A Short Course in Political Economy - Michigan · PDF fileA Short Course in Political Economy...

Overview Do Voters A�ect or Elect Policies? The Value of School Facility Investments Media Bias and In�uence Momentum and Social Learning in Presidential Primaries

A Short Course in Political Economy

Brian Knight1

1Brown University

November 15, 2014


Outline

1 Overview

2 Do Voters A�ect or Elect Policies?

3 The Value of School Facility Investments

4 Media Bias and In�uence

5 Momentum and Social Learning in Presidential Primaries

2/60


Political economy tradition

Political economy: the study of relationships between

economic and political factors.

The public choice tradition emphasized rent-seeking and the

possibility of government failure.

A literature on social choice emphasizes group decision making

when individuals have con�icting preferences.

A literature on rational choice uses game theory to study the

role of political institutions in shaping economic policy, often

with a focus on incomplete voter information.

While these theoretical contributions have continued, the

literature has also recently gained an empirical focus, and my

course provides a selective overview of these contributions.

3/60















3/60















3/60















3/60















3/60


What can political economy learn from public economics?

The tradition of normative analysis in public economics,

focused on characterizing optimal policies, can be extended to

political economy.

In particular, political economists have characterized optimal

institutions, the constitution that leads to the implementation

of optimal policies.

For example, Maskin and Tirole (2004) study whether o�cials

should to accountable (�politicians�) or nonaccountable

("judges") in a model in which elections allow voters to

monitor o�cials but also provide incentive for o�cials to

�pander� to public opinion.

Hummel and Knight (2014) investigate whether sequential

elections or simultaneous elections are better for voter welfare.

4/60





political economy.











4/60





political economy.











4/60





political economy.











4/60


What can public economics learn from political economy?

Public economists interested in measuring the economic impact

of policies can exploit variation in policies induced by politics.

For example, Cellini, Ferreira, and Rothstein (2010) exploit

close bond elections in California to estimate the impact of

education spending on housing values.

Likewise, Knight (2003) uses variation in federal grants

induced by Congressional representation to examine the impact

of federal grants on state government �nances.

5/60











5/60











5/60


Outline

Today I will focus on two areas of recent research activity.

Voting and Regression Discontinuity

Lee, Moretti, and Butler (2004)Cellini, Ferreira, and Rothstein (2010)

Voter Learning, the Media, and Sequential Elections

Chiang and Knight (2010)Knight and Schi� (2010) and Knight and Hummel (2014)

6/60


Outline






6/60


Outline






6/60


Outline

1 Overview





7/60


Introduction

There are two distinct views of the electoral process.

Electoral competition is su�cient to induce candidates tomoderate platforms towards voter preferences. In the extreme,platforms converge to the median voter. In this case, voters�a�ect� policies.Candidate platforms are �xed and voters must choose betweentwo extremists. This is consistent with politicians unable tocommit (Besley and Coate, 1997). In this case, voters �elect�policies.

Whether voters �a�ect� or �elect� policies has important

implications for voter welfare, redistricting, and political

polarization.

8/60


Introduction


Electoral competition is su�cient to induce candidates tomoderate platforms towards voter preferences. In the extreme,platforms converge to the median voter. In this case, voters�a�ect� policies.

Candidate platforms are �xed and voters must choose betweentwo extremists. This is consistent with politicians unable tocommit (Besley and Coate, 1997). In this case, voters �elect�policies.



polarization.

8/60


Introduction





polarization.

8/60


Introduction





polarization.

8/60


Overview

To investigate this issue, the authors exploit variation in the

degree of electoral competition.

Let Dt indicate whether a Democrat holds a seat in time t,and let P∗

t denote the �electoral strength� of the Democrat at

time t.

Incumbency advantage: the probability of a Democrat winning

in t+1 is higher if a Democrat holds a seat in t (PDt+1> PR

t+1).

Platforms at time t are measured by post-election roll-call

votes (RCt).

Then, if voters �a�ect� policies, a Republican winning in tcauses both candidates to shift platforms to the right in t + 1.

If voters �elect� policies, candidate platforms are unchanged.

9/60


Overview





time t.



t+1).


votes (RCt).



9/60


Overview





time t.



t+1).


votes (RCt).



9/60


Overview





time t.



t+1).


votes (RCt).



9/60


Overview





time t.



t+1).


votes (RCt).



9/60


Overview





time t.



t+1).


votes (RCt).



9/60


Voters �elect� policies

D R

D R

D R

D R

Dt=1 Dt=0

RCt

RCt+1

10/60


Voters �a�ect� policies

D R

Dt=1 Dt=0

RCt

RCt+1D R

D R

D R

11/60


Statistical framework

Based upon Alesina (1988), platforms at t can be written as:

RCt = α+ π0P∗t + π1Dt + εt

Two challenges: P∗t is unobserved and Dt is correlated with εt .

The authors overcome these by using close elections.

Platforms in t + 1 as a function of the (randomly chosen)

winner in t can be written as:

γ = E (RCt+1|Dt = 1)− E (RCt+1|Dt = 0)

γ = π0[P∗Dt+1 − P∗R

t+1]︸︷︷︸affect

+π1[PDt+1 − PR

t+1]︸︷︷︸elect

where π1 =E (RCt |Dt = 1)− E (RCt |Dt = 0).

Full convergence (π1 = 0): voters only a�ect policies.

Full divergence (π0 = 0): voters only elect policies.

