5 questions

European Economic Review 135 (2021) 103734

Contents lists available at ScienceDirect

European Economic Review

journal homepage: www.elsevier.com/locate/euroecorev

A theory of cultural revivals �

Murat Iyigun a , ∗, Jared Rubin b , Avner Seror c

a University of Colorado, Boulder, United States b Chapman University, United States c Aix-Marseille School of Economics, France

a r t i c l e i n f o

Article history:

Received 9 January 2020

Revised 1 April 2021

Accepted 3 April 2021

Available online 8 April 2021

JEL classification:

D02

N40

N70

O33

O38

O43

Z10

Keywords:

Institutions

Cultural beliefs

Cultural transmission

Institutional change

a b s t r a c t

Why do some societies have political institutions that support productively inefficient out-

comes? And why does the political power of elites vested in these outcomes often grow

over time, even when they are unable to block more efficient modes of production? We

propose an explanation centered on the interplay between political and cultural change.

We build a model in which cultural values are transmitted inter-generationally. The cul-

tural composition of society, in turn, determines public-goods provision as well as the fu-

ture political power of elites from different cultural groups. We characterize the equilib-

rium of the model and provide sufficient conditions for the emergence of cultural revivals .

These are characterized as movements in which both the cultural composition of society

as well as the political power of elites who are vested in productively inefficient outcomes

grow over time. We reveal the usefulness of our framework by applying it to two case

studies: the Jim Crow South and Turkey’s Gülen Movement.

© 2021 Elsevier B.V. All rights reserved.

1. Introduction

Why do some societies have political institutions that support productively inefficient outcomes? A common view in the

literature is that vested interests block the adoption of new technologies ( Acemoglu and Robinson, 2006; 2012; Chaudhry

and Garner, 20 06; 20 07 ). This interpretation is useful, but its explanatory power is limited to cases in which vested interests

actually have the power to block such changes. More often than not, the political power of vested interests is more limited

than this. As we shall elaborate below, there are salient historical cases in which social groups vested in inefficient produc-

� We are grateful for comments received from three anonymous referees, the editors, David Levine & Steffen Huck, as well as participants at workshops

at Stanford (AALIMS), Chapman (ASREC), the Ostrom Workshop at Indiana University, University of Colorado, UC Irvine, the Washington Area Economic

History Seminar, Hong Kong University’s Quantitative History Webinar, and the 2017 AEA Meetings. Eric Alston, Lee Alston, Cihan Artunç, Alberto Bisin, Lisa

Blaydes, Randall Calvert, Jean-Paul Carvalho, Zhiwu Chen, Jeffry Frieden, Avner Greif, Anna Harvey, Tonja Jacobi, Saumitra Jha, Asim Khwaja, Mark Koyama,

Timur Kuran, Trevon Logan, Debin Ma, Chicheng Ma, Luis Martinez, Victor Menaldo, John Patty, Tom Pepinsky, Armando Razo, Eitan Regev, Dani Rodrik,

Kenneth Shepsle, Tuan Hwee Sng, Jeffrey Staton, Melissa Thomasson, Thierry Verdier, John Wallis, Yongqin Wang, and Hye Young You provided extremely

useful comments. We also thank Kevin Takeda for very useful research assistance. Rubin received research funding from the John Templeton Foundation.

All errors are ours. ∗ Corresponding author.

E-mail addresses: [email protected] (M. Iyigun), [email protected] (J. Rubin), [email protected] (A. Seror).

https://doi.org/10.1016/j.euroecorev.2021.103734

0014-2921/© 2021 Elsevier B.V. All rights reserved.

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

tion modes and technologies were not able to directly block changes detrimental to their interests. Curiously, such groups

were still able to maintain their social and political dominance over time, and in some cases even came out ahead .

In this paper, we develop a theory that explains how elites vested in inefficient economic production are able to gain

political power despite the presence of more efficient modes of production which they are unable to block. We propose

that when elites have limited power to directly block modes of production detrimental to their interests, they can instead

influence a society’s culture . We present a model focusing on interactions between elites and citizens: elites provide public

goods, and their power to do so reflects the proportion of the citizens that share their cultural trait. The main mecha-

nism through which public-goods provision operates is by affecting citizens’ socialization decisions. Citizens care about the

welfare of their children and thus invest more effort in transmitting a cultural trait that better aligns with public-goods

provision. In turn, cultural changes strengthening or weakening a given cultural type lead to commensurate changes in the

political power of the elites. That is, political power changes in response to cultural change.

Our model provides insight into the emergence of cultural revivals . We define cultural revivals as occurring when two

conditions are satisfied. First, given the initial composition of the population, it is more efficient for elites to provide public

goods complementary to only one sector of the economy. Second, in spite of the first feature, the political power of the

elites who benefit from public-goods provision in the inefficient sector increases over time.

The primary insight underlying the existence of cultural revivals is that an economic disadvantage threatens the future

political power of elites who benefit from public-goods provision in the inefficient sector. Hence, on the margin, these

elites benefit significantly from the provisioning of public goods used by citizens that share their cultural trait, as this

affects socialization decisions which in turn affect future political weights. Under some conditions, which we derive in the

model, this consideration results in public-goods provision by the elites favoring the less efficient sector. This in turn triggers

cultural evolution in the direction of that sector.

Two historical examples help motivate the model. First, how did American white planter elites maintain their political

and economic power following Reconstruction? This is a puzzle: poor whites and freed blacks vastly outnumbered the white

elite, and the former two groups were mired in poverty. Political changes favoring the vast majority of the (poor) population

would certainly have improved the economic prospects of most Southerners. Conceivably, the white planter elites could

have lost their political power to any number of groups who tried to unite poor whites and freed blacks into a voting bloc.

Indeed, the Populists and Republicans attempted to create such an alliance. Yet, poor whites largely rejected such an alliance,

aligning culturally and politically with the white economic elites. Public goods favoring whites (e.g., segregated schools and

hospitals) were key to creating a more politically salient “white identity” that aligned much of the former Confederacy on

racial, rather than economic, lines. 1 After Reconstruction, the salience of white identity enabled white elites to strengthen

their grip on the Southern economy and politics. Jim Crow laws were a manifestation of this outcome. In the parlance of

our model, a cultural revival of racist values encouraged poor whites to align with white economic elites, which in turn

facilitated political changes strengthening the old economic and political structures.

Turkey’s Islamist Gülen Movement is another historical narrative that fits our theory well. After the Turkish Republic was

founded in 1923 following the collapse of the Ottoman Empire, there followed a host of mainly top-down sociopolitical and

economic reforms. These reforms were motivated by the fact that the returns to secular schooling and human capital had

risen markedly following the Industrial Revolution, whereas the economic productivity of the more established conservative

Ottoman culture had long been stagnant. These reforms were politically and economically empowering to a new group of

secular elites. Thus, throughout its nearly seven decade nascent existence, the Turkish Republic adhered to its French-style

“laicité” whereby the state hierarchy was fully under the control of secular elites. Nonetheless, this new regime did not tip

the political balance of power completely in one group’s favor given the culturally more conservative leanings of Turkish

society. With a succession of elections starting in 1995, Turkish Islamists were able to firmly regain and consolidate their

political power. This revival was a manifestation of deep-rooted cultural change spearheaded by investment in public goods.

The seeds of this cultural revival were sown at the end of Turkey’s single party era in 1950, when Islamist groups ratcheted

up their social and political activism. At the forefront of this movement was Fethullah Gülen, a religious cleric who mainly

focused on establishing K-12 schools which stressed the importance of quality education with an emphasis on science and

math proficiency. In the half century starting in the mid-1960s, the growth in Gülenist schools was remarkable. Gülenist

supporters and their associated Islamist culture became more prominent in Turkey after the 1990s. In line with the theory

we present below, this subsequently culminated in a shift in the balance of political power from the seculars to the Islamists

in Turkey.

Our model departs from—and adds to—the standard political economy explanation of institutional calcification, in which

stagnation occurs when it is in the interest of the politically powerful for the status quo to prevail ( Acemoglu, 2003; Ace-

moglu and Robinson, 2006; 2012 ). This view is rooted in North ’s (1990) idea that a society’s formal institutions —those polit-

ical, legal, social, and economic mechanisms that establish the formal “rules of the game” and the incentives faced by the

players therein—are the key drivers of economic and political outcomes. The “formal institutions of political economy” view

1 The type of public good investments we describe in these examples and in the model are different from conventional pork barrel spending. Whereas

pork barrel spending occurs when politicians win spending concessions for their own constituents, we are describing a broader effort by political elites of

the same “type” to provide targeted public goods to one part of the population. While both types of spending are targeted to keep the prevailing elites in

power, the type of spending we focus on works through altering society’s cultural composition in the longer run. On the other hand, pork barrel spending

works through a more straight-forward “buying votes” mechanism.

2

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

clearly explains many cases of economic and institutional stagnation. We complement this literature, as our theory implies

two key insights that cannot be explained by the “formal institutions of political economy” view. First, we demonstrate that

political changes which move society away from productively efficient outcomes can be triggered despite vested interests

lacking the capacity to directly enact such changes. Second, we provide an explanation as to why cultural changes are so

often linked to economic stagnation.

This paper is not the first in economics to suggest an interaction between cultural and political change. 2 In a closely

related work, Bénabou et al. (2021) develop a theory on the role that religion can play in preventing scientific progress.

Religion often plays a role in cultural revivals, as we conceptualize them, because the grasp religious elites tend to have

on “eternal truths” often means that new, more productive ways of doing things upset the status quo in which they are

powerful. In this light, Bénabou et al. (2021) focus on the threat that certain technologies pose to religious beliefs and how

this interacts with the political structure. By contrast, we are interested in the interaction between the cultural composition

of society and political power. Hence, the two views are highly complementary and help explain different, although related,

phenomena. More broadly, our theory can explain why not just religious, but also secular values can be leveraged by elites

in order to facilitate political change.

Our paper is related to the theoretical literature on religious and cultural leaders. 3 Hauk and Mueller (2015) present a

model in which individuals transmit their cultural norms and elites seek to spread their culture. They do so by interpreting

cultural aspects of their own and other cultures. Our model is close to Hauk and Mueller (2015) , as we also advance a theory

whereby culture is transmitted intergenerationally à la Bisin and Verdier (2001) and elites affect the incentive of parents to

pass their cultural values to their offspring. 4 We complement the related literature in two ways. First, existing works con-

sider elites that are not constrained in their ability to affect the cultural composition. By contrast, we incorporate our model

of cultural transmission within a broader political economy theory where the elites’ ability to affect the cultural composition

reflects their political power. We further model how the political power of elites changes over time in response to cultural

change. Second, the existing literature typically focuses on the effect of leaders on cultural diversity ( Prummer, 2019 ). 5 We

study the effect of elites on economic outcomes, and thereby connect as well to the growing empirical literature on the

effect of leaders on economic growth ( Jones and Olken, 2005; Yao and Zhang, 2015; George and Ponattu, 2020; Ferraz et al.,

2020 ).

The rest of our paper proceeds as follows. Section 2 lays out the model. Section 3 elaborates further on the formal

analysis to provide an interpretation for the existence of cultural revivals in history, and Section 4 offers some concluding

thoughts.

2. The model

2.1. Setup

We describe the model here first before laying it out formally below. We consider a two-period model with two types of

agents: elites and citizens. The productivity of (adult) citizens is a function of their cultural type and public-goods provision

by the elites. The political power of the elites, which determines their capacity to provide their preferred public good, is a

function of the share of the adult citizenry sharing their cultural trait. In the first period (hereafter, period 0), elites choose

the level of public-goods provision. Then, adult citizens produce and socialize their offspring to their cultural type. The

cultural types of the children are then realized. In the second period (hereafter, period 1), parents die and children become

adults and the cultural profile of the latter determines the political power of elites. Elites then choose the level of public-

goods provision in period 1, adult citizens produce, and the game ends. We employ the Subgame Perfect Nash Equilibrium

concept and solve using backward induction.

2.1.1. The citizens

There are a continuum of adult citizens in period zero, each with one child. Citizens belong to one of two cultural types,

which we label type 1 and type 2. The cultural types complement the two types of public goods, { g 1 t , g 2 t } ∈ [0 , 1] , in the production process, where t ∈ { 0 , 1 } represents the period.

The utility of adult citizens of type i ∈ { 1 , 2 } in period t can expressed as: U i (g i t ) = (η + φi ) g i t , (1)

for i ∈ { 1 , 2 } . The utility of adult citizens of type i depends on the provision of public good g i t in two ways. First, adult citizens consume public good g i t and receive a linear utility ηg

i t , with η ∈ [0 , 1 4 ] . 6 Second, adult citizens produce using public

2 For key insights and overviews of recent developments of various aspects of this literature, see Bisin and Verdier (20 0 0b) , Bénabou and Ti-

role (2006) , Guiso et al. (2006) , Nunn (2012) , Spolaore and Wacziarg (2013) , Ticchi et al. (2013) , Algan and Cahuc (2014) , Alesina and Giuliano (2015) ,

Bénabou et al. (2015) , Bisin and Verdier (2017) , and Bisin et al. (2019) . 3 See, among others, Hauk and Mueller (2015) , Verdier and Zenou (2015) , Carvalho et al. (2017) , Prummer and Siedlarek (2017) , Verdier and Zenou (2018) ,

Prummer (2019) , Almagro and Andrés-Cerezo (2020) and Carvalho and Sacks (2021) . 4 For similar models, see Verdier and Zenou (2018) and Almagro and Andrés-Cerezo (2020) . 5 A notable exception is Almagro and Andrés-Cerezo (2020) , who study the rise of national identities. 6 The assumption η ≤ 1

4 ensures that there exists an interior solution for the parent’s optimization problem, as explained below.

