the voters’ subjective evaluations of national economic performance. To the extent that the state of the national economy affects the decision to vote, it will do so through voter perceptions of the economy. Second, subjective evaluations of the economy vary from voter to voter, unlike the objective state of the economy, which is a constant for any given election. This within-election variation allows for the potential of explaining variation in individual-level turnout for any given election.

The principal disadvantage to using subjective evaluations of the state of the economy is that these subjective assessments are likely endogenous to various political activities, including turnout. To negate this disadvantage we employ an IV approach that relies on objective local economic conditions as instruments for subjective assessments of the national economy. Table S1 presents the two first-stage models for Model 1.1, one for each endogenous variable. Table S2 presents the four first-stage models for Model 1.2 (in which there are four endogenous variables).

To instrument for Sociotropic Evaluations2, we follow Wooldridge (2002, 237) and use the square of the predicted values of Sociotropic Evaluations obtained with a model including all the independent variables and our other excluded instruments. These predicted values are then treated as an instrument in the following first-stage models. The model used to generate these predicted values for Sociotropic Evaluations is presented in Table S3.

Table S1 First-Stage Models for Model 1.1.

Independent Variables / (Sociotropic Evaluations)2 / (Sociotropic Evaluations)2
× Incumbent
/ .119
(.073) / -.058
(.056)
× Incumbent / .477*
(.094) / .650*
(.073)
Female / .003
(.028) / -.005
(.022)
Black / .172*
(.043) / .111*
(.033)
Latino / .176*
(.052) / .025
(.040)
Asian / .009
(.109) / -.031
(.084)
Age / .011*
(.005) / .004
(.004)
Age2 / -.000*
(.000) / -.000
(.000)
Education / .027*
(.009) / .007
(.007)
Income / -.007
(.016) / -.012
(.012)
Unemployed / .071
(.047) / .084*
(.037)
Married / .012
(.030) / .009
(.023)
Union Member / .043
(.035) / .004
(.027)
Religiosity / -.032*
(.011) / -.012
(.009)
Strength of Party ID / .065*
(.014) / .037*
(.011)
Party Contact / .021
(.030) / .002
(.023)
Registration Closing Date / .000
(.001) / .002
(.001)
Constant / .821*
(.152) / 1.04*
(.118)
N / 12,842 / 12,842
F-test for excluded instruments / 34.3* / 59.8*
R2 / .199 / .351

* p ≤ .05 (two-tailed). Entries are OLS estimates (and standard errors). Election fixed effects are included in the model.

Table S2 First-Stage Models for Model 1.2.

Independent Variables / SocEval / SocEval × Incumbent / (SocEval)2 / (SocEval)2 × Incumbent
Δ County Income (in $1,000s) / .059*
(.011) / .001
(.014) / -.009
(.018) / -.001
(.014)
Δ County Unemployment / -.035*
(.009) / -.012
(.007) / .028
(.014) / .003
(.011)
Δ County Income × Incumbent / -.038*
(.018) / .017
(.014) / .048
(.029) / .042
(.022)
Δ County Unemployment × Inc. / -.012
(.011) / -.033*
(.009) / -.011
(.018) / .015
(.014)
/ .313*
(.047) / .283*
(.037) / .083
(.075) / -.065
(.058)
× Incumbent / -.070
(.061) / -.115*
(.048) / .483*
(.097) / .630*
(.076)
Female / -.222*
(.018) / -.156*
(.014) / .008
(.029) / -.002
(.022)
Black / -.220*
(.027) / -.207*
(.021) / .174*
(.043) / .113*
(.033)
Latino / -.084*
(.033) / -.037
(.026) / .187*
(.052) / .024
(.041)
Asian / .138
(.069) / .080
(.054) / .008
(.109) / -.033
(.085)
Age / -.015*
(.003) / -.008*
(.002) / .012*
(.005) / .005
(.004)
Age2 / .000*
(.000) / .000*
(.000) / -.000*
(.000) / .000
(.000)
Education / .014*
(.006) / .015*
(.005) / .027*
(.009) / .007
(.007)
Income / .049*
(.010) / .044*
(.008) / -.008
(.016) / -.012
(.012)
Unemployed / -.081*
(.030) / -.077*
(.023) / .072
(.047) / .083*
(.037)
Married / -.005
(.019) / -.001
(.015) / .011
(.030) / .008
(.023)
Union Member / -.116*
(.022) / -.100*
(.018) / .046
(.035) / .006
(.027)
Religiosity / .012
(.007) / .010
(.005) / -.032*
(.011) / -.012
(.009)
Strength of Party ID / .051*
(.009) / .038*
(.007) / .064*
(.014) / .036*
(.011)
Party Contact / -.010
(.019) / .003
(.015) / .018
(.030) / -.000
(.023)
Registration Closing Date / .003*
(.001) / .002*
(.001) / .000
(.002) / .002
(.001)
Constant / -1.42*
(.100) / -1.44*
(.078) / .839*
(.158) / 1.07*
(.123)
N / 12,842 / 12,842 / 12,842 / 12,842
F-test for excluded instruments / 24.4* / 25.9* / 12.9* / 21.3*
R2 / .374 / .370 / .199 / .352