12/60









γ = E (RCt+1|Dt = 1)− E (RCt+1|Dt = 0)

γ = π0[P∗Dt+1 − P∗R


+π1[PDt+1 − PR





12/60









γ = E (RCt+1|Dt = 1)− E (RCt+1|Dt = 0)

γ = π0[P∗Dt+1 − P∗R


+π1[PDt+1 − PR





12/60









γ = E (RCt+1|Dt = 1)− E (RCt+1|Dt = 0)

γ = π0[P∗Dt+1 − P∗R


+π1[PDt+1 − PR





12/60









γ = E (RCt+1|Dt = 1)− E (RCt+1|Dt = 0)

γ = π0[P∗Dt+1 − P∗R


+π1[PDt+1 − PR





12/60









γ = E (RCt+1|Dt = 1)− E (RCt+1|Dt = 0)

γ = π0[P∗Dt+1 − P∗R


+π1[PDt+1 − PR




Full divergence (π0 = 0): voters only elect policies.12/60


Data

The authors examine various measures of voting records in the

U.S. House of Representatives during the period 1946-1995

Americans for Democratic Action (ADA) scores, which arehigher for more liberal voting patterns.DW-Nominate scores (Poole and Rosenthal, 1985).

The regression analysis is restricted to close elections, de�ned

as cases where the Democratic vote share was between 48 and

52 percent.

The authors show that the standard conditions for the RD

design are satis�ed (e.g., there are no discontinuities in district

characteristics at 50 percent).

13/60


Data






52 percent.




13/60


Data






52 percent.




13/60


Total e�ect of Democrat at t on RCt+1

828 QUARTERLY JOURNAL OF ECONOMICS

100

0.00 0.25 0.50 0.75 1.00

Democratic Vote Share, time t

Figure I

Total Effect of Initial Win on Future ADA Scores: 7 This figure plots ADA scores after the election at time t + 1 against the

Democrat vote share, time t. Each circle is the average ADA score within 0.01 intervals of the Democrat vote share. Solid lines are fitted values from fourth order polynomial regressions on either side of the discontinuity. Dotted lines are

pointwise 95 percent confidence intervals. The discontinuity gap estimates

7 = iroiPKx

- PT+i) + irxGP??!

- PFEi).

N-v-/ V-v-/ "Affect" "Elect"

be a continuous and smooth function of vote shares everywhere,

except at the threshold that determines party membership. There

is a large discontinuous jump in ADA scores at the 50 percent threshold. Compare districts where the Democrat candidate

barely lost in period t (for example, vote share is 49.5 percent), with districts where the Democrat candidate barely won (for

example, vote share is 50.5 percent). If the regression disconti

nuity design is valid, the two groups of districts should appear ex

ante similar in every respect?on average. The difference will be

that in one group, the Democrats will be the incumbent for the next election (t 4- 1), and in the other it will be the Republicans. Districts where the Democrats are the incumbent party for elec

tion t + 1 elect representatives who have much higher ADA

scores, compared with districts where the Republican candidate

14/60


E�ect of Democrat at t on RCt


<

<

0.00 0.25 0.50 0.75


Figure lia

Effect of Party Affiliation: -n1

1.00

M#.-#.)

0.00 0.25 0.50 0.75


1.00

Figure lib

Effect of Initial Win on Winning Next Election: {Pf+1 -

Pf+1)

Top panel plots ADA scores after the election at time t against the Democrat vote share, time t. Bottom panel plots probability of Democrat victory at t + 1

against Democrat vote share, time t. See caption of Figure III for more details.

15/60


Incumbency advantage


<

<

0.00 0.25 0.50 0.75


Figure lia

Effect of Party Affiliation: -n1

1.00

M#.-#.)

0.00 0.25 0.50 0.75


1.00

Figure lib

Effect of Initial Win on Winning Next Election: {Pf+1 -

Pf+1)

Top panel plots ADA scores after the election at time t against the Democrat vote share, time t. Bottom panel plots probability of Democrat victory at t + 1

against Democrat vote share, time t. See caption of Figure III for more details. 16/60


Decomposition into �Elect� and �A�ect�


TABLE I Results Based on ADA Scores?Close Elections Sample

Total effect Elect component Affect component

7 ^ (Pf+1-/f4l)^[(P?+1-Pf+1)] TTolP*?! -

PtR+1\ Variable ADAt + l ADAt DEM, + 1 (col. (2)*(col. (3)) (col. (1))

- (col. (4))

(1) (2) (3) (4) (5)

Estimated gap 21.2 47.6 0.48

(1.9) (1.3) (0.02) 22.84 -1.64

(2.2) (2.0)

Standard errors are in parentheses. The unit of observation is a district-congressional session. The sample includes only observations where the Democrat vote share at time t is strictly between 48 percent and 52 percent. The estimated gap is the difference in the average of the relevant variable for observations for which the Democrat vote share at time t is strictly between 50 percent and 52 percent and observations for which the Democrat vote share at time t is strictly between 48 percent and 50 percent. Time t and t + 1 refer to congressional sessions. ADAt is the adjusted ADA voting score. Higher ADA scores correspond to more liberal roll-call voting records. Sample size is 915.

primarily elect policies (full divergence) rather than affect poli cies (partial convergence).

Here we quantify our estimates more precisely. In the analy sis that follows, we restrict our attention to "close elections"?

where the Democrat vote share in time t is strictly between 48

and 52 percent. As Figures I and II show, the difference between

barely elected Democrat and Republican districts among these elections will provide a reasonable approximation to the discon

tinuity gaps. There are 915 observations, where each observation

is a district-year.20

Table I, column (1), reports the estimated total effect 7, the

size of the jump in Figure I. Specifically, column (1) shows the difference in the average ADAt + 1 for districts for which the

Democrat vote share at time t is strictly between 50 percent and

52 percent and districts for which the Democrat vote share at

time t is strictly between 48 percent and 50 percent. The esti

mated difference is 21.2.

In column (2) we estimate the coefficient nly which is equal to

the size of the jump in Figure Ha. The estimate is the difference

in the average ADAt for districts for which the Democrat vote

20. In 68 percent of cases, the representative in period t + 1 is the same as the representative in period t. The distribution of close elections is fairly uniform across the years. In a typical year there are about 40 close elections. The year with the smallest number is 1988, with twelve close elections. The year with the largest number is 1966, with 92 close elections.