3

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

good g i t and get additional utility based on their production. 7 We take production of adult citizen of type i as φi g i t .

8 The

parameter φi ∈ [0 , 1 2

] represents a fixed marginal productivity of public good i for adult citizens of type i . 9

In period t, fraction q t ∈ [0 , 1] of the adult citizens are of type 1 and fraction 1 − q t are of type 2. q 0 is exogenous, but q 1 evolves according to dynamics we describe below. We assume that q 0 ∈ (0 , 1) , so that both cultural groups are initially present.

Cultural Dynamics : The only decision adult citizens make is investment in socialization of their children in period 0.

Following Bisin and Verdier (2001) , we model the transmission of cultural values as a mechanism which interacts intergen-

erational socialization and socialization by society. Intergenerational socialization to type i occurs with probability τ i , the effort of the parent. If direct intergenerational socialization fails, the child receives its cultural trait through horizontal or

oblique transmission (i.e., via peers or other adults such as teachers). This occurs with probability equaling the trait’s share

in the population.

Let P i j denote the probability that the child of a citizen of type i ∈ { 1 , 2 } is socialized to type j ∈ { 1 , 2 } . We can express P i j for any i, j ∈ { 1 , 2 } as:

P 11 = τ 1 + (1 − τ 1 ) q 0 P 12 = (1 − τ 1 )(1 − q 0 ) P 22 = τ 2 + (1 − τ 2 )(1 − q 0 ) P 21 = (1 − τ 2 ) q 0 .

(2)

As an illustration, the probability that a child from a type 1 parent is socialized to type 1 is equal to the sum of the

probability that direct socialization succeeds, τ 1 , and of horizontal transmission by a peer of type 1, (1 − τ 1 ) q 0 . Assuming that transmission effort s are symmetric, we can express q 1 (the fraction of adult citizens of type 1 in period 1) as:

10

q 1 = q 0 + q 0 (1 − q 0 )(τ 1 − τ 2 ) . (3) The Citizens’ Optimization Problem: Parents are forward-looking, and their time preference is set to 1 for simplicity.

We assume that parents have imperfect empathy towards their offspring. This is a form of altruism where parents evaluate

their children’s utility using their own preferences.

Let U i j t

denote the utility of a child of type j in period t, as perceived by a parent of type i, for i, j ∈ { 1 , 2 } . In period 0, children consume the public goods but do not produce. Under the imperfect empathy assumption,

U ii 0 = ηg i 0 and U i j 0 = 0 when j � = i. (4) According to (4) , if the child is socialized to cultural type i, a parent of type i perceives that the child gets utility ηg i

0 given

that the child consumes the public good g i 0

but does not produce. By contrast, if the child is socialized to cultural type j � = i, then a parent of type i perceives that the child gets no utility, given that the child does not consume public good g i

0 . 11

In period 1, children become adults, produce, and consume the public goods. Hence, as perceived by a parent of type i,

the utility of her child is

U ii 1 = (η + φi ) g i 0 and U i j 1 = 0 when j � = i. (5) The inequality U ii

0 + U ii

1 ≥ U i j

0 + U i j

1 = 0 is always satisfied, so parents have incentive to socialize their children to their

own cultural trait.

Let c(τ i ) denote the socialization cost, where τ i is the probability of direct socialization to type i . Since the value of parental socialization is orthogonal to the parent’s own utility represented in (1) , 12 the optimization problem faced by a

parent of type i in period 0 can be written as:

max τ i ∈ [0 , 1]

P ii (U ii 0 + U ii 1 ) + P i j (U i j 0 + U i j 1 ) − c(τ i ) , (6)

7 The assumption of parent’s deriving utility from their production is tantamount to parents earning a piece-rate wage or keeping their after-tax produc-

tion. Since we are not concerned with wages or taxes in the present model, we have chosen for the sake of parsimony to ignore these considerations. For

a theory of cultural evolution that accounts for production and taxation, see, for example, Bisin et al. (2021) . 8 This is a simplified version of a model in which a citizen of type i provides an effort e i t ≥ 0 , and the cost of effort, d(e i t ) , is a function of φi and g i t . For

instance, setting utility such that U i (e i t ) = ηg i t + e i t − (e i t )

2

2 φi g i t would yield indirect utility (at optimization over e i t ) of (

1 2 φi + η) g i t (for g i t > 0 ), which is similar

to (1) . 9 As we explain further below, the assumption φi ≤ 1

2 ensures an interior solution for the parent’s optimization problem.

10 The symmetry assumption of the transmission effort s is a common feature of the studies in the related literature ( Bisin and Verdier, 20 0 0b; 20 0 0a;

20 01; Tabellini, 20 08; Hauk and Mueller, 2015 ). Eq. (3) follows from the fact that there are a proportion (1 − q 0 ) P 21 children of type 2 parents socialized by peers of type 1, and there are a proportion q 0 P

12 children of type 1 parents socialized by peers of type 2. We can therefore write q 1 = q 0 + (1 − q 0 ) P 21 − q 0 P

12 . Substituting P 21 and P 12 from (2) , we derive (3) . 11 The imperfect empathy assumption simplifies the model. Our results are robust to weaker assumptions, as long as the parents derive more utility

from having a child belonging to their own type. For a related theoretical application of the imperfect empathy concept to public good consumption, see

Bisin and Verdier (20 0 0b) . Bisin and Verdier (2011) provide a review of the related literature. 12 Based on the formulation in (6) , we abstract from the parent’s own utility from production in period 0. One could easily incorporate this, however,

by adding a parameter of “altrusim” which would gauge the weight of the child’s utility relative to that of the parent. Doing so would not impact the

qualitative nature of our key results. For a similar specification of the parent’s optimization problem, see, for example, Bisin and Verdier (20 0 0b, 20 01) and

Hauk and Mueller (2015) .

4

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

Fig. 1. Timeline.

with P i j given by (2) , U i j 0

by (4) and U i j 1

by (5) for i, j ∈ { 1 , 2 } , i � = j. We assume that c(τ i ) = 1 2 (τ i ) 2 for simplicity.

2.1.2. The elites

Elites also derive utility from public goods 1 and 2. We therefore denote them as either type 1 or type 2, depending on

whether they derive utility from public good 1 or public good 2. The utility of elites of type i ∈ { 1 , 2 } in period t can be expressed as:

V i (g i t ) = log (φi g i t ) , (7) for t ∈ { 0 , 1 } . The log specification is taken for simplicity and ensures the concavity of the utility function V i (. ) . 13

The primary idea behind (7) is that elites have a vested interest in the provision of a particular type of public good.

For instance, merchants desire protection of property rights as well as transport infrastructure ( North, 1981; Acemoglu and

Robinson, 2012 ), military elites desire spending on defense ( Tilly, 1990; Hoffman, 2015 ), religious authorities advocate for

spending on religious infrastructure and education (possibly to the detriment of spending on secular public education; see

Gill (1998) , Co ̧s gel and Miceli (2009) , Chaudhary and Rubin (2016) , and Rubin (2017) ), and elites in declining industries may

push for subsidies or tariffs to revitalize their industry (e.g., coal mining in the United States).

Public-Goods Provision by the Elites: In each period t, the allocation of two public goods g 1 t and g 2 t is determined

through a political process that involves the two types of elites. We normalize the resources available to the elite to 1 in

both periods, so g 1 t + g 2 t ≤ 1 for t ∈ { 0 , 1 } . In the political process, we assume that the political weights of the elites are monotonic in the population fractions of

the types of the citizens. Put differently, the elites operate under the constraints of the political institutions that give them

power based on the proportion of cultural types that sympathize with them. The weights of the elites therefore reflect the

cultural composition, so provision decisions represent adult citizens’ preferences. 14 In reality, these weights are presumably

influenced by the composition of the citizenry, the inclusiveness of political institutions, the likelihood of social unrest, and

various other dimensions that mediate the effect of the cultural composition of the citizenry on public good provision. 15 We

employ this allocation mechanism to “stack the deck” against frictions or non-representative political power being the root

cause of public good provision and cultural change.

In other words, the political process results in an allocation (g 1 t , g 2 t ) , for t ∈ { 0 , 1 } , that maximizes the weighted dis-

counted utility of the elites under the constraint g 1 t + g 2 t ≤ 1 . In period 1, the allocation mechanism maximizes:

W 1 (g 1 1 , g

2 1 ) = q 1 V 1 (g 1 1 ) + (1 − q 1 ) V 2 (g 2 1 ) . (8)

We denote β ∈ [0 , 1] the time preference of the elites. β is therefore a characteristic of the political process, which in period 0 maximizes the discounted weighted utility of the elites. That is, in period 0, the allocation mechanism maximizes:

W 0 (g 1 0 , g

2 0 ) = q 0 V 1 (g 1 0 ) + (1 − q 0 ) V 2 (g 2 0 ) + β max

g 1 1 ,g 2

1

W 1 (g 1 1 , g

2 1 ) , (9)

given that the constraints g 1 t + g 2 t ≤ 1 are satisfied, for t ∈ { 0 , 1 } , and the elites internalize the dynamics of cultural change (3) .

2.1.3. Timeline and solution concept

The timeline of the model is summarized in Fig. 1 . At the beginning of period 0, children are born. The provision of

the two public goods g 1 0

and g 2 0

in period 0 is then decided via optimization of (9) and adults produce. Adults then choose

their intergenerational socialization effort s τ 1 and τ 2 via optimization of (6) . Socialization then occurs and the types of the

13 It is important that the utility function of the elites is different from that of the citizenry. This ensures that the elites make self-interested decisions

that are not completely aligned with the interests of citizens of their own type. 14 This is a simplified version of a model where the political weights correspond to the prevailing institutions and change so as to keep the political

power of the elites in line with the prevailing cultural composition. For related models of institutional change, see, for example, Bisin and Verdier (2017) ,

Bisin et al. (2021) , and Hiller and Touré (2021) . Additionally, as long as there is a positive relationship between the weights of the elites and the cultural

composition, then the results established in this paper remain robust, as demonstrated in the working paper version Iyigun et al. (2019) . 15 On how social unrest affect policies, see, for instance, Passarelli and Tabellini (2017) and Almagro and Andrés-Cerezo (2020) .

5

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

children are realized. At the beginning of period 1, the children become adults, the parents die, and q 1 is realized. 16 The

provision of the public goods (g 1 1 , g 2

1 ) in period 1 is then decided via the optimization of the allocation mechanism in (8) .

The adults then produce and the game concludes. 17

Our solution concept is Subgame Perfect Nash Equilibrium (SPE). A SPE consists of the optimal provision scheme in both

periods and the intergenerational cultural transmission effort s. Accordingly, the SPE will be denoted { (g 1 ∗ 0

, g 2 ∗ 0

) ; (g 1 ∗ 1

, g 2 ∗ 1

) ; (τ 1 ∗, τ 2 ∗ ) } in the remainder of the paper.

2.2. Solution

We solve the model via backward induction. First, we solve for the provision of public goods in period 1. We proceed to

solve for the socialization efforts of the citizens as well as public good provision in period 0.

2.2.1. Period 1

The public good allocation mechanism chooses (g 1 1 , g 2

1 ) to optimize (8) in period 1 under the constraint g 1

1 + g 2

1 ≤ 1 .

Taking the first-order conditions, we find that:

g 1 ∗1 = q 1 and g 2 ∗1 = 1 − q 1 , (10) with q 1 given by (3) . The optimal provision scheme in period 1 perfectly reflects population shares. An increase in the

fraction of individuals of type 1 implies a commensurate increase in the weight of type 1 elites and the optimal provision

of public good 1.

2.2.2. Period 0

In period 0, the citizens solve (6) for the optimal socialization effort s. We find that

τ 1 ∗ = (1 − q 0 )(ηg 1 0 + (η + φ1 ) g 1 ∗1 ) and τ 2 ∗ = q 0 (ηg 2 0 + (η + φ2 ) g 2 ∗1 ) , (11) with g i ∗

1 given by (10) for i ∈ { 1 , 2 } . Since φi ≤ 1 / 2 , η ≤ 1 / 4 and g i t ≤ 1 , the optimization problem faced by adults always

admits a solution τ i ∗ ∈ [0 , 1] , i ∈ { 1 , 2 } . This is consistent with our probabilistic interpretation of this parameter. Further- more, as in Bisin and Verdier (2001) , there is a substitution between vertical and horizontal socialization mechanisms. All

else equal, when the initial fraction of individuals of type i increases, citizens of type i invest less effort in socializing their

offspring. Likewise, when the initial fraction of individuals of type i increases, the likelihood that their offspring will switch

their cultural affiliation through horizontal socialization decreases.

Using (3) , we establish the following result:

Remark 1. q 1 is the unique solution of the fixed point equation

q 1 = q 0 + q 0 (1 − q 0 )(τ 1 ∗ − τ 2 ∗ ) , (12) with τ i ∗ given by (11) for i ∈ { 1 , 2 } . Proof. See Appendix A.1 . �

Parents’ socialization efforts affect the cultural composition in period 1. However, socialization decisions depend on the

optimal provision of the public goods in period 1, which is a function of the cultural composition in that period. Hence, q 1 solves a fixed point equation. We find in the Appendix that this fixed point equation admits a unique interior solution.