* p ≤ .05 (two-tailed). Entries are OLS estimates (and standard errors). Election fixed effects are included in the model.

Table S3 Predicting Sociotropic Evaluations.

Independent Variables / Coefficient Estimates
(Standard Errors) / Coefficient Estimates
(Standard Errors)
Δ County Income (in $1,000s) / .037*
(.009) / .044*
(.010)
Δ County Unemployment / -.032*
(.005) / -.038*
(.006)
Female / -.175*
(.017) / -.206*
(.020)
Black / -.160*
(.026) / -.192*
(.031)
Latino / -.073*
(.033) / -.091*
(.039)
Asian / .118
(.069) / .135
(.080)
Age / -.011*
(.003) / -.014*
(.003)
Age2 / .000*
(.000) / .000*
(.000)
Education / .012*
(.006) / .013
(.007)
Income / .040*
(.010) / .043*
(.011)
Unemployed / -.057
(.030) / -.069*
(.035)
Married / -.005
(.019) / -.008
(.022)
Union Member / -.091*
(.022) / -.107*
(.026)
Religiosity / .008
(.007) / .011
(.008)
Strength of Party ID / .040*
(.008) / .045*
(.010)
Party Contact / -.009
(.019) / -.013
(.022)
Registration Closing Date / .003*
(.001) / .003*
(.001)
Constant / -1.12*
(.072) / ---
Estimator / OLS / Ordered Probit
N / 12,842 / 12,842
F-test, Δ County Income = Δ County Unemploy. = 0 / 41.5* / ---
R2 / .374 / ---

* p ≤ .05 (two-tailed). Election fixed effects are included in the model. The cut points for the ordered probit model are -.218, .583, 1.68, and 2.77.

Table S4 Full Results of IV Models 1.1 and 1.2.

Independent Variables / Model 1.1 / Model 1.2
Sociotropic Evaluation / --- / .067
(.095)
Sociotropic Evaluation × Incumbent / --- / -.002
(.127)
(Sociotropic Evaluation)2 / .374*
(.147) / .220*
(.131)
(Sociotropic Evaluation)2 × Incumbent / -.379†
(.143) / -.236
(.136)
Female / -.004
(.010) / .011
(.011)
Black / -.014
(.021) / .010
(.021)
Latino / -.090†
(.029) / -.063†
(.025)
Asian / -.146†
(.039) / -.150†
(.033)
Age / .008†
(.002) / .010†
(.002)
Age2 / -.000
(.000) / -.000†
(.000)
Education / .052†
(.004) / .054†
(.003)
Income / .040†
(.006) / .037†
(.005)
Unemployed / -.051†
(.017) / -.047†
(.015)
Married / .014
(.011) / .015
(.009)
Union Member / .016
(.014) / .029†
(.013)
Religiosity / .043†
(.005) / .039†
(.004)
Strength of Party ID / .066†
(.006) / .068†
(.005)
Party Contact / .100†
(.011) / .105†
(.009)
Registration Closing Date / -.002†
(.001) / -.002†
(.000)
Constant / .089
(.089) / .145
(.087)
N / 12,842 / 12,842

Note: * p ≤ .05 (one-tailed, for hypothesized relationships). † p ≤ .05 (two-tailed, for control variables). Election fixed effects are also included, which absorb the direct effect of an Incumbent on the ballot.