17/60


Outline

1 Overview





18/60


Introduction

Is public spending capitalized into house prices?

If so, is public spending above or below e�cient levels?

If above, an increase in public spending should decreasehousing prices.If below, an increase in public spending should increasehousing prices.

The authors examine the passage of bond elections in

California, which require voter approval, during the period

1988-2005 using data on housing transactions.

19/60


Introduction







19/60


Introduction



If above, an increase in public spending should decreasehousing prices.

If below, an increase in public spending should increasehousing prices.




19/60


Introduction







19/60


Introduction







19/60


Framework

Consider the following regression equation linking housing

prices (yj) to an indicator for the passage of a bond (bj):

yj = κ+ bjθ + uj

If θ > 0 then public spending is too low

If θ < 0 then public spending is too high.

The key identi�cation problem is that unobserved preferences

for education are likely correlated with both support for bonds

and housing prices.

The authors address this using a regression discontinuity

design, under the assumption that failing municipalities are

otherwise equivalent to passing municipalities.

20/60


Framework



yj = κ+ bjθ + uj





and housing prices.




20/60


Framework



yj = κ+ bjθ + uj





and housing prices.




20/60


Framework



yj = κ+ bjθ + uj





and housing prices.




20/60


Framework



yj = κ+ bjθ + uj





and housing prices.




20/60


Bond passage increases expenditures and capital outlays

240 QUARTERLY JOURNAL OF ECONOMICS0

50

01

,00

01

,50

0M

ea

n t

ota

l exp

en

ditu

res

pe

r p

up

il

0 5 10Vote share relative to threshold (2 pp bins)

Total expenditures

05

00

1,0

00

1,5

00

Me

an

ca

pita

l ou

tlays

pe

r p

up

il

0 5 10Vote share relative to threshold (2 pp bins)

Year before election

Three years after election

Capital outlays

FIGURE IITotal Spending and Capital Outlays per Pupil, by Vote Share, One Year before

and Three Years after ElectionGraph shows average total expenditures (left panel) and capital outlays (right

panel) per pupil, by the vote share in the focal bond election. Focal elections aregrouped into bins two percentage points wide: measures that passed by between0.001% and 2% are assigned to the 1 bin; those that failed by similar margins areassigned to the −1 bin. Averages are conditional on year fixed effects, and the −1bin is normalized to zero.

after the election, districts where the measure just passed spendabout $1,000 more per pupil, essentially all of it in the capitalaccount.31

Panel A of Table IV presents estimates of the intent-to-treateffect of bond passage on district spending and on state and fed-eral transfers (all in per-pupil terms) over the six years followingthe election, using equation (7).32 Bond passage has no significanteffect on any of the fiscal variables in the first year. We see largeincreases in capital expenditures in years 2, 3, and 4. These in-creases fade by the fifth year following the election. There is noindication of any effect on current spending in any year, and con-fidence intervals rule out effects amounting to more than about

31. It is possible that districts use bond revenues for operating expenses butreport these expenditures in their capital accounts. The CCD data are not used forfinancial oversight, so districts have no obvious incentive to misreport.

32. We make one modification to equation (7): We constrain the τ = 0 coeffi-cients to zero. It is not plausible that bond passage can have effects on that year’sdistrict budget, which will typically have been set well before the election. In anycase, results are insensitive to removing this constraint.

21/60


Bond passage increases housing pricesTHE VALUE OF SCHOOL FACILITY INVESTMENTS 245

0–0.1

0.1

0.2

0.3

Mean lo

g h

ousi

ng p

rice

s

–10 –5 0 5 10

Vote share relative to threshold (2 pp bins)

Year before election Three years after election

FIGURE VLog Housing Prices by Vote Share, One Year before and Three Years after

ElectionGraph shows average log housing prices by the vote share in the focal bond

election. Focal elections are grouped into bins two percentage points wide: mea-sures that passed by between 0.001% and 2% are assigned to the 1 bin; those thatfailed by similar margins are assigned to the −1 bin. Averages are conditional onyear fixed effects, and the −1 bin is normalized to zero.

uniformly significant after year 0. The estimates indicate that theTOT effect of bond approval in year t is to increase average pricesby 2.8%–3.0% that year, 3.6%–4.1% in year t + 1, 4.2%–8.6% inyears t + 2 through t + 5, and 6.7%–10.1% in t + 6. Figure VI plotsthe coefficients and confidence intervals from the two dynamicspecifications, showing estimates out to year 15. The recursiveestimator shows growing effects through almost the entire period,whereas the one-step estimator yields a flatter profile. Confidenceintervals are wide, particularly for the recursive estimator in laterperiods, and a zero effect is typically at or near the lower boundof these intervals.35

As discussed in Section IV, the TOT estimators assume thathouse prices are unaffected by the likelihood of a future bond

35. We have also estimated models that constrain the TOT to be constantover time. With our one-step estimator, we obtain a point estimate of 4.9% and astandard error of 1.7%.

22/60


WTP for education spending

In addition to the sign of the e�ect, the authors are interested

in the WTP for an additional dollar of spending.

Consider the passage of a bond increasing spending per-pupil

by $6,000 and �nanced by a $6,000 increase in per-pupil taxes.

If housing prices in a passing municipality increase by θ,relative to a failing municipality, then we have that:

θ = 6, 000 ∗WTP − 6000

Solving for WTP , we have that:

WTP = 1+θ

6, 000

If θ > 0, then WTP > 1 and if θ < 0, then WTP < 1.