As q 1 solves a fixed point equation, there is a multiplier effect in the socialization decision. All else equal, an increase in

the provision of good 1 in the first period, g 1 0 , leads to higher socialization effort s by citizens of type 1, given (11) . In turn,

adult citizens expect an increase in the fraction of citizens of type 1 in period 1. Therefore, they expect a higher provision

of public good 1 and a lower provision of good 2 in period 1, as g 1 ∗ 1

= q 1 and g 2 ∗1 = 1 − q 1 . Adult citizens of type 1 therefore invest even more in socializing their offspring, and adult citizens of type 2 invest less. Marginal changes in the provision

of the public goods in the initial period can thus have substantial effects on socialization efforts and the evolution of the

weights of the elites.

The optimal allocation of the public goods (g 1 ∗ 0

, g 2 ∗ 0

) maximizes (9) under the constraints g 1 t + g 2 t ≤ 1 for t ∈ { 0 , 1 } , and given the dynamics of cultural change (3) . As demonstrated in the Appendix, the optimal allocation of the public goods in

period 0 is characterized by the following first-order condition:

∂W 0 ∂g 1

0

= q 0 g 1

0

− 1 − q 0 1 − g 1

0

+ β ∂q 1 ∂g 1

0

log

( φ1 q 1

φ2 (1 − q 1 )

) = 0 . (13)

16 We assume that the old generation is entirely replaced by the new one for simplicity, although such an assumption could be relaxed. For example, in

a closely related model of cultural transmission, Hauk and Mueller (2015) assume an overlapping structure where a Poisson birth and death process keeps

the population size constant. 17 There is no commitment issue in this model because the provision of the public goods in the two periods is optimal given the political process.

6

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

Only one first-order condition is sufficient to find the optimal allocation of the public goods in period 0 because the bud-

get constraint g 1 0

+ g 2 0

≤ 1 is necessarily satisfied at equality in equilibrium. The first two terms on the RHS of Eq. (13) de- scribe the trade-off in period 0 between allocating resources to either good 1 or good 2, given the initial weights of the

elites. The third term gives the effect of a marginal increase in the provision of good 1 on the weighted utility of the elites

in period 1. Since ∂q 1 ∂g 1

0

> 0 , an increase in the provision of public good 1 in period 0 shifts the citizens’ socialization deci-

sions, and it affects the cultural composition in period 1 and the period 1 weights of the elites. The elites thus internalize

the effect of the initial provision of the public goods on the future cultural composition.

2.2.3. Characterization of subgame perfect equilibria

We establish the following result in the Appendix:

Proposition 1. There exists a threshold ˜ β > 0 such that:

• If β ≤ ˜ β, there exists a unique stable SPE. • If β > ˜ β, there exists two thresholds q and q in (0,1) with q < q such that:

• If q 0 ∈ [ q , q ] , there exists two stable SPE. • If q 0 / ∈ [ q , q ] , there exists a unique stable SPE.

Proof. See Appendix A.2 . �

The intuition behind Proposition 1 is that there can be increasing marginal returns to provisioning public goods. Incre-

mental changes in the initial provision of public goods can lead to increasingly higher future utility levels for the elites by

increasing their weights in provision decisions and shifting the cultural composition. This non-convexity can formally be ob-

served in the first-order condition given in (13) . The marginal benefit of increasing the provision of good 1 in period 0 is pro-

portional to the period-1 relative utility of the elites of type 1, which necessarily increases with g 1 0 , as ∂

∂g 1 0

log [ φ1 q 1

φ2 (1 −q 1 ) ] > 0 .

Since the non-convexity arises from the inter-temporal concerns of the elites, its magnitude is related to the time pref-

erences of the elites. When β is lower than the threshold ˜ β, the elites’ concern for the future is minimal. It follows that the non-convexity does not meaningfully affect the period-0 optimization problem, so there is a unique stable SPE.

By contrast, when the time preferences of the elites are such that β > ˜ β, then the elites care enough about the future that the non-convexity substantially affects their period-0 decision problem. When the initial fraction of individuals of type

1, q 0 , has intermediate values, there are two solutions to the optimization problem faced by the elites in period 0. This is

because the multiplier effect in socialization decisions is strong, so it is conceivable that either type of elite could increase

its future weight in the provision decision of period 1 if enough individuals adopt their cultural type. The optimization

problem faced by the elites in period 0 therefore admits two stable solutions. In both cases, the elites are able to affect

citizens’ socialization decisions in order to shift the cultural composition in their favor. Alternatively, when one cultural type

has a clear initial majority (i.e., q 0 is above q or below q ), then the extent of cultural change is limited. In such cases, it is

too costly for the elites from the minority group to affect the trajectory of culture in order to increase their future weight in

provision decisions. Consequently, the optimization problem faced by the elites in period 0 admits a single stable solution.

2.3. Cultural revivals

We now extend our model to account for the type of cultural revivals highlighted in the introduction. In the case of the

postbellum South, white elites faced a serious threat to their political and economic power following emancipation. Poor

whites and freedmen could have joined forces to improve their economic and political power, both of which were previ-

ously almost non-existent. In order to prevent this from happening, white elites funneled resources into public goods used

by whites (e.g., white-only schools), which increased the returns to a racist cultural ideology. Ultimately, the equilibrium

that was reached was one in which racist policies and cultural ideologies reinforced each other, and poor whites largely

aligned with the wealthy elite. Importantly, this outcome arose in spite of the fact that a political and economic alliance

with African-Americans would have likely improved labor market outcomes for poor whites (see Section 3.1 ). Likewise, the

nascent Turkish Republic’s strictly secular reforms and principles were not enough to block conservative Islamists from re-

asserting their influence in Turkish society. With a succession of elections starting in 1995, Islamists in Turkey were able to

firmly consolidate their political power. This was a manifestation of deep-rooted cultural change spearheaded by investment

in public goods (see Section 3.2 ).

One commonality unites these examples. Cultural change induced political change despite these changes entailing a

movement away from what might have been more efficient outcomes.

How do we relate these insights to the model? In particular, what do we mean by “more efficient” outcomes? We address

this issue by introducing the notion of dynamic production efficiency (DPE). Accordingly, we consider a public good allocation

to be dynamic production efficient if, given the initial cultural profile of society ( q 0 ), the period-0 allocation maximizes

production across both periods and the period-1 allocation maximizes production in that period. This conceptualization is

important for our understanding of cultural revivals, because we are seeking to understand the conditions under which

cultural change triggers political change despite these changes being (dynamically) production inefficient.

7

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

We operationalize these insights in the context of our model by first formalizing dynamic production efficiency. To that

end, we can express the aggregate production of the citizens in period t ∈ { 0 , 1 } as follows: p t (g

1 t , g

2 t ) = q t φ1 g 1 t + (1 − q t ) φ2 g 2 t . (14)

Denoting the DPE allocation of public goods as { (g 1 ,DPE 0

, g 2 ,DPE 0

) , (g 1 ,DPE 1

, g 2 ,DPE 1

) } , we can now formally introduce the concept of Dynamic Production Efficiency (DPE):

Definition 1. The production of public goods { (g 1 0 , g 2

0 ) , (g 1

1 , g 2

1 ) } is dynamically production efficient when:

(g 1 ,DPE 1

, g 2 ,DPE 1

) = arg max (g 1 1 , g 2

1 ) p 1 (g

1 1 , g

2 1 ) (15)

under the constraint g 1 1

+ g 2 1

≤ 1 , and (g 1 ,DPE

0 , g 2 ,DPE

0 ) = arg max (g 1

0 , g 2

0 ) p 0 (g

1 0 , g

2 0 ) + βp 1 (g 1 , DPE 1 , g

2 , DPE 1

) , (16)

given that g 1 0

+ g 2 0

≤ 1 , the dynamics of cultural change (12) are internalized, and (g 1 ,DPE 1

, g 2 ,DPE 1

) maximizes production in

period 1.

Hence, if the elites were to maximize the citizens’ production, they would choose a provision scheme

{ (g 1 ,DPE 0

, g 2 ,DPE 0

) , (g 1 ,DPE 1

, g 2 ,DPE 1

) } . We can now characterize dynamic efficient production as follows: Proposition 2. There exists a threshold β

DPE > 0 and a threshold ˜ q DPE ∈ [0 , 1] such that if β > βDPE , there is a unique efficient

dynamic production path such that

• if q 0 ≥ q DPE , it is dynamically efficient to produce good 1, g 1 ,DPE

0 = g 1 ,DPE

1 = 1 and g 2 ,DPE

0 = g 2 ,DPE

1 = 0 . (17)

• if q 0 < q DPE , it is dynamically efficient to produce good 2,

g 1 ,DPE 0

= g 1 ,DPE 1

= 0 and g 2 ,DPE 0

= g 2 ,DPE 1

= 1 . (18) • q DPE is non-decreasing in β when φ2 > φ1 , and non-increasing in β otherwise.

Proof. See Appendix A.3 . �

From Proposition 2 , which good will be produced along the dynamic efficient path depends on the initial cultural com-

position. If type 1 adults are initially sufficiently numerous (i.e. q 0 > ˜ q DPE ), then only good 1 will be the efficient one to

produce in the two periods. Conversely, when adults of type 1 are not initially numerous (i.e., q 0 < ˜ q DPE ), then only good 2

will be the efficient one to produce. 18

While we derive the full characterization of the dynamic efficient production in the Appendix, we restrict our attention

in Proposition 2 and in the rest of the paper to the parameter values such that β > β DPE

. 19 By doing so, we abstract from

the less interesting cases where production in period 0 and the ensuing evolution of cultural norms do not affect efficient

production in period 1. When β > β DPE

, the dynamic efficient production is path dependent . When only good 1 is produced

in period 0, then the fraction of individuals of type 1 increases sufficiently between the two periods so that producing

good 1 remains efficient in period 1. Conversely, if only good 2 is produced initially, then the fraction of type-2 individuals

increases and it remains optimal to produce good 2 in period 1. Intuitively, when the time preferences are sufficiently large,

the evolution of cultural norms and the period-1 allocation substantially affect the optimization problem in period 0.

With the definition of DPE and Proposition 2 in hand, we can now turn to cultural revivals. As we noted before, cultural

revivals have two features: i) given the initial cultural composition, it is dynamically efficient to produce one public good; ii)

the cultural type and the elites associated with the productionally inefficient sector becomes predominant. The first condition

entails that the share in the population of one type and the political weight of the corresponding elite are initially low

enough that it is dynamically inefficient to produce the public good favored by that elite. This is important, because we are

not interested in the case in which cultural and political change is either efficient or driven by vested interests.

Moreover, we are not interested in the case in which the cultural and political profile moves in the direction of the

type with higher marginal productivity. While our model can account for such movements, there are other explanations for

such an outcome that we cannot rule out. These include external influence (i.e., a country will fall behind if it does not

adopt the cutting-edge production method) and externalities (i.e., adopting a more productive ‘type’ gives a society access

to other, unforeseeable windfalls). We do not deny the importance of such influences. However, this is not the phenomenon

we are interested in, nor is it salient for our motivating examples. Rather, we are interested in the case in which the sector

with lower marginal productivity becomes more predominant over time. For this reason, for the remainder of the paper we

assume without loss of generality that φ1 < φ2 . We therefore focus on cultural revivals favoring type 1, since economic activities performed by this type have lower marginal productivity.

18 In the proof of Proposition 2 , we assume that when ˜ q DPE = q 0 , the default option is to produce good 2. 19 A formal characterization of β

DPE is provided in the proof of Proposition 2 .

8

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

Fig. 2. Period-0 Allocation in a SPE with and without a Cultural Revival Note: the red line represents the DPE path and the blue line represents equilibrium

production.

We can therefore define cultural revivals in the context of our model as follows:

Definition 2. A cultural revival in favor of type 1 occurs in a SPE when the following conditions are satisfied:

i) q 0 < ˜ q DPE , and

ii) q 1 > q 0 .

The first condition above, q 0 < ˜ q DPE , implies (via Proposition 2 ) that it is dynamically efficient to produce good 2 in both

periods of the game. However, the second condition, q 1 > q 0 , implies that despite type 2 individuals being more marginally

productive (since φ1 < φ2 ) and public good provision associated with them being more production efficient in the two periods, the fraction of type 1 individuals increases between periods. Furthermore, the weight of the elites of type 1 in

provision decisions increases in a cultural revival, while it would have decreased if the provision of the public good was

close to the DPE.

The effect of a cultural revival on the equilibrium of the game is illustrated in Fig. 2 . The red line represents the dy-

namically efficient production path. Since q 0 < ˜ q DPE , it is dynamically efficient to provision good 2 in the two periods of the

game (i.e., g 1 ∗ 0

= g 1 ∗ 1

= 0 , by Proposition 2 ). The left panel illustrates a typical case where there is no revival in the SPE. The period-0 allocation (g 1 ∗

0 , 1 − g 1 ∗

0 ) is close to the efficient provision in period zero (g 1 ,DPE

0 = 0 , g 2 ,DPE

0 = 1) , and the share of

citizens of type 1 decreases over time. The equilibrium production represented by the blue line converges towards the dy-

namic efficient path. The right panel of the figure illustrates an equilibrium in which there is a cultural revival. The period-0

allocation (g 1 ∗ 0

, 1 − g 1 ∗ 0

) is far from the dynamic efficient production. The high provision of good 1 in period 0 triggers sig-

nificant cultural changes and the fraction of citizens of type 1, q 1 , increases. Since g 1 ∗ 1

= q 1 , equilibrium production diverges from the dynamic efficient production path.

Given our characterization of the SPE of this model, we are thus able to derive sufficient conditions under which cultural

revivals emerge in at least one SPE:

Proposition 3 Cultural Revivals. Assuming that β > β DPE

, there exists a threshold β and a threshold ˜ q 0 < ˜ q DPE such that there

is a cultural revival favoring type 1 in at least one SPE if ˜ q 0 < q 0 < ˜ q DPE and β > β.

