Partisan Accountability and Economic Voting

Transcription

1 Evidence from Exchange Rate Fluctuations L. Jason Anastasopoulos 1 Aaron Chalfin 2 1 Department of Political Science UC Berkeley 2 Goldman School of Public Policy UC Berkeley November 16, 2011

2 Congressional Economic Voting: Literature Review Preview Major questions Macro-Level Do changes in the economy impact congressional voting? How do they impact congressional voting? Micro-Level By what process Rational Accountability Partisan Accountability

3 Congressional Economic Voting: Literature Review Preview Economic Voting: Macro-Level Kramer (1971) - It s the economy, stupid. Study of Midterm elections. DV: Aggregate House Member Vote Share: Party of the President IVs: Per capita income, unemployment. Improvements help incumbents in the party of the President, declines hurt them. Subsequent studies support/contradict Kramer (1971) It s the economy, stupid.(bloom and Price 1975; Kiewiet and Udell 1998) It s NOT the economy, stupid. (Erikson 1990)

4 Congressional Economic Voting: Literature Review Preview Economic Voting: Micro-level Focus on accountability. Rational Accountability: Incumbents held responsible for economic performance over the course of their term. Retrospective Voting.(Fiorina 1978, 1981). Sophisticated Retrospective Voting. (Ebeid and Rodden 2006) Partisan Accountability: Incumbents held responsible for economic performance in a partisan fashion. Group attribution bias. (Peffley, Feldman and Sigelman 1987; Sigelman and Knight 1987 ; Rudolph 2003) Economic ideology. (Rudolph 2003)

5 Congressional Economic Voting: Literature Review Preview : Macro-Level It s STILL the economy, stupid. Data: CCES + Census State-Level Panel. 1% increase in (instrumented) state unemployment decreases House incumbent vote share by 5%. Effect driven almost entirely by Democratic House incumbents. 1% increase in (instrumented) state unemployment decreases Democratic House incumbent vote share by 6.3%. No effect found for Republican House incumbents.

6 Congressional Economic Voting: Literature Review Preview : Micro-Level Support for partisan accountability. Data: ANES State-Level Panel. Question: Did you vote for a candidate for Congress? Whom did you vote for? Which party was that? Coded response: R s Vote: for major party candidate from party different than R s partisanship 1 Pooled: 1% in (instrumented) state unemployment 5.9% in R vote for candidate of different partisanship. Democratic Partisans: 1% in (instrumented) state unemployment 6.8% in R vote for candidate of different partisanship. Republican Partisans: No effect found. 1 ANES Note: The respondent has been considered a partisan if R was strong, weak or leaning

7 OLS + Fixed Effects Instrumental Variables : OLS w Fixed Effects Omitted variables/endogeneity May not return consistent estimates. State-level economic policy decisions that affect unemployment and vote share vary w/in states over time. Fixed effects/differencing won t solve this problem. Measurement Error in the Independent Variable Measurement error in unemployment will produce inconsistent OLS estimates. Made worse with the inclusion of fixed effects/differencing.

8 OLS + Fixed Effects Instrumental Variables : Measurement Error with OLS Measurement error in the independent variable biases coefficient estimates toward zero. W/ no omitted variables, β is unbiased and consistent when: y = βx + ɛ, ɛ N(0, 1) (1) cov(x, ɛ) = 0 (2) y = Vote share, X = unemployment.

9 OLS + Fixed Effects Instrumental Variables : Measurement Error with OLS Classical Measurement Error - Measurement uncorrelated with the variable of interest. Found to be present in Current Population Survey estimates of local and state unemployment. (Ullman 1962; Buss 1986; Poterba and Summers 1986; Norwood 1988; Bound and Krueger 1991; Bollinger 1998; Feng 2004). With classical measurement error, however. We estimate: y = βx + ɛ, ɛ N(0, 1) (3) X = X + υ, υ N(0, συ) 2 (4) cov(υ, X) = 0 (5) cov(ɛ, υ) = 0 (6) y = Vote share, X = unemployment measured with error, X = unemployment.

10 OLS + Fixed Effects Instrumental Variables : Measurement Error with OLS Measurement error biases β toward zero. ˆβ = cov(x,y) plim ˆβ = β λ = var(x ) = cov(x+υ,xβ+ɛ) var(x+υ) (7) σ2 x = λβ (8) σx 2 +συ 2 σ2 x σ 2 x +σ 2 υ (9) Since 0 < λ < 1, β is biased toward zero. y = Vote share, X = unemployment measured with error, X = unemployment.

11 OLS + Fixed Effects Instrumental Variables : Measurement Error with Fixed Effects Inclusion of fixed effects can increase measurement error attenuation by a factor of four. Panel data set with repeated measures over two time periods, T = 2 (equivalent to first differences). Now: β fe = cov( X, Y ) var( X ) (10) where X = (X st + υ st ) (X st 1 + υ st 1 ) (11) Y = Y st Y st 1 (12)

12 OLS + Fixed Effects Instrumental Variables : Measurement Error with Fixed Effects It can be shown that, with measurement error: plimβ fe = β σx 2 (1 π X ) σx 2 (1 π (13) X )+συ 2 (1 πυ π X = cov(x st, X st1 )/var(x st ) (14) π υ = cov(υ st, υ st 1 )/var(υ st (15) Rearranging terms, we find that the bias is greatest when π X > π υ. This generally occurs in fixed effects models as serial correlation within units over time is generally very high.