23/60








θ = 6, 000 ∗WTP − 6000


WTP = 1+θ

6, 000


23/60








θ = 6, 000 ∗WTP − 6000


WTP = 1+θ

6, 000


23/60








θ = 6, 000 ∗WTP − 6000


WTP = 1+θ

6, 000


23/60








θ = 6, 000 ∗WTP − 6000


WTP = 1+θ

6, 000


23/60


But failing municipalities are future passing municipalities

THE VALUE OF SCHOOL FACILITY INVESTMENTS 243

–0.5

–0.2

50

0.2

50.5

0.7

51

Effect

on p

r(pass

bond)

0 3 6 9 12 15Year (relative to election)

95% CI

Reduced-form RD estimate

FIGURE IIIEstimates of the Effect of Bond Passage on the Probability of Passing a Later

Bond, by Years since the Focal ElectionGraph shows coefficients and 95% confidence intervals for the effect of measure

passage in year t on the probability of passing a measure in year t + τ . The spec-ification is the ITT regression described in equation (7). Sample includes relativeyears −19 through +19, excluding relative year 0 (when the effect is mechanicallyone).

confidence intervals, particularly at long lags. When we discountall of the estimated effects from the one-step estimator back to thedate of the election, using a discount rate of 7.33% as in Barrowand Rouse (2004), the effect of authorizing a bond is to increasethe present value of future spending by $5,671. This is quite sim-ilar to the size of the average bond proposal in close elections,$6,309.

VII. RESULTS

VII.A. Housing Prices

Figure V provides a graphical analysis of the impact of bondpassage on log housing prices corresponding to the analyses offiscal outcomes in Figure II. Two important patterns emerge.First, housing prices in the year before the election are positivelycorrelated with vote shares, indicating that higher priced districtsare more likely to pass bond measures with larger margins of

24/60


WTP with dynamics

To address this, consider a two-period case in which failing

municipalities in t pass a bond in t + 1 with probability π.

Then, the relationship between the increase in housing prices

and WTP is given by:

θ = 6, 000 ∗WTP − 6000− πδ[6, 000 ∗WTP − 6000]

WTP is then adjusted as follows:

WTP = 1+θ

6, 000(1− πδ)

See the paper for more general dynamic relationships.

With estimates of π, WTP is estimated to be between $1.13

and $1.50.

25/60


WTP with dynamics





θ = 6, 000 ∗WTP − 6000− πδ[6, 000 ∗WTP − 6000]


WTP = 1+θ

6, 000(1− πδ)



and $1.50.

25/60


WTP with dynamics





θ = 6, 000 ∗WTP − 6000− πδ[6, 000 ∗WTP − 6000]


WTP = 1+θ

6, 000(1− πδ)



and $1.50.

25/60


WTP with dynamics





θ = 6, 000 ∗WTP − 6000− πδ[6, 000 ∗WTP − 6000]


WTP = 1+θ

6, 000(1− πδ)



and $1.50.

25/60


WTP with dynamics





θ = 6, 000 ∗WTP − 6000− πδ[6, 000 ∗WTP − 6000]


WTP = 1+θ

6, 000(1− πδ)



and $1.50.

25/60


Other Voting RD Applications

Ferreira and Gyourko, �Do political parties matter? Evidence

from US cities�, Quarterly Journal of Economics, 2009.

Ferraz and Finan, �Electoral accountability and corruption:

Evidence from the audits of local governments�, American

Economic Review, 2011.

Pettersson-Lidbom, Per, "Do Parties Matter for Economic

Outcomes: A Regression Discontinuity Approach", Journal of

the European Economic Association, 2008.

Dal Bó, Dal Bó, and Snyder, �Political dynasties�, The Review

of Economic Studies, 2009.

Folke and Snyder. "Gubernatorial midterm slumps." American

Journal of Political Science, 2012.

Brollo and Nannicini. "Tying your enemy's hands in close

races: The politics of federal transfers in Brazil." American

Political Science Review, 2012.26/60


Outline

1 Overview





27/60


Introduction

While the media may provide valuable information to voters,

they may also be biased.

If voters do not account for bias, then biased outlets may lead

to poor voter decisions, such as electing low quality candidates.

Research on this is mixed:

DellaVigna and Kaplan (2007) show that the introduction ofFox News increased votes for Republican candidatesGentzkow, Shapiro and Sinkinson (2011) show that the entryof partisan newspapers did not in�uence voting.In a �eld experiment, Bergan, Gerber, and Karlan (2009) showthat newspapers increase votes for Democrats.

This paper examines newspaper endorsements in Presidential

elections. If voters account for bias, then a Republican

endorsement from a right-leaning paper should have less

in�uence than a Democratic endorsement.

28/60


Introduction











28/60


Introduction











28/60


Introduction






DellaVigna and Kaplan (2007) show that the introduction ofFox News increased votes for Republican candidates

Gentzkow, Shapiro and Sinkinson (2011) show that the entryof partisan newspapers did not in�uence voting.In a �eld experiment, Bergan, Gerber, and Karlan (2009) showthat newspapers increase votes for Democrats.





28/60


Introduction






DellaVigna and Kaplan (2007) show that the introduction ofFox News increased votes for Republican candidatesGentzkow, Shapiro and Sinkinson (2011) show that the entryof partisan newspapers did not in�uence voting.

In a �eld experiment, Bergan, Gerber, and Karlan (2009) showthat newspapers increase votes for Democrats.





28/60


Introduction











28/60


Introduction










in�uence than a Democratic endorsement.28/60


Model

Voters di�er in their ideology (iv ), with increases in ideology

associated with movements to the right.

Two candidates c ∈ D,R up for election di�er in terms of their

quality (qc) and their ideology (ic).

Voters value quality independent of their ideology, have a

preference for candidates with similar ideology, and thus

receive the following payo� from candidate c winning:

Uvc = qc − (ω/2)(iv − ic)²

Under certain conditions, the di�erence in utility for voter vcan be written as a function of the di�erence in candidate

quality (q = qD − qR).