Proof. See Appendix A.4 . �

Although the result of Proposition 3 is related to cultural revivals favoring type 1, by symmetry, a similar result can

be established for cultural revivals favoring type 2 for φ1 > φ2 . The assumption that β > β DPE

ensures that the dynamic

efficient production is characterized by Proposition 2 .

The intuition associated with Proposition 3 is that when φ1 < φ2 , elites of type 1 face a marginal productivity disad- vantage that threatens their future weight in provision decisions. Hence, type 1 elites receive a particularly high marginal

benefit, relative to type 2 elites, from citizens’ socialization decisions when good 1 is over-provided. Formally, ∂ 2 q 1

∂ φ2 ∂ g 1 0

> 0 ,

meaning that when φ2 is large, so is the marginal (positive) effect of provisioning good 1 on the fraction of citizens of type

1, ∂q 1 ∂g 1

0

.

We find that type 1 elites can still thrive over time, despite good 1 being dynamically inefficient to produce (i.e. q 0 <

˜ q DPE ) and good 1 yielding lower marginal production (i.e. φ2 > φ1 ). In order for this to happen, two conditions must be met. First, there must be a sufficiently large fraction of individuals of type 1, (i.e. q > ˜ q ). It must be conceivable for the

0 0

9

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

elites of type 1 to increase their future weight in provision decisions by affecting socialization decisions. This necessitates a

sufficiently high population of type 1 individuals. Second, q 0 must be less than ˜ q DPE from the definition of cultural revivals.

Finally, the time preference parameter β must be sufficiently large (i.e. β > β ) . If β is too small, the elites have a limited effect on the evolution of the cultural composition. The evolution of the cultural composition is rather driven by citizens’

socialization decisions, which reflect the economic conditions of their children. Hence, if good 1 is less efficient to produce

than good 2, the citizens of type 2 tend to invest higher socialization effort s than their peers of type 1. The fraction of

citizens of type 2 increases, and the elites provisioning good 2 thrive. Conversely, if β > β, the elites of type 1 care enough about the future to shift the cultural composition in their favor and can set in motion a cultural revival.

To summarize, despite the fact that elites cannot affect structural differences by force or directly alter politics to their

benefit, they can still leverage the resources at their disposal to change the prevailing cultural norms and, ultimately, their

future political power. We have demonstrated that even when providing public goods to a cultural type is inefficient and is

associated with a productivity disadvantage, cultural and political change favoring that type is still possible. By influencing

the course of cultural change, self-interested elites that provision inefficient public goods might still be able to thrive.

3. Historical evidence of cultural revivals

3.1. “Poor Whites” and Jim Crow in the Postbellum South

Historians of the postbellum South have long been fascinated by the acquiescence of poor whites after the Civil War

to racist policies that limited the rights of blacks and excluded them from many basic services ( Frazier, 1949; Woodward,

1974 ). The large literature on this topic tends to frame this as a puzzle. While poor whites did face some degree of economic

competition from blacks, the potential gains from cooperation—both in labor relations and at the ballot box—seemed to have

been much greater. Indeed, for a brief period, many poor whites joined black men in the Republican Party. For instance, in

North Carolina, a “biracial coalition of freedmen and disaffected lower-class whites, resentful of planter domination, chan-

neled their frustrations into politics, ushering into state and local offices Republican administrations of a reformist bent,

pursuing measures calculated to end aristocratic privilege and forge a more democratic society” ( Forret, 2006 , p. 229). Some

of these reforms were led by recently-freed slaves who were elected to office during Reconstruction.

Poor whites were poor , and they faced similar class and employment relations with wealthy whites as did blacks. An

alliance between the two groups would have allowed them to dominate Southern politics and ultimately receive the as-

sociated economic benefits. Yet, such alliances, where they existed, did not last. The white economic elite recognized the

potential threat to their political power, and they successfully prevented the alliance from happening. In the end, poor

whites tended to align with the rich white elite on political and social issues. How can this be explained, given that (as a

class) this alliance was to the economic detriment of poor whites?

The dominant theory in the literature focusing on economic issues stresses the role of economic competition between

blacks and poor whites. With the freeing of slaves, blacks and poor whites were now in competition for the same jobs.

Hence, racist laws were favored by poor whites because it limited economic competition ( Marshall, 1961; Wilson, 1976 ). Yet,

a purely economic explanation has a difficult time explaining why poor whites continued to align with the same elites that

kept them in such a subjugated economic position. Alternatives did exist. Every southern state had some experience with

interracial coalitions in the decades following the Civil War ( Forret, 2006 ). In the last two decades of the 19th century, the

Populist party attempted to bring together poor whites and blacks. Yet, every time such a coalition was attempted it failed.

Race, not economic relations, tended to draw poor whites back to the Democratic Party fold: “flagrant race-baiting, playing

to whites’ racial fears and anxieties, cemented loyalty to the party of white supremacy . . . When poor white voters aided in

the restoration of Democratic governments, they removed from power the very politicians most sympathetic to their plight”

( Forret, 2006 , p. 231).

Theories focusing primarily on the role of economic competition are likely correct in many respects. It is not our inten-

tion to undermine the importance of such economic factors. Yet, these theories have shortcomings. Our theory of cultural

revivals helps address these shortcomings. We can think of there being two types of elites in the postbellum South: the old

planter elite and Republican or Populist mobilizers. Both offered the possibility of providing goods via the political process.

Under Jim Crow, the planter elite offered goods that favored poor whites (white schools, white churches, white hospitals,

white drinking fountains, and many “public” goods discriminated by race), while the Populists promised policies that would

improve the plight of both the poorest whites and blacks. Such policies were briefly enacted during Reconstruction, when

many recently-freed slaves were elected to office. Logan (2020) finds that counties with more black officials had greater tax

revenue, which was spent on improving literacy (for both black and white children) and land redistribution. These were

clearly policies favoring poor Southerners.

From the perspective of poor whites, the marginal productivity of public goods ( φ2 ) complementary to Populist politics was almost certainly greater than that of public goods complementary to the planter elite ( φ1 ). In the context of our model, Populist cultural values would have been those conducive to an alliance between poor whites and blacks, whereas the cul-

tural values of the planter elite were racist and meant to undermine any such alliance. Given that poor whites and blacks

made up a vast majority of the Southern population, it was almost certain that the provision of public goods complementary

to Populist politics was dynamically production efficient (to employ the terminology of our model). Hence, multiple equilib-

10

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

ria were possible. One (non-DPE) equilibrium consisted of the old white planter elite dominating the political process, while

the other (DPE) equilibrium consisted of poor whites and blacks gaining a greater political voice.

Clearly, the cultural values of the planter elite won out, and this was manifested in Jim Crow laws. Our model helps

explain this outcome. The planter elite, facing the prospect of losing their political and economic power following the Civil

War, could not simply alter political institutions to their benefit, especially during Reconstruction. Widespread suffrage—

for males, at least—meant that freed blacks and poor whites could unite to upend their political dominance. This outcome

would have been dynamically production efficient. How could the white elites prevent this outcome from happening? That

is, how could they trigger a cultural revival? The logic of our model, as formalized in Proposition 3 , indicates one possible

solution: overinvestment in public goods favoring poor whites, which in turn triggered a change in cultural beliefs conducive

to their desired outcomes.

These were precisely the actions taken by white southern elites. It was a departure from the antebellum period, in

which the relationship between the planter elite and poor whites was much more antagonistic. As recently explained by

Merritt (2017) , the planter elite limited educational opportunities for poor whites prior to the Civil War, because keeping

them ignorant was best for limiting social unrest. More generally, poor whites in the antebellum South were marginalized.

Yet, after the Civil War, and especially after Reconstruction, “formerly marginalized poor whites were welcomed into a fuller

participation in the benefits of whiteness . . . by the end of the nineteenth century, antebellum cooperation between slaves

and poor whites was forgotten, replaced by the reality of racial hatred and Jim Crow segregation” ( Forret, 2006 , p. 228,

231). As explained by Feldman (2004 , p. 164), “race repeatedly exerted pressure on poor whites to ally themselves with

their privileged white ‘betters’ against their own class interests and potential biracial alliance, and race produced a series

of reforms that largely made life better for whites and worse or no better for blacks.” Some of these benefits, which took

the form of racially segregated public goods, are well-known. Segregated schools funneled resources to white communities—

poor or not—at the expense of black communities. These are a prototypical example of public goods that are complementary

to a certain cultural identity. Numerous other types of public goods complementary to a racist, white identity were provided

by elites. One important example is parades and statues commemorating the Confederacy. This was the period when a new

narrative around the “struggles of the Confederacy” emerged. In Alabama, for instance, this period was characterized by

“attempts to memorialize the Lost Cause, disparage Reconstruction, glorify Redemption, romanticize the Reconstruction Klan,

and paint a dark picture of [Reconstruction] as a tragic time of black rule, Yankee pillage, federal repression, corruption, and

chaos” ( Feldman, 2004 , p. 165). Memorials and public events were a common means of forging such values. In the terms of

our model, this narrative was part of a larger “cultural revival” that increased the cultural imprint of white identity, which

in turn allowed for the codification of racist laws that primarily benefited the white elite.

This glorification of the past was clearly intended to establish and cement a white cultural identity. Another mechanism

through which the old white elite attempted to affect culture was propaganda. Ottinger and Winkler (2020) find that the

emergence of the Populist party caused a rise in anti-Black propaganda in the media (i.e., the word “rape” in co-occurrence

with the word “negro”). It seemingly worked. The Populist-desired alliance between poor whites and blacks never came to

fruition, and the white elite were largely able to co-opt the former and suppress the latter in the century following the Civil

War.

In short, the rise of a white supremacist culture among many poor southern whites in the decades following the Civil

War was part of a broader “cultural revival.” This is precisely what Proposition 3 predicts can happen when established

elites face a threat to their political power. An elite-driven “overinvestment” in goods complementary to a white supremacist

cultural identity helped the elite maintain their political power in the face of an alternative that promised greater potential

returns to the masses of blacks and poor whites. Yet, most poor whites never gained the cultural capital to take advantage

of this alternative. This would have required a cultural change in which an alliance with working class blacks would have

been desirable. This cultural revival had numerous long-run, negative consequences, many of which are still with us today.

While it is not within the scope of this paper to explore these long-run consequences, they are indicative of just how hard

it is to escape from an equilibrium in which culture and political power reinforce each other.

3.2. The Gülen movement in Turkey

The Turkish Republic was founded in 1923 after the Ottoman Empire collapsed following a longer than six century tenure.

The modern republic was founded on the back of mainly top-down sociopolitical and economic reforms that were primarily

inspired by Western Enlightenment principles and strictly secular social and political norms. The main impetus for these

reforms was provided by the fact that the returns to secular schooling and human capital had markedly risen following

the Industrial Revolution. Meanwhile, the economic productivity of the more established but conservative Ottoman culture

had long been stagnant. As Kuran (2011) documents, this was a new reality that had been borne out by the upturn in

economic fortunes of the better educated, non-Muslim citizens of the crumbling empire. These reforms were implemented

fairly swiftly in a country whose population was more than 95 percent Muslim and most of them highly devout.

Religious groups have been a constituent element of Turkey’s Ottoman legacy. They managed to persevere even the Ot-

toman modernization campaigns in the mid- to late-19th century, collectively known as the Tanzimat Era. Nevertheless, they

were disbanded and outlawed by Kemal Ataturk in the early years of the Turkish Republic. The outlawing of religious edu-

cation and the introduction of the Latin alphabet in 1927 further limited their influence and forced them to go underground

( Tee, 2016; Bozça ̆ga and Christia, 2020 ).

11

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

Fig. 3. Gülen Schools and Businesses Source: Bozça ̆ga and Christia (2020) .

Throughout its first seven decades of existence, the Turkish Republic adhered to a French-style “laicité” whereby the

state hierarchy was fully under the secular elites’ control and all public goods and services—most notably, all three levels

of education—were established and guided by secular norms. The upshot is that the demise of the Ottoman Empire and

the fledgling new republic very clearly and swiftly upended the well-rooted Ottoman hierarchy and elite who derived their

political legitimacy from Islam and the Muslim clerics ( Rubin, 2017 ).

By the early 1990s, however, the Turkish seculars’ grip on political power and institutions began to wane. With a suc-

cession of elections starting in 1995, Turkish Islamists were able to firmly regain and consolidate their political power. This

consolidation was evident in the wake of the 2018 Presidential elections, in which the century old Turkish Parliament dis-

banded and unprecedented powers were bestowed upon the president-elect Tayyip Erdo ̆gan, who had been the Turkish

Prime Minister since late 2002.

How was this complete reversal achieved? The model we outlined above provides some insight. In the parlance of the

model, two types of elites existed in Turkish politics: secularists and Islamists. Even though the latter were forced under-

ground for decades after the fall of the Ottoman Empire, as Proposition 3 dictates, the Islamist ideology had enough support

and adherents in Turkey so as to enable an eventual revival. Moreover, and in line with our theory, such a reversal came on

the back of deep-rooted cultural change spearheaded by investment in public goods.

The seeds of this transformation were planted in the 1950s by a religious revival—the Gülen Movement—whose primary

emphasis lay in public goods provision in the form of primary and secondary schooling. The foundations of such a cultural

revival were sown at the end of Turkey’s single party era in 1950, when Islamist groups ratcheted up their social and

political activism. At the forefront of this movement was Fethullah Gülen, a religious cleric in the Western coastal city of

Izmir who mainly focused on establishing K-12 schools. Gülen purported to preach an inclusive brand of Sunni Islam that

emphasized cooperation and tolerance, and he viewed Western capitalism and economic modernity as generally compatible

with Islam ( Matthews, 2020 ).