13 OLS + Fixed Effects Instrumental Variables : Instrumental Variables as a Solution An instrument, Z can identify β if it meets the following conditions: Condition 1: Instrument Strength: It is strongly correlated with X* (unemployment measured with error). Condition 2: Exclusion Restriction: It is uncorrelated with ɛ.

14 OLS + Fixed Effects Instrumental Variables : Instrumental Variables as a Solution Measurement error in X becomes irrelevant when we instrument: βˆ IV = cov(y,z ) cov(x,z ) = cov(βx+ɛ,z ) cov(x+υ,z ) plim ˆ β IV = β σ2 XZ σ 2 XZ = β

15 Model Instrument : Model The Model: X st = β 1 Z st + β 2 %M st + W st γ + ψ s + δ t + η st y st = α 1 ˆXst + α 2 %M st + W st γ + ψ s + δ t + ɛ st Estimate α 1 using 2SLS. X st = State unemployment, y st = Incumbent vote share, Z st = Exchange rate instrument, ψ s = State fixed effects, δ t Year fixed effects.

16 Model Instrument : Instrument Instrument for state unemployment: real exchange rate shocks to state manufacturing employment as an instrument for state unemployment. (Lin 2006) Z st = E %M st = E t E t 1 E t 1 %M st Intuition - States unemployment differentially determined by quasi-random exogenous shocks (set by the world market). E = Percent change in real exchange Rate, %M st = Percent employed in manufacturing.

17 Model Instrument : Testing IV Assumptions Condition 1: Instrument Strength - results below, show that instrument meets criteria for strong instrument. F-Statistic > 10 on the excluded instrument. Condition 2: Exclusion Restriction Untestable in practice. Given our model, a violation would require correlation of Z with omitted time-varying characteristics of states.

18 OLS + Fixed Effects 2SLS IV TABLE 1. LEAST SQUARES MODELS OF THE EFFECT OF UNEMPLOYMENT RATES ON INCUMBENT VOTE SHARE (1) (2) (3) (4) (5) PANEL A. CONGRESSIONAL QUARTERLY DATA Incumbent Vote ** ** (0.002) (0.002) (0.002) (0.003) (0.003) Democratic Incumbent Vote * (0.004) (0.007) (0.005) (0.005) (0.006) Republican Incumbent Vote *** (0.003) (0.006) (0.005) (0.005) (0.005) PANEL B. ANES DATA % Party Switch (0.003) (0.005) (0.006) (0.008) (0.009) % Democrat Switch (0.003) (0.006) (0.006) (0.009) (0.012) % Republican Switch (0.006) (0.012) (0.005) (0.015) (0.015) year effects no yes yes yes yes state effects no no yes yes yes covariates no no no yes yes time trends no no no no yes

19 Introduction OLS + Fixed Effects 2SLS IV OLS Coefficient Plot: Unemployment, Aggregate OLS Coefficient Plot: Unemployment, ANES IncumbVote % Party Switch Democratic Incumbent Vote Republican Incumbent Vote % Democrat Switch Democrat Vote Republican Vote % Republican Switch (a) CCES Data (b) ANES Data Figure: Model (5): LS Estimates of DVs on Unemployment

20 OLS + Fixed Effects 2SLS IV TABLE 2. 2SLS MODELS OF THE EFFECT OF UNEMPLOYMENT RATES ON INCUMBENT VOTE SHARE (1) (2) (3) (4) (5) PANEL A. CONGRESSIONAL QUARTERLY DATA Incumbent Vote ** -0.04** ** ** *** (0.005) (0.019) (0.016) (0.018) (0.018) Democratic Incumbent Vote *** ** ** *** *** (0.009) (0.024) (0.021) (0.021) (0.021) Republican Incumbent Vote 0.022*** (0.008) (0.017) (0.016) (0.017) (0.018) F-Statistic on instrument PANEL B. ANES DATA % Party Switch ** (0.012) (0.027) (0.026) (0.026) (0.029) % Democrat Switch 0.031** ** (0.014) (0.032) (0.030) (0.028) (0.013) % Republican Switch *** 0.066** (0.027) (0.033) (0.032) (0.038) (0.041) F-Statistic on instrument year effects no yes yes yes yes state effects no no yes yes yes covariates no no no yes yes time trends no no no no yes

21 Introduction OLS + Fixed Effects 2SLS IV IV Coefficient Plot: Unemployment, Aggregate IV Coefficient Plot: Unemployment, ANES IncumbVote % Party Switch Democratic Incumbent Vote Republican Incumbent Vote % Democrat Switch Democrat Vote Republican Vote % Republican Switch (a) CCES Data (b) ANES Data Figure: Model (5): 2SLS Estimates of DVs on Unemployment

22 OLS + Fixed Effects 2SLS IV Conclusions Congressional Economic Voting and Local Unemployment Local unemployment matters in congressional voting. Support for the conclusion that its importance has been underestimated in the past. Partisan Accountability Strong support for partisan accountability. Only Democratic incumbents punished for increases in state unemployment. Suggests that Democratic partisans tend to hold their incumbents accountable while Republicans do not.