Table S5 Full Results of OLS Models 1.3 and 1.4

Independent Variables / Model 1.3 / Model 1.4
Sociotropic Evaluation / --- / .005
(.006)
Sociotropic Evaluation × Incumbent / --- / .001
(.008)
(Sociotropic Evaluation)2 / -.005
(.004) / -.004
(.004)
(Sociotropic Evaluation)2 × Incumbent / .009
(.005) / .009
(.005)
Female / .004
(.007) / .005
(.007)
Black / .016
(.011) / .016
(.011)
Latino / -.034†
(.013) / -.034†
(.013)
Asian / -.135†
(.028) / -.135†
(.028)
Age / .011†
(.001) / .011†
(.001)
Age2 / -.000†
(.000) / -.000†
(.000)
Education / .059†
(.002) / .059†
(.002)
Income / .041†
(.004) / .041†
(.004)
Unemployed / -.054†
(.012) / -.053†
(.012)
Married / .015
(.008) / .015
(.008)
Union Member / .033†
(.009) / .033†
(.009)
Religiosity / .035†
(.003) / .035†
(.003)
Strength of Party ID / .075†
(.003) / .075†
(.003)
Party Contact / .109†
(.008) / .109†
(.008)
Registration Closing Date / -.002†
(.000) / -.002†
(.000)
Constant / -.016
(.030) / -.011
(.030)
N / 12,842 / 12,842

Table S6 Full Results of Models 1.3 and 1.4, Estimated via Probit

Independent Variables / Model 1.3 / Model 1.4
Sociotropic Evaluation / --- / .013
(.025)
Sociotropic Evaluation × Incumbent / --- / .017
(.033)
(Sociotropic Evaluation)2 / -.026
(.014) / -.023
(.016)
(Sociotropic Evaluation)2 × Incumbent / .035
(.019) / .041†
(.021)
Female / .008
(.028) / .012
(.028)
Black / .060
(.041) / .063
(.041)
Latino / -.095
(.050) / -.095
(.050)
Asian / -.531†
(.106) / -.533†
(.106)
Age / .035†
(.004) / .035†
(.004)
Age2 / -.000†
(.000) / -.000†
(.000)
Education / .238†
(.010) / .238†
(.010)
Income / .158†
(.016) / .157†
(.016)
Unemployed / -.138†
(.045) / -.137†
(.045)
Married / .084†
(.031) / .084†
(.031)
Union Member / .103†
(.038) / .105†
(.038)
Religiosity / .138†
(.011) / .138†
(.011)
Strength of Party ID / .282†
(.014) / .281†
(.014)
Party Contact / .526†
(.035) / .525†
(.035)
Registration Closing Date / -.010†
(.002) / -.010†
(.002)
Constant / -1.98†
(.117) / -1.96
(.118)
N / 12,842 / 12,842

Table S7 IV Models 1.1 and 1.2, by Incumbent.

Independent Variables / Model 1.1 / Model 1.1 / Model 1.2 / Model 1.2
Sociotropic Evaluation / --- / --- / .167
(.106) / .027
(.042)
(Sociotropic Evaluation)2 / .416*
(.227) / -.015
(.035) / .298*
(.147) / -.021
(.035)
Female / -.018
(.024) / .005
(.009) / .012
(.021) / .010
(.013)
Black / -.035
(.056) / -.007
(.017) / -.008
(.038) / .002
(.022)
Latino / -.151
(.079) / -.043
(.020) / -.108†
(.052) / -.041†
(.021)
Asian / -.174†
(.080) / -.125†
(.039) / -.191†
(.065) / -.128†
(.040)
Age / .002
(.005) / .012†
(.002) / .007†
(.004) / .012†
(.002)
Age2 / .000
(.000) / -.000†
(.000) / -.000
(.000) / -.000†
(.000)
Education / .037†
(.014) / .060†
(.003) / .043†
(.009) / .059†
(.003)
Income / .038†
(.012) / .040†
(.005) / .038†
(.010) / .038†
(.006)
Unemployed / -.048
(.035) / -.054†
(.018) / -.049
(.027) / -.049†
(.019)
Married / .015
(.024) / .013
(.010) / .018
(.018) / .013
(.010)
Union Member / -.048
(.041) / .050†
(.011) / -.029
(.028) / .053†
(.013)
Religiosity / .056†
(.015) / .035†
(.004) / .049†
(.011) / .035†
(.004)
Strength of Party ID / .058†
(.017) / .071†
(.005) / .060†
(.013) / .070†
(.005)
Party Contact / .106†
(.025) / .095†
(.010) / .119†
(.019) / .095†
(.010)
Registration Closing Date / -.001
(.001) / -.002†
(.000) / -.002†
(.001) / -.002†
(.001)
Constant / -.928
(.533) / -.002
(.069) / -.443
(.308) / .040
(.095)
Incumbent Candidate? / No / Yes / No / Yes
N / 5,077 / 7,765 / 5,077 / 7,765

Note: * p ≤ .05 (one-tailed, for hypothesized relationships). † p ≤ .05 (two-tailed, for control variables). Election fixed effects are also included.