The Gülen Movement stressed charity and public goods provision to the lower and middle-income classes, but by far

the most important element in that drive was investment in public schools. As Bozça ̆ga and Christia (2020) note, “Hizmet’s

primary emphasis was on education services and, similar to many Islamist movements that have viewed the school system

as a way to yield control over the hearts and minds of students... Gülenists used educational institutions as a way to spread

their ideas, win over youth, and strengthen the movement.”

In the half century between the mid-1960s and 2016 but especially after the 1970s, the growth in Gülenist “Hizmet”

schools and other educational institutions was quite remarkable. Based on Bozça ̆ga and Christia ’s data, while there were 7

Hizmet schools between 1965 and 1981, their numbers grew to 28 by the close of the latter decade; to 292 by the end of

the 1990s; to 524 by the end of the aughts; and to 960 by 2016 (see Fig. 3 ). In fact, “the proportion of Gülenist educational

institutions as compared to all private ones varied from about 5 percent for tutoring centers, to 11 percent for schools and

12

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

18 percent for dorms” ( Bozça ̆ga and Christia, 2020 ). More importantly, the growth of the Hizmet movement represented

only a subset of the expanding weight and social influence of Islamism in Turkey.

Based on our model, the number of those who were sympathetic to the Hizmet movement ought to have increased more

rapidly in Turkey after the 1990s, and their associated Islamist culture should have become more prominent and influential

over time. Moreover, politics should have subsequently evolved in ways that were more amenable to the Islamists. This is

precisely what happened starting in the late 1990s and the early 21st century. It culminated in a permanent shift in the

balance of power from the Kemalist seculars to Islamists. This is the mechanism through which Proposition 3 predicts a

cultural revival will occur (if it occurs at all).

According to Bozça ̆ga and Christia (2020) , the proportion of Gülen-affiliated officials across different civil service sectors

ranged between about 1.5% in healthcare to roughly 5% in the judiciary and 11.3% in the police. These estimates are indicative

of the extent of the penetration of the Gülenist movement among the high-ranking officials within in the government

bureaucracy, judiciary and the police. Relatedly, and as depicted in Fig. 3 , the number of Gülen affiliated businesses in Turkey

began to rapidly increase starting in the 1990s, following more than a decade lag in the ascension of Gülenist schools.

The Gülenist Hizmet movement was part of a broader Islamist revival in Turkey. In fact, while the Islamists remain

in power and their political control is further entrenched through institutional interventions, Gülenist political influence

and control came to an abrupt end following the failed military coup attempt in the summer of 2016. The Islamist Tayyip

Erdo ̆gan government, which was in a tight and decades-old alliance with the Gülenists in their collective power struggle

against the Kemalist seculars, ascribed the failed coup attempt to Fethullah Gülen and his followers and it began a sweeping

purge of Gülenists from all levels of governmental, educational, and economic hierarchies which continues to this day.

In sum, the birth, spread and growth of the Gülenist Hizmet movement and Islamism in Turkey is a historical example

which supports our model. It was achieved almost exclusively on the back of a focused emphasis on educational supply,

followed by a subsequent and unambiguous political transformation of the country. This shift in the balance of political

power came about only after the cultural dynamics of the country were altered in ways that slowly but steadily favored the

Islamists.

3.3. Other examples of cultural revivals

In this section, we briefly provide more examples of cultural revivals upon which our model provides insight.

Squicciarini (2020) provides a prototypical example. She finds that the Catholic Church responded to the second wave of

industrialization in the 19th century by imposing an anti-scientific curriculum in Catholic schools, which harmed the eco-

nomic outcomes of students in highly Catholic regions of France. Much as in our model, one type of elites (the Church)

altered cultural norms by investing in public goods (schools). This shifted the institutional and cultural paths against the

headwinds of modernization in highly Catholic regions of France. 20

Cultural revivals need not be associated solely with religious elites, however. Iyigun and Rubin (2017) study macro-level

cultural revivals in the 17th-century Ottoman Empire, 19th-century Imperial China, and 18th–19th century Tokugawa Japan.

Only in the first of these cases were religious elites important in facilitating the cultural revival. In each of these cases, rulers

and elites were confronted with Western institutions and technologies that had the potential to upend the economic and

social order. In the context of our model, the old political, military, and economic elite had cultural values complementary

to the production of “traditional” goods, such as tımars or waqf in the Ottoman Empire or Confucian education in Imperial

China and Tokugawa Japan. Meanwhile, certain types of merchants, producers, and others with access to capital but not

social prestige or political power had values consistent with more “non-traditional” goods. This latter group would have seen

their returns rise immensely with the adoption of Western technologies, education, and modes of production. Such adoption

would have almost certainly been dynamically production efficient. Yet, in each of these cases , the reaction to the West was

what we call a “cultural revival”: cultural values favoring the established elites became more predominant in society, and

modes of production suited for the pre-industrial world became further entrenched. In many ways, these cultural revivals

mimic the revival of white supremacist culture in the postbellum South discussed in greater detail in Section 3.1 . In all of

these cases, modes of production ended up supporting the interests of the established elites all the more.

4. Conclusion

There have been many historical cases in which social groups vested in inefficient production modes and technologies

were able to maintain their social influence and political dominance over time despite having no capacity to block modes

of production detrimental to their interests. This paper develops a theory to explain this phenomenon. We propose that

when elites have limited power to directly block economic activities detrimental to their interests, they can instead affect a

20 There are many other relevant historical examples. For instance, Chaney (2016) studies the decline of Islamic science, finding that the decline be-

gan in the 12th and 13th centuries and that scientific learning was replaced by more traditional modes of religious education in madrasas. Carvalho and

Koyama (2016) and Carvalho et al. (2017) find that ultra-Orthodox European Jews responded to emancipation in the 19th century by imposing unprece-

dented restrictions on secular education, further closing themselves off from society. Fouka (2020) provides an example of a cultural backlash from post-

WWI US education policy. A prohibition of German in public schools, which was intended to promote assimilation, had the effect of heightening cultural

identity among Germans.

13

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

society’s culture . We call such outcomes “cultural revivals.” The primary intuition for the existence of cultural revivals is that

when an economic disadvantage threatens the future political weight of one type of elites, these elites have particularly high

incentive to affect citizens’ socialization decisions by provisioning public goods. If they are successful, they increase their

political weight and, as a result, economic activities complementary to their interests also increase. This happens despite the

fact that these economic activities are associated with less productively efficient outcomes.

The insights provided by the model offer an explanation for two historical case studies: the Jim Crow South and Turkey’s

Gülen Movement. In both of these cases, one group of elites—whose economic power and political influence were threatened

by new economic realities—could not prevent institutional changes by simply altering political institutions. Yet, they were

still successful in preventing changes that would have undermined their power. They did so by altering society’s cultural

composition via the provision of public goods. In both cases, cultural changes favoring more “traditional” values became

predominant. This in turn allowed the prevailing elites to strengthen their grip on political and economic power.

Such a series of events has been shown time and again to be a potent means for established elites to maintain their

power in the face of social and economic headwinds pushing to undermine their power. Our theory highlights why elites so

often succeed in pushing against these headwinds even when they cannot directly alter political and economic institutions

to their benefit.

Appendix A. Proofs

A1. Proof of Remark 1

Plugging in (11) into (12) , we get:

q 1 = q 0 + q 0 (1 − q 0 ) { (1 − q 0 ) { ηg 1 0 + (η + φ1 ) q 1 } − q 0 { ηg 2 0 + 1 − (η + φ2 )(1 − q 1 ) }} (A.1) The solution q 1 (g

1 0 , g 2

0 ) is unique, as represented in Fig. A .1 . The RHS of (A .1) (the thick black line in Fig. A.1 ) belongs to

(0,1) when q 0 ∈ (0 , 1) ; it is linearly increasing in q 1 with a slope below 1. 21 Hence, the RHS of (A.1) only crosses the 45 ◦ line once. The solution q 1 (g

1 0 , g 2

0 ) is necessarily stable: if q 1 > q 1 (g

1 0 , g 2

0 ) , then the socialization effort s are such that q 1 is too

high to be an equilibrium. The inverse is true if q 1 < q 1 (g 1 0 , g 2

0 ) . We find that

q 1 (g 1 0 , g

2 0 ) =

q 0 (1 − q 0 (1 − q 0 )(η + φ2 )) + ηq 0 (1 − q 0 ) { (1 − q 0 ) g 1 0 − q 0 g 2 0 } 1 − q 0 ( 1 − q 0 )( η + φ)

, (A.2)

with φ = (1 − q 0 ) φ1 + q 0 φ2 .

A2. Proof of Proposition 1

The optimal allocation in period 0, (g 1 0 , g 2

0 ) , solves:

max (g 1

0 ,g 2

0 ) W 0 (g

1 0 , g

2 0 ) = q 0 V 1 (g 1 0 ) + (1 − q 0 ) V 2 (g 2 0 ) + βW 1 (g 1 ∗1 , g 2 ∗1 ) , (A.3)

with W 1 given by (8) under the following constraints: ⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩

g 1 0

+ g 2 0

≤ 1 , g 1 ∗

1 = q 1 ,

g 2 ∗ 1

= 1 − q 1 , and

q 1 = q 0 (1 − q 0 (1 − q 0 )(η + φ2 )) + ηq 0 (1 − q 0 ) { (1 − q 0 ) g 1 0 − q 0 g 2 0 }

1 − q 0 (1 − q 0 )(η + φ) ,

(A.4)

with φ = (1 − q 0 ) φ1 + q 0 φ2 . As the elites always benefit at the margin from more spending on their most preferred good, the budget constraint is

necessarily satisfied at equality, g 1 0

+ g 2 0

= 1 . Hence, only one first-order condition can be written for public good g 1 0 , while

we substitute g 2 0

with 1 − g 1 0 . We denote g 1

0 = g, g 2

0 = 1 − g, q 1 (g 1 0 , g 2 0 ) = q 1 (g) and W 0 (g 1 0 , g 2 0 ) = W 0 (g) in the rest of the proof

to simplify the notation.

Substituting g 2 0

by 1 − g and maximizing W 0 (g) , we find the following FOC:

∂W 0 (g)

∂g = q 0

g − 1 − q 0

1 − g + β ηq 0 (1 − q 0 )

1 − q 0 (1 − q 0 )(η + φ) log

( φ1 g 1 ∗

1

φ2 g 2 ∗ 1

) + ∂g

1 ∗ 1

∂g

∂W 0 (g)

∂g 1 ∗ 1

+ ∂g 2 ∗ 1

∂g

∂W 0 (g)

∂g 2 ∗ 1

= 0 . (A.5)

21 Formally, if we denote Z(q 1 ) the RHS of (A.1) , we find that Z ′ (q 1 ) = q 0 (1 − q 0 ) ηφ < 1 , with φ = (1 − q 0 ) φ1 + q 0 φ2 . If q 1 = 0 , Z(0) = q 0 { 1 + (1 −

q 0 )[(1 − q 0 ) ηg 1 0 − q 0 ηg 2 0 − q 0 (η + φ2 )] } . A g 2 0 = 1 − g 1 0 , Z(0) ≥ q 0 { 1 − (1 − q 0 ) q 0 [2 η + φ2 ] } > 0 for any φ1 , φ2 ∈ [0 , 1 / 2] and η ∈ [0 , 1 / 4] . By symmetry, Z(1) < 1 .

14

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

Fig. A.1. Determination of q 1 (g 1 0 , g 2

0 ) Note: the thick black line represents the RHS of (A.1) .

In the previous FOC, and given our equilibrium concept, the envelope theorem can be applied. In period 0, the elites

internalize their optimal choice of period 1, so

∂g 1 ∗ 1

∂g

∂W 0 (g)

∂g 1 ∗ 1

+ ∂g 2 ∗ 1

∂g

∂W 0 (g)

∂g 2 ∗ 1

= β ∂q 1 ∂g

{ ∂w 1 ∂g 1 ∗

1

− ∂w 1 ∂g 2 ∗

1

} = β ηq 0 (1 − q 0 )

1 − q 0 (1 − q 0 )(η + φ)

[ q 1

g 1 ∗ 1

− 1 − q 1 g 2 ∗

1

] = 0 , (A.6)

given that g 1 ∗ 1

= q 1 and g 2 ∗1 = 1 − q 1 . Hence, ∂W 0 (g)

∂g can be written as:

∂W 0 (g)

∂g = q 0

g − 1 − q 0

1 − g + β ηq 0 (1 − q 0 )

1 − q 0 (1 − q 0 )(η + φ) log (

φ1 q 1 φ2 (1 − q 1 )

) = 0 , (A.7)

with q 1 is given by (A.2) .

The sign of ∂W 0 ∂g

is ambiguous. Writing the second-order derivative of W 0 , we find that:

∂ 2 W 0 (g)

∂g 2 = −

{ q 0 g 2

+ 1 − q 0 (1 − g) 2

} + β

[ ηq 0 (1 − q 0 )

1 − q 0 (1 − q 0 )(η + φ)

]2 1

q 1 (1 − q 1 ) . (A.8)

The second term in the expression above is positive. Indeed, as long as β > 0 non-convexities may arise in the optimization problem of the elites for the following reason. For the elites of type 1, the marginal benefit of increasing their weight is

equal to their period-1 utility, which is necessarily increasing in g, the provision of public good 1.