∆v = UvD − UvR = q − ωiv (iR − iD)

29/60


Model








Uvc = qc − (ω/2)(iv − ic)²




29/60


Model








Uvc = qc − (ω/2)(iv − ic)²




29/60


Model








Uvc = qc − (ω/2)(iv − ic)²




29/60


Newspapers

Newspapers receive a signal about relative quality: θn = q + εn

Each newspaper has an editorial position (pn), the �standard�

for endorsing the Democrat:

en = 1[θn/√σ2q + σ2ε > pn]

Voters reading an endorsement for the Democrat update via:

E (q|en) = βλd(pn)

where the �credibility� of the endorsement is given by:

λd(pn) =φ(pn)

1− Φ(pn)

Endorsements for Democrats from right-leaning papers have

more in�uence.

30/60


Newspapers




en = 1[θn/√σ2q + σ2ε > pn]




λd(pn) =φ(pn)

1− Φ(pn)


more in�uence.

30/60


Newspapers




en = 1[θn/√σ2q + σ2ε > pn]




λd(pn) =φ(pn)

1− Φ(pn)


more in�uence.

30/60


Newspapers




en = 1[θn/√σ2q + σ2ε > pn]




λd(pn) =φ(pn)

1− Φ(pn)


more in�uence.

30/60


Newspapers




en = 1[θn/√σ2q + σ2ε > pn]




λd(pn) =φ(pn)

1− Φ(pn)


more in�uence.30/60


Illustration

0.1

.2.3

.4y

-2 -1 0 1 2x

Example without noise--standard normal prior

31/60


Illustration

0.1

.2.3

.4y

-1.25 0 .2x

Endorsement for Democrat from left-leaning paper

32/60


Illustration

0.1

.2.3

.4y

0 1.25 1.73x

Endorsement for Democrat from right-leaning paper

33/60


Data

Key prediction: if voters account for media bias, then

surprising endorsements should have more in�uence than

unsurprising endorsements.

We test this prediction using data on the timing of newspaper

endorsements during the 2000 and 2004 Presidential elections.

We use three variables in the Annenberg National Election

Surveys.

voting intentionsdate of the interviewnewspaper readership

34/60


Data







Surveys.


34/60


Data







Surveys.


34/60


Data







Surveys.

voting intentions

date of the interviewnewspaper readership

34/60


Data







Surveys.

voting intentionsdate of the interview

newspaper readership

34/60


Data







Surveys.


34/60


Measuring newspaper ideology

The literature on media bias has focused on two sources of

media bias:

According to the demand side view, bias originates from voterpreferences for like-minded news and pro�t-maximizing mediaoutlets. This includes Gentzkow and Shapiro (2010) andMullainathan and Schleifer (2005)According to the supply side view, bias originates from ownersof media outlets desiring to in�uence voters, as emphasized byDjankov et. al. (2003) and Baron (2006) .

Given these two sources, we measure newspaper ideology via a

statistical model that includes demand side factors (reader

preferences prior to the endorsement) and supply side factors

(group ownership).

We also estimate speci�cations that use information on the

historical pattern of Presidential endorsements.

35/60




media bias:

According to the demand side view, bias originates from voterpreferences for like-minded news and pro�t-maximizing mediaoutlets. This includes Gentzkow and Shapiro (2010) andMullainathan and Schleifer (2005)

According to the supply side view, bias originates from ownersof media outlets desiring to in�uence voters, as emphasized byDjankov et. al. (2003) and Baron (2006) .




(group ownership).



35/60




media bias:





(group ownership).



35/60




media bias:





(group ownership).



35/60




media bias:





(group ownership).



35/60


Endorsements and voting Intentions

“rdq037” — 2011/6/14 — 6:59 — page 807 — #13

CHIANG & KNIGHT MEDIA BIAS AND INFLUENCE 807

TABLE 4Influence of top20newspapers in 2000†

Reader support Probability of Actual ImpliedNewspaper for Gore (%) Group owner‡ endorsing Gore (%) endorsement influence(%)

New York Times 75 New York Times 90 Gore 0∙50Washington Post 64 – 54 Gore 2∙10New York Daily News 67 – 58 Gore 1∙90Chicago Tribune 53 – 36 Bush -1∙70Newsday 57 – 44 Gore 2∙60Houston Chronicle 39 Hearst 34 Bush -1∙60Dallas Morning News 35 – 17 Bush -0∙87Chicago Sun Times 67 – 58 Bush -2∙70Boston Globe 72 New York Times 89 Gore 0∙50San Francisco Chronicle 74 Hearst 82 Gore 0∙90Arizona Republic 41 – 20 Bush -1∙00New York Post 49 – 31 Bush -1∙50Rocky Mountain News 47 – 28 Bush -1∙30Denver Post 52 – 35 Gore 3∙10Philadelphia Inquirer 59 Knight Ridder 82 Gore 0∙90Union-Tribune 51 – 34 Bush -1∙60

†USA Today,Wall Street Journal, andLA Timesare not in this table because those newspapers did not make an endorse-ment or made a non-endorsement in 2000. Star Ledger is not included because the newspaper is not in the sample dueto missing ownership information in the ownership data in 2000.‡Missing (–) means that the newspaper is not owned by a group owner. Group owner is defined as a company that ownsmore than ten daily newspapers in the survey.

FIGURE 2Endorsements and voting

school, relative to high school dropouts, who are male, who attend religious services or considerthemselves born-again Christians are more likely to vote for the Republican.