Using Objective National Conditions Instead of Subjective Perceptions

As a robustness check, we also estimate our model using objective national conditions instead of individual-level perceptions. Specifically, we use Δ U.S. Median Income (in 10,000s of inflation-adjusted dollars) and Δ U.S. Unemployment as our national economic indicators. To test for the presence of the curvilinear relationship that we theorize, both of these variables are squared, and we expect these squared variables to have positive coefficients. To allow these effects to vary according to whether one of the candidates is the incumbent president, we interact these variables with Incumbent. All of the control variables included in Models 1.1, 1.2, and 2.1 are also included in this model. The important exception that we cannot include the election fixed effects since the economic variables are election-specific. This is a clear disadvantage, yet by using national economic conditions instead of Sociotropic Evaluation we are able to include six more presidential elections (we add 1956 through 1976), yielding a total of 14 elections. The inclusion of these earlier elections leads us to add three additional control variables related to voter registration requirements: Property Requirement, Literacy Test, and Poll Tax. These variables indicate whether individuals resided in a state with one or more of these restrictive voting laws.

The results for this model estimation are presented in Table S7. The pattern of results for Δ U.S. Unemployment is fully consistent with what we find with Sociotropic Evaluation. Large increases or decreases in the unemployment rate increase the probability of voting, as long as there is not an incumbent candidate. The results for Δ U.S. Median Income are not as similar, though the estimate for (Δ U.S. Median Income)2 is in the predicted direction.

Table S8 Effect of National Economic Circumstances on Individual-Level Turnout, 1956-2008.

Independent Variables / Coefficient Estimate / (Standard Error)
Δ U.S. Median Income (in $10,000s) / 1.25† / (.551)
(Δ U.S. Median Income)2 / .0002 / (.0002)
Δ U.S. Unemployment / .209† / (.026)
(Δ U.S. Unemployment)2 / .017* / (.007)
Δ U.S. Median Income × Incumbent / -1.42† / (.625)
(Δ U.S. Median Income)2 × Incumbent / .0003 / (.0002)
Δ U.S. Unemployment × Incumbent / -.197† / (.029)
(Δ U.S. Unemployment)2 × Incumbent / -.016* / (.007)
Incumbent / .395 / (.260)
Female / -.086† / (.022)
Black / -.049 / (.040)
Latino / -.136† / (.051)
Asian / -.474† / (.105)
Age / .046† / (.004)
Age2 / -.0003† / (.0000)
Education / .218† / (.008)
Income / .154† / (.012)
Unemployed / -.128† / (.041)
Married / .076† / (.025)
Union Member / .118† / (.028)
Religiosity / .157† / (.010)
Strength of Party ID / .250† / (.011)
Party Contact / .470† / (.027)
Registration Closing Date / -.006† / (.001)
Property Requirement / -1.45† / (.151)
Literacy Test / -.043 / (.062)
Poll Tax / -.894† / (.086)
Constant / -2.63 / (.263)
N
Log likelihood
Wald (df: 27) / 21,182
- 9469.0
3,466.4*

Note: * p ≤ .05 (one-tailed, for hypothesized relationships). † p ≤ .05 (two-tailed, for control variables). Entries are probit estimates.

Table S9 Full Results of Models 2.1 and 2.2

Independent Variables / Model 1.3 / Model 1.4
Δ County Income / -.021†
(.008) / -.070†
(.031)
(Δ County Income)2 / (3.5×10-6)*
(1.2×10-6) / (1.3×10-5)*
(5.3×10-6)
Δ County Income × Incumbent / .020†
(.009) / .066
(.037)
(Δ County Income)2 × Incumbent / (-7.6×10-6)†
(2.8×10-6) / (-3.3×10-5)†
(1.1×10-5)
Female / -.010
(.006) / -.047
(.025)
Black / .009
(.010) / .043
(.038)
Latino / -.037†
(.013) / -.109†
(.049)
Asian / -.126†
(.027) / -.494†
(.102)
Age / .012†
(.001) / .038†
(.004)
Age2 / -.000†
(.000) / -.000†
(.000)
Education / .059†
(.002) / .234†
(.009)
Income / .040†
(.004) / .155†
(.014)
Unemployed / -.049†
(.011) / -.123†
(.041)
Married / .022†
(.007) / .105†
(.028)
Union Member / .040†
(.008) / .125†
(.033)
Religiosity / .038†
(.003) / .147†
(.010)
Strength of Party ID / .075†
(.003) / .277†
(.012)
Party Contact / .107†
(.007) / .502†
(.031)
Registration Closing Date / -.002†
(.000) / -.007†
(.001)
Constant / -.006
(.028) / -1.94
(.107)
Estimator / OLS / Probit
N / 15,929 / 15,929