From (A.8) , we deduce that the second-order derivative is equal to zero when:

{ q 0 g 2

+ 1 − q 0 (1 − g) 2 } = β[

ηq 0 (1 − q 0 ) 1 − q 0 (1 − q 0 )(η + φ)

] 2 1

q 1 (g)(1 − q 1 (g)) , (A.9)

which rewrites

q 1 (g)(1 − q 1 (g)) = β[ ηq 0 (1 − q 0 )

1 − q 0 (1 − q 0 )(η + φ) ] 2

g 2 (1 − g) 2 (1 − q 0 ) g 2 + q 0 (1 − g) 2

, (A.10)

with q 1 (g) given by (A.2) , so

∂ 2 W 0 (g)

∂g 2 < 0 when q 1 (g)(1 − q 1 (g)) > β

[ ηq 0 (1 − q 0 )

1 − q 0 (1 − q 0 )(η + φ)

]2 g 2 (1 − g) 2

(1 − q 0 ) g 2 + q 0 (1 − g) 2 , and (A.11)

∂ 2 W 0 (g)

∂g 2 ≥ 0 when q 1 (g)(1 − q 1 (g)) ≤ β

[ ηq 0 (1 − q 0 )

1 − q 0 (1 − q 0 )(η + φ)

]2 g 2 (1 − g) 2

(1 − q 0 ) g 2 + q 0 (1 − g) 2 . (A.12)

q 1 (g)(1 − q 1 (g)) as a function of g is represented by the black curve in Fig. A.2 in the case where there exists g ∈ [0 , 1] such that

∂ 2 W 0 (g)

∂g 2 ≥ 0 (this case will formally be characterized below). As the function q 1 (g) is linear and increasing, the

15

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

Fig. A.2. Solutions of Eq. (A.8) . Note: the blue curve is the RHS of (A.10) , and the black curve is the LHS of (A.10) .

function q 1 (g)(1 − q 1 (g)) is concave in g, has an inverted U shape, and takes strictly positive values on [0,1], as q 1 (g) ∈ (0 , 1) .

β[ ηq 0 (1 −q 0 )

1 −q 0 (1 −q 0 )(η+ φ) ] 2

g 2 (1 −g) 2 (1 −q 0 ) g 2 + q 0 (1 −g) 2

as a function of g is represented by the blue curve in Fig. A.2 . The function is single

peaked, and necessarily equal to zero in the corners.

The following result follows from the previous discussion, as summarized in Fig. A.2 :

Lemma 1. Let

˜ β = 1 [

ηq 0 (1 −q 0 ) 1 −q 0 (1 −q 0 )(η+ φ)

] 2 min

g∈ [0 , 1] q 1 (g)(1 − q 1 (g))

g 2 (1 −g) 2 (1 −q 0 ) g 2 + q 0 (1 −g) 2

. (A.13)

• If β ≤ ˜ β, then ∂ 2 W 0 (g)

∂g 2 ≤ 0 for any g ∈ [0 , 1] .

• If β > ˜ β, then the equation ∂ 2 W 0 ∂g 2

= 0 admits two solutions ˜ g 1 , ̃ g 2 ∈ (0 , 1) , ˜ g 1 � = ˜ g 2 such that: • If g < ˜ g 1 or g > ˜ g 2 then

∂ 2 W 0 ∂g 2

< 0 .

• If g ∈ [ ̃ g 1 , ̃ g 2 ] , then ∂ 2 W 0 ∂g 2

≥ 0 .

Since the non-convexity arises from the inter-temporal concerns of the elites, the magnitude of the non-convexity can

be simply related to the magnitude of the time preference parameter β. In the case where β ≤ ˜ β, the non-convexity in the optimization problem is weak, so that the function W 0 (. ) is concave.

In the case where β > ˜ β, Fig. A.2 depicts the determination of the solutions of the equation ∂ 2 W 0 ∂g 2

= 0 . We deduce that the convexity of the function

∂W 0 (. ) ∂g

changes twice. The function is first decreasing, then increasing, and

decreases again. Additionally, when g → 0 , then ∂W 0 (g) ∂g

→ ∞ . When g → 1 , then ∂W 0 (g) ∂g

→ −∞ . Given these results, we have represented the function

∂W 0 (. ) ∂g

in Fig. A.3 .

As represented in Fig. A.3 , the function ∂W 0 (g)

∂g can at most cross the horizontal axis three times. When the function

crosses the horizontal line and is decreasing, then the solution of the equation ∂W 0 (g)

∂g = 0 is stable. This is the case of the

two extreme solutions ˜ g L and ˜ g H , where the subscripts L and H stand for “low” and “high” respectively. When the function ∂W 0 (g)

∂g crosses the horizontal axis and is increasing, then the solution of the equation

∂W 0 (g) ∂g

= 0 is unstable. This is the case of the intermediate solution ˜ g U , as represented in Fig. A.3 .

The second step of the proof consists in establishing the following Lemma:

Lemma 2. ∂ 2 W 0 (g) ∂ g∂ q 0

> 0 when g is such that ∂W 0 (g)

∂g = 0

In order to establish this result, we need to prove the following Lemma first:

Lemma 3. ∂q 1 ∂q 0

> 0 , with q 1 given by (A.2) .

From (A.1) , q 1 solves the following fixed point equation:

q 1 (g) = q 0 + q 0 (1 − q 0 )((1 − q 0 ) U 11 (g) − q 0 U 22 (g)) , (A.14) with U 11 (g) = ηg + (η + φ1 ) q 1 (g) and U 22 (g) = η(1 − g) + (η + φ2 )(1 − q 1 (g)) .

In order to prove that ∂q 1 (g) ∂q 0

> 0 , we prove that the RHS of (A.14) is increasing in q 0 . As represented in Figure A.4 , this

would shift the RHS of (A.14) upward and prove that q 1 (g) increases with q 0 , as the intersection between the LHS and RHS

of (A.14) is shifted to the right.

16

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

Fig. A.3. ∂W 0 ∂ g

as a function of g.

Fig. A.4. Solutions of the fixed point Eq. (A.14) Note: the black line is the RHS of (A.14) .

After denoting the RHS of (A.14) as RHS(g) to ease the notation, we find that ⎧ ⎪ ⎨ ⎪ ⎩

∂RHS(g)

∂q 0 = 1 + (1 − q 0 )(1 − 3 q 0 ) U 11 (g) − q 0 (2 − 3 q 0 ) U 22 (g) , and

∂ 2 RHS(g)

∂q 2 0

= (−4 + 6 q 0 ) U 11 (g) + (−2 + 6 q 0 ) U 22 (g) . (A.15)

Hence, the second-order derivative of RHS(g) is linearly increasing in q 0 and it equals zero at

q 0 = ˜ q 0 = 2 U 11 + U 22

3(U 11 + U 22 ) ∈ (0 , 1) . (A.16)

Thus, ∂RHS(g)

∂q 0 is U -shaped, and is minimum in ˜ q 0 . Notwithstanding a few computations, we find that in q 0 = ˜ q 0 ,

∂RHS(g)

∂q = 1 − [ U

11 (g)] 2 (U 11 (g) + 2 U 22 (g)) 3(U 11 (g) + U 22 (g)) 2 −

[ U 22 (g)] 2 (U 22 (g) + 2 U 11 (g)) 3(U 11 (g) + U 22 (g)) 2 , (A.17)

0

17

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

which rewrites

∂RHS(g)

∂q 0 = 1 − 1

3 (U 11 (g) + U 22 (g)) + 1

3

U 11 (g) U 22 (g)

U 11 (g) + U 22 (g) . (A.18)

As

1

3 (U 11 (g) + U 22 (g)) = 1

3 { 2 η + φ1 q 1 + φ2 (1 − q 2 ) } , (A.19)

1

3 (U 11 (g) + U 22 (g)) < 1

3 { 2 η + max (φ1 , φ2 ) } < 1 , (A.20)

given that η ∈ [0 , 1 / 4] and max (φ1 , φ2 ) ≤ 1 / 2 . We deduce that ∂RHS(g)

∂q 0 > 0 . (A.21)

This concludes the proof of Lemma 3 . We can now turn to the proof of Lemma 2 .

By differentiating ∂W 0 (g)

∂g given in (A.7) with respect to q 0 , we find:

∂ 2 W 0 (g)

∂ g∂ q 0 = 1

g + 1

1 − g + βη (1 − 2 q 0 ) + [ q 0 (1 − q 0 )] 2 (φ2 − φ1 )

(1 − q 0 (1 − q 0 )(η + φ)) 2 log

( φ1 q 1

φ2 (1 − q 1 )

)

+ β ηq 0 (1 − q 0 ) 1 − q 0 (1 − q 0 )(η + φ)

∂q 1 ∂q 0

1

q 1 (1 − q 1 ) . (A.22)

When ∂W 0 (g)

∂g = 0 ,

β log ( φ1 q 1 (g)

φ2 (1 − q 1 (g)) ) =

1 −q 0 1 −g −

q 0 g

ηq 0 (1 −q 0 ) 1 −q 0 (1 −q 0 )(η+ φ)

. (A.23)

Rewriting the cross derivative of W 0 by substituting (A.23) in (A.22) , we find

∂ 2 W 0 (g)

∂ g∂ q 0 = 1

g + 1

1 − g + (1 − 2 q 0 ) + [ q 0 (1 − q 0 )] 2 (φ2 − φ1 ) q 0 (1 − q 0 )(1 − q 0 (1 − q 0 )(η + φ))

{ 1 − q 0 1 − g −

q 0 g

} + β ηq 0 (1 − q 0 )

1 − q 0 (1 − q 0 )(η + φ) ∂q 1 (g)

∂q 0

1

q 1 (g)(1 − q 1 (g)) , (A.24)

which can be rewritten:

∂ 2 W 0 (g)

∂ g∂ q 0 = q 0

g

{ 1 − (1 − q 0 ) 2 (η + φ + q 0 (φ2 − φ1 ))

(1 − q 0 )(1 − q 0 (1 − q 0 )(η + φ))

} + 1 − q 0

1 − g

{ 1 − q 2

0 (η + φ + (1 − q 0 )(φ2 − φ1 ))

q 0 (1 − q 0 (1 − q 0 )(η + φ))

}

+ β ηq 0 (1 − q 0 ) 1 − q 0 (1 − q 0 )(η + φ)

∂q 1 (g)

∂q 0

1

q 1 (g)(1 − q 1 (g)) > 0 . (A.25)

Take the case where φ2 ≥ φ1 without loss of generality. The second term in the RHS of the previous equation is then necessarily positive. We find that the first term in the RHS of the previous equation is also positive when η ∈ [0 , 1 / 4] and φ2 ∈ [0 , 1 / 2] . 22 Finally, the third term in the RHS is also positive, as ∂q 1 (g)

∂q 0 > 0 from Lemma 3 . We have proven that

∂ 2 W 0 (g) ∂ g∂ q 0

>

0 when ∂W 0 (g)

∂g = 0 . The result holds when φ1 ≥ φ2 by symmetry of the problem.

Combining this last result with our previous analysis, we deduce our final intermediary result:

Lemma 4.

• If β ≤ ˜ β, then there is a single stable solution ˜ g a ∈ (0 , 1) that solves ∂W 0 (g) ∂g = 0 . • If β > ˜ β, there exist two threshold values of the initial fraction of type 1 individuals q 0 , q and q , with 0 < q ≤ q < 1 such

that:

- If q 0 ≥ q or q 0 ≤ q , then there is a single stable solution ˜ g that solves ∂W 0 (g) ∂g = 0 . - If q 0 ∈ ( q , q ) , then the equation ∂W 0 (g) ∂g = 0 admits two stable solutions ˜ g L and ˜ g H , and one unstable solution ˜ g U with

˜ g < ˜ g < ˜ g and ˜ g , ̃ g , ̃ g ∈ (0 , 1) .

L U H L U H

22 To see this, denote N(q 0 ) = 1 − (1 − q 0 ) 2 (η + φ + q 0 (φ2 − φ1 )) the numerator of the first term in the RHS of (A.25) . Hence, N(q 0 ) > 1 − (1 − q ) 2 (η + (1 + q 0 ) max (φ1 , φ2 ) . Since 1 − (1 − q ) 2 (η + (1 + q 0 ) max (φ1 , φ2 ) is a decreasing function of q 0 , it is maximum in q 0 = 0 . We deduce that N(q 0 ) > 1 − (η + max (φ1 , φ2 ) ≥ 0 when φi ∈ [0 , 1 / 2] and η ∈ [0 , 1 / 4] .

18

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

Fig. A.5. Effect on g → ∂W 0 (g) ∂ g

of an increase in q 0 .

Fig. A.6. Effect on g → ∂W 0 (g) ∂ g

of an increase in q 0 when q 0 = q .

In the case where β ≤ ˜ β, then from Lemma 1 , ∂ 2 W 0 (g)

∂g 2 < 0 for any g ∈ [0 , 1] . Hence, the function W 0 (g) is concave, so the

optimization problem admits a unique stable solution ˜ g a . This solution belongs to (0,1), as ∂W 0 (g)

∂g → −∞ when g → 0 , and

∂W 0 (g) ∂g

→ + ∞ when g → 1 . This concludes the proof of the first point of Lemma 4 . In the case where β > ˜ β the results of Lemma 4 can be illustrated with graphs. The effect of an increase in q 0 on the

function ∂W 0 (. )

∂g is represented in Fig. A.5 . Given our results in Lemma 2 , the effect of an increase in q 0 on the function

∂W 0 (. ) ∂g

can be represented as in Fig. A.5 . ∂ 2 W 0 (g) ∂ g∂ q 0

> 0 when ∂W 0 (g)

∂g = 0 , so ∂W 0 (. )

∂g is shifted upwardly along the horizontal axis. We

can deduce from Lemma 2 that ˜ g L and ˜ g H increase with q 0 , and that ˜ g U necessarily decreases with q 0 .