To provide a sense of the magnitude of this effect of endorsements and endorsement credibil-ity, the final column of Table4 provides our implied estimates of the influence of endorsementsin the largest newspapers in the U.S. during the 2000 campaign. As shown, the least credible

at Brow

n University on O

ctober 24, 2011http://restud.oxfordjournals.org/

Dow

nloaded from

36/60


High credibility endorsements

“rdq037” — 2011/6/14 — 6:59 — page 808 — #14

808 REVIEW OF ECONOMIC STUDIES

FIGURE 3High-credibility endorsements and voting

FIGURE 4Low-credibility endorsements and voting

endorsements were for Gore from theNew York Timesand for Bush from theDallas MorningNews. According to the logic of our model, the higher probability of endorsing Gore by theNewYork Times, e.g.can be interpreted as having a lower standard in terms of information regardingthe quality of Gore, relative to Bush. Thus, these low-credibility endorsements from theNewYork Timesand theDallas Morning Newsconvinced less than 1% of their readers to switch theirallegiance to the endorsed candidate. The endorsements with the largest effect, by contrast, camefrom theDenver Postand theChicago Sun Times, both of which had surprising endorsements.According to our estimates, these endorsements convinced about 3% of readers to switch theirallegiance to the endorsed candidate. Interestingly, both of these newspapers switched their en-dorsements in 2004, when theChicago Sun TimesendorsedKerry and theDenver Postendorsed

at Brow

n University on O


Dow

nloaded from

37/60


Low credibility endorsements

“rdq037” — 2011/6/14 — 6:59 — page 808 — #14


FIGURE 3High-credibility endorsements and voting

FIGURE 4Low-credibility endorsements and voting

endorsements were for Gore from theNew York Timesand for Bush from theDallas MorningNews. According to the logic of our model, the higher probability of endorsing Gore by theNewYork Times, e.g.can be interpreted as having a lower standard in terms of information regardingthe quality of Gore, relative to Bush. Thus, these low-credibility endorsements from theNewYork Timesand theDallas Morning Newsconvinced less than 1% of their readers to switch theirallegiance to the endorsed candidate. The endorsements with the largest effect, by contrast, camefrom theDenver Postand theChicago Sun Times, both of which had surprising endorsements.According to our estimates, these endorsements convinced about 3% of readers to switch theirallegiance to the endorsed candidate. Interestingly, both of these newspapers switched their en-dorsements in 2004, when theChicago Sun TimesendorsedKerry and theDenver Postendorsed

at Brow

n University on O


Dow

nloaded from

38/60


Regression results

“rdq037” — 2011/6/14 — 6:59 — page 806 — #12


TABLE 3Second Stage: effect of newspaper endorsements on voteintention

Dependentvariable: 1 if intend to vote for the DemocratI II III

After×Credibility 0∙029** 0∙055**(0∙013) (0∙026)

After×Endorsement 0∙011 −0∙020(0∙008) (0∙017)

High school −0∙047*** −0∙047*** −0∙047***(0∙016) (0∙015) (0∙016)

College −0∙013 −0∙013 −0∙013(0∙016) (0∙016) (0∙016)

Male −0∙088*** −0∙087*** −0∙088***(0∙006) (0∙006) (0∙006)

Black 0∙440*** 0∙440*** 0∙440***(0∙009) (0∙008) (0∙009)

Age 0∙002** 0∙002** 0∙002**(0∙001) (0∙001) (0∙001)

Age squared 0∙000 0∙000 0∙000(0∙000) (0∙000) (0∙000)

Born-again Christian −0∙150*** −0∙150*** −0∙150***(0∙007) (0∙007) (0∙007)

Attend religious activities −0∙123*** −0∙123*** −0∙123***(0∙006) (0∙006) (0∙006)

Constant 0∙740*** 0∙740*** 0∙741***(0∙183) (0∙189) (0∙183)

Income categories Yes Yes YesNewspaper fixed effects Yes Yes YesDate fixed effects Yes Yes YesObservations 32,014 32,014 32,014

Notes:Standarderrors in parentheses, * denotes 90% significance, ** denotes 95% significance, and *** denotes 99%significance.

doesseem to be a slight widening in the gap between readers of Democrat-endorsing newspa-pers and readers of Republican-endorsing newspapers after publication of the endorsement. Theeffect is relatively small, however, and is somewhat difficult to detect. As evidence regardingthe prediction that endorsement influence depends upon its credibility, we next split the sam-ple into readers of newspapers with high-credibility endorsements and readers of newspaperswith low-credibility endorsements. As shown in Figure3, which focuses on high-credibility, orsurprising, endorsements, there appears to be an immediate and significant effect of the endorse-ment on reader voting intentions. In Figure4, by contrast, which focuses on low-credibility, orunsurprising endorsements, the effect is again small and difficult to visually detect. These re-sults suggest that high-credibility endorsements do have more influence than do low-credibilityendorsements, and we turn next to a more formal econometric examination of this hypothesis.

As shown in column 1 of Table3, which presents our second-stage results based upon ourfirst-stage estimates of credibility, we find support for the idea that endorsement credibility is akey determinant of the influence of the endorsement, as the coefficient on our credibility measureis positive and statistically significant. Regarding the other controls, we also find that voterswho are older or black are more likely to vote for Democrats and voters who complete high

at Brow

n University on O


Dow

nloaded from

39/60


In�uence of large newspapers“rdq037” — 2011/6/14 — 6:59 — page 807 — #13

CHIANG & KNIGHT MEDIA BIAS AND INFLUENCE 807

TABLE 4Influence of top20newspapers in 2000†

Reader support Probability of Actual ImpliedNewspaper for Gore (%) Group owner‡ endorsing Gore (%) endorsement influence(%)

New York Times 75 New York Times 90 Gore 0∙50Washington Post 64 – 54 Gore 2∙10New York Daily News 67 – 58 Gore 1∙90Chicago Tribune 53 – 36 Bush -1∙70Newsday 57 – 44 Gore 2∙60Houston Chronicle 39 Hearst 34 Bush -1∙60Dallas Morning News 35 – 17 Bush -0∙87Chicago Sun Times 67 – 58 Bush -2∙70Boston Globe 72 New York Times 89 Gore 0∙50San Francisco Chronicle 74 Hearst 82 Gore 0∙90Arizona Republic 41 – 20 Bush -1∙00New York Post 49 – 31 Bush -1∙50Rocky Mountain News 47 – 28 Bush -1∙30Denver Post 52 – 35 Gore 3∙10Philadelphia Inquirer 59 Knight Ridder 82 Gore 0∙90Union-Tribune 51 – 34 Bush -1∙60

†USA Today,Wall Street Journal, andLA Timesare not in this table because those newspapers did not make an endorse-ment or made a non-endorsement in 2000. Star Ledger is not included because the newspaper is not in the sample dueto missing ownership information in the ownership data in 2000.‡Missing (–) means that the newspaper is not owned by a group owner. Group owner is defined as a company that ownsmore than ten daily newspapers in the survey.