Hence, when q 0 is sufficiently high, then the U -shaped part of the graph between ˜ g L and ˜ g U is shifted above the hori-

zontal axis, as represented in Fig. A.6 , and only one equilibrium remains: ˜ g H . The value of q 0 such that the function ∂W 0 (. )

∂g

is exactly tangent to the horizontal axis in ˜ g L is denoted q , and belongs to (0,1). Indeed, q is strictly in the segment (0,1), as

when q 0 = 0 , ∂W 0 (g) ∂g < 0 for any value of g, while when q 0 = 1 , then ∂W 0 (g)

∂g > 0 for any value of g. Since

∂W 0 (. ) ∂g

must switch

sign when q 0 = q , we deduce that q ∈ (0 , 1) . The reasoning is similar for q . When q 0 decreases, then the inverted- U shape on the segment of the graph representing

∂W 0 (. ) ∂g

on [ ̃ g U , ̃ g H ] is shifted below the horizontal axis, as represented in Fig. A.7 , and only one equilibrium remains: ˜ g L . The

value of q 0 such that the function ∂W 0 (. )

∂g is tangent to the horizontal axis in ˜ g H is denoted q , and belongs to (0,1), given the

same reasoning as the one developed above.

19

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

Fig. A.7. Effect on g → ∂W 0 (g) ∂ g

of a decrease in q 0 when q 0 = q .

Finally, when q ∈ [ q , q ] , then there are two stable solutions of the equation ∂W 0 ∂g

= 0 . These two solutions, ˜ g L and ˜ g H , have already been represented in Fig. A.3 . 23

To summarize, we have proven that depending on the initial cultural composition of the population, there can be either

one or two stable Subgame Perfect Equilibria.

When β ≤ ˜ β: the game admits a single stable Subgame Perfect Equilibria. The period-0 equilibrium allocation (g 1 ∗ 0

, g 2 ∗ 0

)

is: { g 1 ∗

0 = ˜ g a , and

g 2 ∗ 0

= 1 − ˜ g a (A.26)

with ˜ g a is the unique solution of ∂W 0 (g)

∂g = 0 , as established in Lemma 4 .

In period 1, the elites choose: { g 1 ∗

1 = q 1 ( ̃ g a ) ,

g 2 ∗ 1

= 1 − q 1 ( ̃ g a ) , (A.27)

with

q 1 ( ̃ g a ) = q 0 (1 − q 0 (1 − q 0 )(η + φ2 )) + ηq 0 (1 − q 0 ) { (1 − q 0 ) ̃ g a − q 0 (1 − ˜ g a ) }

1 − q 0 (1 − q 0 )(η + φ) , (A.28)

The citizens socialize their offspring, and choose the equilibrium effort s { τ 1 ∗( ̃ g a ) = (1 − q 0 ) { η ˜ g a + (η + φ1 ) q 1 ( ̃ g a ) } τ 2 ∗( ̃ g a ) = q 0 { η(1 − ˜ g a ) + (η + φ2 )(1 − q 1 ( ̃ g a )) } . (A.29)

The SPE can then be written as: { ( ̃ g a , 1 − ˜ g a ) ; (q 1 ( ̃ g a ) , 1 − q 1 ( ̃ g a )) ; (τ 1 ∗( ̃ g a ) , τ 2 ∗( ̃ g a )) } . When β > ˜ β and q < q : The game admits a single stable Subgame Perfect Equilibria. The period-0 equilibrium allocation

(g 1 ∗ 0

, g 2 ∗ 0

) is: { g 1 ∗

0 = ˜ g L , and

g 2 ∗ 0

= 1 − ˜ g L (A.30)

with ˜ g L the unique solution of ∂W 0 ∂g

= 0 . In period 1, the elites choose: {

g 1 ∗ 1

= q 1 ( ̃ g L ) , g 2 ∗ = 1 − q 1 ( ̃ g L ) ,

(A.31)

1

23 In order to formally characterize q and q , one can define the following system,

{ ∂W 0 ∂ g

= 0 ∂ 2 W 0 ∂ g 2

= 0 .

Given our analysis of the functions ∂W 0 (g)

∂ g and

∂ 2 W 0 (g) ∂ g 2

, this system necessarily admits two solutions ( ̃ g L , q ) , and ( ̃ g H , q ) .

20

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

with

q 1 ( ̃ g L ) = q 0 (1 − q 0 (1 − q 0 )(η + φ2 )) + ηq 0 (1 − q 0 ) { (1 − q 0 ) ̃ g L − q 0 (1 − ˜ g L ) }

1 − q 0 (1 − q 0 )(η + φ) , (A.32)

.

The citizens socialize their offspring, and choose the equilibrium effort s { τ 1 ∗

L = (1 − q 0 ) { η ˜ g L + (η + φ1 ) q 1 ( ̃ g L ) }

τ 2 ∗ L

= q 0 { η(1 − ˜ g L ) + (η + φ2 )(1 − q 1 ( ̃ g L )) } . (A.33)

The SPE can then be written as: { ( ̃ g L , 1 − ˜ g L ) ; (q 1 ( ̃ g L ) , 1 − q 1 ( ̃ g L )) ; (τ 1 ∗L , τ 2 ∗L ) } . When β > ˜ β and q > q : The game admits a single stable Subgame Perfect Equilibria.

The period-0 equilibrium allocation (g 1 ∗ 0

, g 2 ∗ 0

) is: { g 1 ∗

0 = ˜ g H , and

g 2 ∗ 0

= 1 − ˜ g H (A.34)

with ˜ g H the unique solution of ∂W 0 ∂g

= 0 . In period 1, the elites choose: {

g 1 ∗ 1

= q 1 ( ˜ g H ) , g 2 ∗

1 = 1 − q 1 ( ˜ g H ) ,

(A.35)

with

q 1 ( ˜ g H ) = q 0 (1 − q 0 (1 − q 0 )(η + φ2 )) + ηq 0 (1 − q 0 ) { (1 − q 0 ) ˜ g H − q 0 (1 − ˜ g H ) }

1 − q 0 (1 − q 0 )(η + φ) , (A.36)

.

The citizens socialize their offspring, and choose the equilibrium effort s { τ 1 ∗

H = (1 − q 0 ) { η ˜ g H + (η + φ1 ) q 1 ( ˜ g H ) }

τ 2 ∗ H

= q 0 { η(1 − ˜ g H ) + (η + φ2 )(1 − q 1 ( ˜ g H )) } . (A.37)

The SPE can then be written as: { ( ˜ g H , 1 − ˜ g H ) ; (q 1 ( ˜ g H ) , 1 − q 1 ( ˜ g H )) ; (τ 1 ∗H , τ 2 ∗H ) } . Finally, when β > ˜ β and q ∈ [ q , q ] : the game admits two stable SPE (and one unstable SPE). In the first stable SPE, the period-0 equilibrium allocation (g 1 ∗

0 , g 2 ∗

0 ) is: {

g 1 ∗ 0

= ˜ g H , and g 2 ∗

0 = 1 − ˜ g H

(A.38)

with ˜ g H the first solution of ∂W 0 ∂g

= 0 such that ∂ 2 W 0 ∂g 2

< 0 .

In the second stable SPE, the period-0 equilibrium allocation (g 1 ∗ 0

, g 2 ∗ 0

) is: { g 1 ∗

0 = ˜ g L , and

g 2 ∗ 0

= 1 − ˜ g L (A.39)

with ˜ g H the second solution of ∂W 0 ∂g

= 0 such that ∂ 2 W 0 ∂g 2

< 0 .

In period 1, the elites choose: { g 1 ∗

1 = q 1 ( ̃ g K ) ,

g 2 ∗ 1

= 1 − q 1 ( ̃ g K ) , (A.40)

with

q 1 ( ˜ g K ) = q 0 (1 − q 0 (1 − q 0 )(η + φ2 )) + ηq 0 (1 − q 0 ) { (1 − q 0 ) ̃ g K − q 0 (1 − ˜ g K ) }

1 − q 0 (1 − q 0 )(η + φ) , (A.41)

for K ∈ { L, H} . The citizens socialize their offspring, and choose the equilibrium efforts { τ 1 ∗

K = (1 − q 0 ) { η ˜ g K + (η + φ1 ) q 1 ( ˜ g K ) }

τ 2 ∗ K

= q 0 { η(1 − ˜ g K ) + (η + φ2 )(1 − q 1 ( ˜ g K )) } . (A.42)

The two SPEs can then be written as: { ( ˜ g H , 1 − ˜ g H ) ; (q 1 ( ˜ g H ) , 1 − q 1 ( ˜ g H )) ; (τ 1 ∗H , τ 2 ∗H ) } and { ( ̃ g L , 1 − ˜ g L ) ; (q 1 ( ̃ g L ) , 1 − q 1 ( ̃ g L )) ; (τ 1 ∗, τ 2 ∗ ) } . This concludes the proof of Proposition 1 .

L L

21

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

A3. Proof of Proposition 2

First, we establish the following intermediary result:

Lemma 5. There exists two thresholds q DPE and q DPE

in [0,1] such that if q 0 ∈ ( q DPE , q DPE ) , then

q 1 (0 , 1) = min (g 1

0 ,g 2

0 ) q 1 (g

1 0 , g

2 0 ) <

φ2

φ1 + φ2 < max (g 1 0 ,g 2

0 ) q 1 (g

1 0 , g

2 0 ) = q 1 (1 , 0) , (A.43)

with g 1 0

+ g 2 0

= 1 and q 1 (g 1 0 , g 2 0 ) the solution of (A.1) . First, notice that

min (g 1

0 , 1 −g 1

0 ) q 1 (g

1 0 , 1 − g 1 0 ) = q 1 (0 , 1) . (A.44)

In words, the minimum value of q 1 (g 1 0 , g 2

0 ) when g 1

0 + g 2

0 = 1 is such that only good two is provided in period 0. But since

q 0 → q 1 (0 , 1) is increasing in q 0 , there exists a unique threshold q DPE in [0,1] such that if q 0 ≤ q DPE , then q 1 (0 , 1) ≤ φ 2

φ1 + φ2 . Following the same reasoning, since

max (g 1

0 , 1 −g 1

0 ) q 1 (g

1 0 , 1 − g 1 0 ) = q 1 (1 , 0) , (A.45)

and that q 0 → q 1 (1 , 0) is increasing in q 0 , there exists a unique threshold q DPE in [0,1] such that if q 0 ≥ q DPE , then q 1 (1 , 0) ≥ φ2

φ1 + φ2 .

Since max (g 1

0 , 1 −g 1

0 )

q 1 (g 1 0 , 1 − g 1

0 ) = q 1 (1 , 0) > min (g 1

0 , 1 −g 1

0 )

q 1 (g 1 0 , 1 − g 1

0 ) = q 1 (0 , 1) , then q DPE < q DPE is necessarily true.

This concludes the proof of Lemma 5 .

Now, we solve the dynamically production efficient provision by backward induction. We assume in this proof that, when

indifferent between producing goods 1 and 2, the default option is to produce good 2.

We distinguish three cases in this proof. Case A: q 0 ∈ [ q DPE , q DPE ] , case B: q 0 < q DPE , and Case C: q 0 > q DPE . In each case, we fully characterize the DPE of the model.

Case A: q 0 ∈ [ q DPE , q DPE ] . A dynamically efficient Production will necessarily be such that the constraint g 1 t + g 2 t = 1 is satisfied at equality in any period t ∈ { 0 , 1 } : all the available resources are used for the production along the efficient path. We find that in period 1,

∂ p 1 (g 1 1 , 1 − g 1

1 )

∂g 1 1

= q 1 (g 1 0 , g 2 0 ) φ1 − (1 − q 1 (g 1 0 , g 2 0 )) φ2 , (A.46)

so

∂ p t (g 1 1 , 1 − g 1

1 )

∂g 1 1

> 0 if and only if q 1 (g 1 0 , 1 − g 1 0 ) >

φ2

φ1 + φ2 . (A.47)

Hence, { g 1 ,DPE

1 = 1

g 2 ,DPE 1

= 0 if q 1 (g

1 0 , 1 − g 1 0 ) >

φ2

φ1 + φ2 , and (A.48)

{ g 1 ,DPE

1 = 0

g 2 ,DPE 1

= 1 otherwise. (A.49)

From (A.2) , since ∂q 1 ∂g 1

0

= ηq 0 (1 −q 0 ) 1 −q 0 (1 −q 0 )(η+ φ)

, we deduce that

∂ p 0 (g 1 0 , 1 − g 1

0 )

∂g 1 0

= q 0 φ1 − (1 − q 0 ) φ2 + β ηq 0 (1 − q 0 )

1 − q 0 (1 − q 0 )(η + φ) { φ1 g 1 ,DPE

1 − φ2 (1 − g 1 ,DPE

1 ) } , (A.50)

Hence, there are two possible outcomes in period 1. In the first outcome, the equilibrium is such that q 1 (g 1 0 , 1 − g 1

0 ) >

φ2

φ1 + φ2 . In this case, the first-order condition in period 0 is:

∂ p 0 (g 1 0 , 1 − g 1

0 )

∂g 1 0

= Z (β ) = q 0 φ1 − (1 − q 0 ) φ2 + β ηq 0 (1 − q 0 )

1 − q 0 (1 − q 0 )(η + φ) φ1 . (A.51)

In the second outcome, the equilibrium is such that q 1 (g 1 0 , 1 − g 1

0 ) ≤ φ2

φ1 + φ2 . In this case, the first-order condition in period 0 is:

∂ p 0 (g 1 0 , 1 − g 1

0 )

∂g 1 = Z (β ) = q 0 φ1 − (1 − q 0 ) φ2 − β

ηq 0 (1 − q 0 ) 1 − q 0 (1 − q 0 )(η + φ)

φ2 . (A.52)

0

22

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

Fig. A.8. Characterization of DPE.