FIGURE 2Endorsements and voting

school, relative to high school dropouts, who are male, who attend religious services or considerthemselves born-again Christians are more likely to vote for the Republican.

To provide a sense of the magnitude of this effect of endorsements and endorsement credibil-ity, the final column of Table4 provides our implied estimates of the influence of endorsementsin the largest newspapers in the U.S. during the 2000 campaign. As shown, the least credible

at Brow

n University on O


Dow

nloaded from

40/60


Implications

We �nd that voters account for media bias.

While moving a newspaper to the right increases the likelihood

of a Republican endorsement, the in�uence of a Republican

endorsement by this newspaper falls.

These e�ects cancel out such that, from an ex-ante

perspective, changing the ideology of a newspaper has no

impact on voting.

41/60


Implications







impact on voting.

41/60


Implications







impact on voting.

41/60


More on the media economics literature

Groseclose and Milyo, �A Measure of Media Bias�, Quarterly

Journal of Economics, 2005.

Durante and Knight, �Partisan Control, Media Bias, and

Viewer Responses: Evidence from Berlusconi's Italy�, Journal

of the European Economic Association, 2012.

Gentzkow and Shapiro, �Ideological Segregation Online and

O�ine�, Quarterly Journal of Economics, 2011.

Snyder and Stromberg, �Press Coverage and Political

Accountability�, Journal of Political Economy, 2010.

Stromberg, �Radio's Impact on Public Spending, �, Quarterly

Journal of Economics, 2004.

Gentzkow, Shapiro and Sinkinson, �Competition and

Ideological Diversity: Historical Evidence from US

Newspapers�, American Economic Review, 2014.

42/60


Outline

1 Overview





43/60


Introduction

While all voters go to the polls on the same day in many

elections, states choose on di�erent days in other cases.

The most prominent example of a sequential election is the

Presidential primary system, under which Iowa and New

Hampshire vote �rst.

The key di�erence between sequential and simultaneous

systems is the ability of late voters to observe early returns

under sequential.

In this paper, we investigate whether late voters can learn

about the desirability of candidates from early returns.

44/60


Introduction








under sequential.



44/60


Introduction








under sequential.



44/60


Introduction








under sequential.



44/60


Challenges to measuring social learning via

Vt = ft(Vt−1,Vt−2, ...)

What is the information content of early returns?

How to account for unmeasured candidate quality?

How to weigh past returns (i.e. Vt−1 versus Vt−2)?

How to account for early learning:

Vt−1 = ft−1(Vt−2,Vt−3, ...)

45/60



Vt = ft(Vt−1,Vt−2, ...)





Vt−1 = ft−1(Vt−2,Vt−3, ...)

45/60



Vt = ft(Vt−1,Vt−2, ...)





Vt−1 = ft−1(Vt−2,Vt−3, ...)

45/60



Vt = ft(Vt−1,Vt−2, ...)





Vt−1 = ft−1(Vt−2,Vt−3, ...)

45/60


Simple model of sequential voting

Two states, early (A) and late (B) choose between two

candidates, the front-runner and the dark horse..

Voter i in state s receives the following utility from the

front-runner, relative to the dark horse, winning:

uis = q + ηs + νis

where q represents the quality of the front-runner,

ηs ∼ N(0, ση2) is a preference for the front-runner in state s,

and

νis is distributed type-I extreme value.

46/60










and


46/60










and


46/60










and


46/60










and


46/60


Information structure

Quality of the front-runner is unknown. Voters have normal

priors characterized by a mean µ1≥0 and variance σ12.

Voters in state s receive a signal (θs) over quality:

θs = q + εs

This signal is common within a state, private across states,

unbiased, and normally-distributed: εs ∼ N(0, σε2).

State-level preferences (ηs) are private across states.

All voters observe their signal before voting. Late voters also

observe the returns from state A.

47/60






θs = q + εs






47/60






θs = q + εs






47/60






θs = q + εs






47/60






θs = q + εs






47/60


Voting

Given their information, voters support the candidate that

provides the highest expected utility.

Given their signal over front-runner quality, early voters update

as follows:

E (q|θA) = αθA + (1− α)µ

where α = σ21/(σ21 + σ2ε )

Voting returns in the early state are then given by:

vA =exp[αθA + (1− α)µ+ ηA]

1+ exp[αθA + (1− α)µ+ ηA]

Late voters observe vA and attempt to infer θA. This is

complicated by the fact that late voters do not observe ηA.

48/60


Voting




as follows:

E (q|θA) = αθA + (1− α)µ

where α = σ21/(σ21 + σ2ε )






48/60


Voting




as follows:

E (q|θA) = αθA + (1− α)µ

where α = σ21/(σ21 + σ2ε )






48/60


Voting




as follows:

E (q|θA) = αθA + (1− α)µ

where α = σ21/(σ21 + σ2ε )






48/60


Voting




as follows:

E (q|θA) = αθA + (1− α)µ

where α = σ21/(σ21 + σ2ε )






48/60


Empirical application

Key prediction: support should increase among late voters for

candidates performing better than expected in early states.

We test this prediction and estimate the parameters of this

model using the 2004 Democratic primary, in which Iowa voted

�rst, followed by New Hampshire, etc.

We focus on the three key candidates (Dean, Kerry, and

Edwards) and use aggregate voting returns and daily survey

data from the 2004 Annenberg National Election Survey

In particular, we examine how voting intentions of late voters

in the survey respond to the release of early returns.

49/60













49/60













49/60













49/60


Support for Dean before and after Iowa

Fig. 2.—Dean before and after the Iowa primary

Fig. 3.—Kerry before and after the Iowa primary

50/60


Support for Kerry

Fig. 2.—Dean before and after the Iowa primary

Fig. 3.—Kerry before and after the Iowa primary

51/60


Support for Edwards

momentum and social learning 1123

Fig. 4.—Edwards before and after the Iowa primary

appropriately.14 Kerry, by contrast, outperformed expectations in Iowa,and survey respondents updated accordingly. Edwards also outper-formed his pre-Iowa polling numbers, and his polling numbers in-creased following Iowa. After a few days, however, his support fell backto pre-Iowa levels.15

B. Empirical Model

As noted above, our empirical strategy involves comparing support forcandidates among late voters before and after the release of votingreturns from early states. In our econometric specification, we assumethat voters receive their signals just prior to the date of their primary.That is, these respondents from late states have not yet observed their

14 The “Dean scream” will be discussed in detail below.15 These patterns are similar to those in prices from the Iowa Electronic Market, in

which market participants purchased contracts that would pay $1 in the event that Kerry,e.g., is the party’s nominee in the general election, and the price of this contract can thusbe interpreted as the probability that a given candidate wins the nomination (Wolfers andZitzewitz 2004). We choose to focus on polling data, rather than these prices from pre-diction markets, for two reasons. First, the mapping from voting in primaries to theprobability of nomination, as provided by the prediction market prices, is confounded bythe presence of superdelegates, who are not pledged to the winning candidate, as well asthe possibility that no candidate wins a majority of the delegates, in which case the nomineeis chosen through a bargaining process at the party convention. Second, the daily pollingdata, but not prediction market data, include additional measures of candidate quality,and we will make use of these in our discussion of alternative explanations to follow.

52/60


Model parameter estimates

Using this approach, we also estimate the key model

parameters: the initial prior over candidate quality (µc1), the

degree of variation in state-level preferences (ση2), and the key

learning parameters (σε2, σ1

2).

μD1 0.938**[0.773,1.14]

μE1 ‐0.701**[‐0.913,‐0.433]

ση2 0.815**

[0.551,1.194]σ1

2 3.577**[1.497,7.129]

σε2 1.197**

[0.062,4.097][bootstrap95%confidenceinterval],**denotessignificanceatthe95‐percentlevel

Table1:ParameterEstimatesfromKnightandSchiff(2010)

53/60


Supporting evidence of learning

Survey questions involving candidate quality (e.g. inspiring,

strong leader, trustworthy) move with support for candidates.

Early voters are better informed about candidate

characteristics than later voters. Moreover, this gap narrows as

the campaign progresses.

A structural break analysis documents breaks in support on

key early election dates (e.g. the Iowa Caucuses and the New

Hampshire Primaries).

54/60











54/60











54/60


Welfare analysis: sequential or simultaneous

In a companion paper, Hummel and Knight (2014) investigate

alternative systems, including a National Primary system.

Opponents of the current system argue that early states have

disproportionate in�uence, while supporters argue that it

enhances competition.

"...for years concerns have been raised regarding the calendarthat some believe gives a disproportionate in�uence to thesetwo early states", David Price, Commission on PresidentialNomination Timing and Scheduling, 10/1/2005."We need to preserve the possibility for lesser known, lesserfunded candidates to compete, and a national primary onFebruary 5th will not do that", Terry Shumaker, 12/5/2005.

55/60









55/60








"...for years concerns have been raised regarding the calendarthat some believe gives a disproportionate in�uence to thesetwo early states", David Price, Commission on PresidentialNomination Timing and Scheduling, 10/1/2005.

"We need to preserve the possibility for lesser known, lesserfunded candidates to compete, and a national primary onFebruary 5th will not do that", Terry Shumaker, 12/5/2005.

55/60









55/60



We show that sequential elections place too much weight on

information of early states but also provide late voters with

valuable information.

Under simultaneous elections, voters equally weigh the

available information but place too much weight on their

priors, providing an inappropriate advantage to front-runners.

Given these trade-o�s, simultaneous elections are preferred if

the front-runner initially has a small advantage, but sequential

elections are preferred if the front-runner has a large

advantage.

To evaluate this trade-o�, we simulate the model using the

Knight and Schi� (2010) parameter estimates and consider a

simultaneous system and an all public system.

56/60












advantage.




56/60












advantage.




56/60












advantage.




56/60


Information of early states is over-weighted

0.0

5.1

.15

coef

ficie

nt

0 5 10 15 20t

sequential publicsimultaneous

FIG 2: THE RESPONSE OF VOTES TO SIGNALS

57/60


Preferences of early states are over-weighted

0.0

5.1

.15

coef

ficie

nt

0 5 10 15 20t

sequential publicsimultaneous

FIG 3: THE RESPONSE OF VOTES TO PREFERENCES

58/60


Sequential is more competitive

System Pr(Dean) Pr(Edwards) Pr(Kerry)

all public 60.88% 12.61% 26.50%

simultaneous 68.91% 8.11% 22.98%

sequential 62.24% 11.84% 25.92%

59/60


Implications

Given the front runner advantage in 2004, the simultaneous

system outperforms the sequential system.

But the di�erences in welfare are small, and the result reverses

with a larger lead for the front-runner.

Finally, we consider a weighted sequential system. Under the

optimal weights, which gives voters from the latest state 18

times more in�uence than voters from the earliest state, this

system dominates both sequential and simultaneous.

Deltas, Herrera and Polborn (2010) examine a similar model

but allow for di�erences in ideology. With more than two

candidates, sequential primaries allow voters to better

coordinate as the election unfolds.

60/60


Implications













60/60


Implications













60/60


Implications













60/60

Date post:	09-Mar-2018
Category:	Documents
Upload:	nguyentuong
View:	220 times
Download:	2 times

A Short Course in Political Economy - Michigan · PDF fileA Short Course in Political Economy...

Documents