In order to characterize the dynamic efficient production, we consider all the possible cases, depending on the parameter

values.

Case 1: q 0 φ 1 − (1 − q 0 ) φ2 > 0 and β < β1 (q 0 ) . This case is depicted in the left panel of Fig. A.8 . There exists a certain

threshold β1 (q 0 ) such that if β < β1 (q 0 ) , then Z (β ) > 0 and Z (β ) > 0 . First, since Z (β ) > 0 , then there is no equilibrium such that good 2 is provisioned in period 0. Since Z (β ) > 0 , an equi-

librium is necessarily such that

g 1 ,DPE 0

= 1 and g 2 ,DPE 0

= 0 . (A.53) As only good 1 is provided in period 0, the fraction of individuals of type 1 reaches q 1 (1 , 0) in period 1. From Lemma 5 ,

q 1 (1 , 0) > φ2

φ1 + φ2 is satisfied, so

g 1 ,DPE 1

= 1 and g 2 ,DPE 1

= 0 . (A.54) Case 2: q 0 φ

1 − (1 − q 0 ) φ2 > 0 and β ≥ β1 (q 0 ) . From Fig. A.8 , we see that there can be two potential equilibrium out- comes. In the first outcome, since Z (β ) ≥ 0 ,

g 1 , DPE 0

= 1 , g 2 , DPE 0

= 0 , and (A.55)

g 1 , DPE 1

= 1 , g 2 , DPE 1

= 0 (A.56) from Lemma 5 . If this outcome is realized, the production in period 0 will be

p 0 (1 , 0) = q 0 φ1 + βq (1 , 0) φ1 . (A.57) In the second outcome, since Z (β ) ≤ 0 ,

g 1 ,DPE 0

= 0 , g 2 ,DPE 0

= 1 and (A.58)

g 1 , DPE 1

= 0 , g 2 , DPE 1

= 1 (A.59) from Lemma 5 . If this outcome is realized, the production in period 0 will be

p 0 (0 , 1) = (1 − q 0 ) φ2 + β(1 − q (0 , 1)) φ2 . (A.60) Hence, the first outcome is realized if

p 0 (1 , 0) > p 0 (0 , 1) , (A.61)

or

q 0 > φ2

φ1 + φ2 + β

φ1 + φ2 { (1 − q (0 , 1)) φ 2 − q (1 , 0) φ1 } . (A.62)

Let denote G (q 0 ) = q 0 − βφ1 + φ2 { (1 − q (0 , 1)) φ2 − q (1 , 0) φ1 } , so that the previous inequality rewrites

G (q 0 ) > φ2

φ1 + φ2 . (A.63)

23

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

We find that

∂G (q 0 )

∂q 0 = 1 + β

φ1 + φ2 {

∂q (0 , 1)

∂q 0 φ2 + ∂q (1 , 0)

∂q 0 φ1

} . (A.64)

As ∂q (0 , 1)

∂q 0 > 0 and

∂q (1 , 0) ∂q 0

> 0 , ∂G (q 0 ) ∂q 0

> 0 . Hence, it is direct that there exists a unique threshold ˜ q T EMP such that

G (q 0 ) > φ2

φ1 + φ2 if q 0 > ˜ q T EMP , and

G (q 0 ) ≤ φ 2

φ1 + φ2 otherwise. (A.65)

Two important properties are worth stating before pursuing the proof. First, we find that ˜ q T EMP increases with β if and only if (1 − q (0 , 1)) φ2 − q (1 , 0) φ1 > 0 . By symmetry, (1 − q (0 , 1)) φ2 − q (1 , 0) φ1 > 0 iff φ2 > φ1 , we have demonstrated that ˜ q T EMP increases with β iff φ2 > φ1 . Second, we can establish that ˜ q T EMP >

φ2

φ1 + φ2 because G ( φ2

φ1 + φ2 ) < φ2

φ1 + φ2 .

Case 3: q 0 φ 1 − (1 − q 0 ) φ2 < 0 and β < β2 (q 0 ) . As represented on the right panel of Fig. A.8 . There exists a certain thresh-

old β2 (q 0 ) such that if β < β2 (q 0 ) , Z (β ) < 0 , and Z (β ) ≥ 0 otherwise. First, since Z (β ) < 0 , then there is no equilibrium such that good 1 is provisioned in period 0. The equilibrium is neces-

sarily such that

g 1 ,DPE 0

= 0 and g 2 ,DPE 0

= 1 . (A.66) As only good 2 is provided in period 0, the fraction of individuals of type 1 reaches q 1 (0 , 1) in period 1. From Lemma 5 ,

q 1 (0 , 1) < φ2

φ1 + φ2 is satisfied, so

g 1 ,DPE 1

= 0 and g 2 ,DPE 1

= 1 . (A.67) Case 4: q 0 φ

1 − (1 − q 0 ) φ2 < 0 and β ≥ β2 (q 0 ) . From Fig. A.8 , we see that there can be two potential equilibrium out- comes. In the first outcome, since Z (β ) ≥ 0 ,

g 1 , DPE 0

= 1 , g 2 , DPE 0

= 0 and (A.68)

g 1 , DPE 1

= 1 , g 2 , DPE 1

= 0 (A.69) from Lemma 5 . If this outcome is realized, the production in period 0 will be

p 0 (1 , 0) = q 0 φ1 + βq (1 , 0) φ1 . (A.70) In the second outcome, since Z (β ) ≤ 0 ,

g 1 , DPE 0

= 0 , g 2 , DPE 0

= 1 and (A.71)

g 1 , DPE 1

= 0 , g 2 , DPE 1

= 1 (A.72) from Lemma 5 . If this outcome is realized, the production in period 0 will be

p 0 (0 , 1) = (1 − q 0 ) φ2 + β(1 − q (0 , 1)) φ2 . (A.73) Hence, the first outcome is realized if

p 0 (1 , 0) > p 0 (0 , 1) , (A.74)

or

q 0 > φ2

φ1 + φ2 + β

φ1 + φ2 { (1 − q (0 , 1)) φ 2 − q (1 , 0) φ1 } . (A.75)

But since φ2 > φ1 , 1 − q (0 , 1) > q (1 , 0) by symmetry of the model. Hence, (1 − q (0 , 1)) φ2 − q (1 , 0) φ1 > 0 . This implies that the inequalities

q 0 > φ2

φ1 + φ2 + β

φ1 + φ2 { (1 − q (0 , 1)) φ 2 − q (1 , 0) φ1 } . (A.76)

and

φ1 q 0 + φ2 (1 − q 0 ) < 0 or equivalently q 0 < φ2

φ1 + φ2 (A.77)

cannot be simultaneously satisfied. We deduce that

p 0 (1 , 0) ≤ p 0 (0 , 1) (A.78)

24

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

necessarily holds. In case 4, the only equilibrium is then such that

g 1 , DPE 0

= 0 , g 2 , DPE 0

= 1 and (A.79)

g 1 , DPE 1

= 0 , g 2 , DPE 1

= 1 (A.80) Case B: q 0 < q

DPE . In this case, independently from what is provided by the elites in period 0, q (1 , 0) < φ2

φ1 + φ2 , so the elites always provide good 2 in period 1. In period 0, the elites will provide good 1 if and only if β < β1 (q 0 ) , and good 2 otherwise.

Case C: q 0 > q DPE

. In this case, independently from what is provided by the elites in period 0, q (0 , 1) > φ2

φ1 + φ2 , so the elites always provide good 1 in period 1. In period 0, the elites will provide good 1 if and only if β > β1 (q 0 ) , and good 2 otherwise.

This concludes the full characterization of the DPE.

In order to focus on a case where the DPE is continuous relative to the parameter values, we assume that β is sufficiently

large, in that Z (q 0 ) < 0 and Z (q 0 ) > 0 for any q 0 ∈ [0 , 1] . Equivalently, β ≥ β DPE = max q 0 ∈ [0 , 1] (β1 (q 0 ) , β2 (q 0 )) , with ⎧ ⎨

⎩ β1 (q 0 ) = max q 0 ∈ [0 , 1]

[ (q 0 φ

1 − (1 − q 0 ) φ2 ) 1 −q 0 (1 −q 0 )(η+ φ) ηq 0 (1 −q 0 ) φ2 ]

β2 (q 0 ) = max q 0 ∈ [0 , 1] [ (−q 0 φ1 + (1 − q 0 ) φ2 ) 1 −q 0 (1 −q 0 )(η+ φ) ηq 0 (1 −q 0 ) φ1

] .

(A.81)

When β ≥ βDPE and q 0 < q DPE (Case B), the equilibrium is necessarily such that g 1 , DPE

0 = 0 , g 2 , DPE

0 = 1 and (A.82)

g 1 , DPE 1

= 0 , g 2 , DPE 1

= 1 . (A.83) When β ≥ βDPE and q 0 > q DPE (Case C), the equilibrium is necessarily such that

g 1 , DPE 0

= 1 , g 2 , DPE 0

= 0 and (A.84)

g 1 ,DPE 1

= 1 , g 2 ,DPE 1

= 0 . (A.85)

When β ≥ βDPE and q 0 ∈ [ q DPE , q DPE ] (Case A), the equilibrium is necessarily such that g 1 , DPE

0 = 1 , g 2 , DPE

0 = 0 and (A.86)

g 1 , DPE 1

= 1 , g 2 , DPE 1

= 0 (A.87) if q 0 ≥ ˜ q T EMP , and

g 1 , DPE 0

= 1 , g 2 , DPE 0

= 0 and (A.88)

g 1 , DPE 1

= 1 , g 2 , DPE 1

= 0 (A.89) otherwise. Hence, denoting ˜ q DPE the threshold such that

˜ q DPE =

⎧ ⎨ ⎩

˜ q T EMP if ˜ q T EMP ∈ [ q DPE , q DPE ] q DPE if ˜ q T EMP < q DPE

q DPE

otherwise,

(A.90)

we have established that for any q 0 ∈ [0 , 1] , g 1 , DPE

0 = 1 , g 2 , DPE

0 = 0 and (A.91)

g 1 , DPE 1

= 1 , g 2 , DPE 1

= 0 (A.92) if q 0 ≥ ˜ q DPE , and

g 1 , DPE 0

= 0 , g 2 , DPE 0

= 1 and (A.93)

g 1 , DPE 1

= 0 , g 2 , DPE 1

= 1 (A.94) otherwise. This concludes the proof of Proposition 2 .

25

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

A4. Proof of Proposition 3

Consider a SPE { (g 1 ∗ 0

, g 2 ∗ 0

) ; (g 1 ∗ 1

, g 2 ∗ 1

) ; (τ 1 ∗, τ 2 ∗ ) } . Given the cultural dynamics (A.2) , q 1 (g

1 ∗ 0 , 1 − g 1 ∗0 ) > q 0 if and only if g 1 ∗0 > g 0 . (A.95)

with

g 0 = {

q 0 +(φ2 −φ1 )(1 −q 0 ) η

if q 0 +(φ2 −φ1 )(1 −q 0 )

η < 1 , and

1 otherwise. . (A.96)

We define an “excess provision” function Z(β ) as

Z(β ) = g 1 ∗0 − g 0 . (A.97) From Definition 2 , a revival occurs when: {

Z(β ) > 0 q 0 < ˜ q

DPE . (A.98)

We will denote g 1 0

= g to simplify the notations. We compute ∂Z(β ) ∂β

:

∂Z(β )

∂β = ∂g

1 ∗ 0

∂β = ∂

2 W 0 (g 1 ∗ 0

, 1 − g 1 ∗ 0

) /∂ g∂ β

−∂ 2 W 0 (g 1 ∗0 , 1 − g 1 ∗0 ) /∂g 2 . (A.99)

Since −∂ 2 W 0 (g 1 ∗0 , 1 − g 1 ∗0 ) /∂g 2 > 0 in the SPE, ∂Z(β ) ∂β

and ∂ 2 W 0 (g 1 ∗ 0

, 1 − g 1 ∗ 0

) /∂ g∂ β have the same sign, with

∂ 2 W 0 (g 1 ∗ 0

, 1 − g 1 ∗ 0

)

∂ g∂ β = ηq 0 (1 − q 0 )

1 − q 0 (1 − q 0 )(η + φ) log (

φ1 q 1 (g 1 ∗ 0

, 1 − g 1 ∗ 0

)

φ2 (1 − q 1 (g 1 ∗0 , 1 − g 1 ∗0 )) ) . (A.100)

Hence,

∂Z(β )

∂β > 0 (A.101)

iff

φ1 q 1 (g 1 ∗ 0 , 1 − g 1 ∗0 ) > φ2 (1 − q 1 (g 1 ∗0 , 1 − g 1 ∗0 )) , (A.102)

or equivalently iff

q 1 (g 1 ∗ 0 , 1 − g 1 ∗0 ) >

φ2

φ1 + φ2 . (A.103)

We deduce the following intermediary result:

Lemma 6. ∂g 1 ∗

0 ∂β

> 0 in at least one SPE if q 0 > k q 1 >

φ2

φ1 + φ2 (q 0 ) , with

k q 1 >

φ2

φ1 + φ2 (q 0 ) =

{ p(q 0 ) if p(q 0 ) ∈ [0 , 1] 1 if p(q 0 ) > 1 0 otherwise ,

(A.104)

with

p(q 0 ) = q 0 + 1

ηq 0 (1 − q 0 ) { φ

2

φ1 + φ2 (1 − q 0 (1 − q 0 )(η + φ) − q 0 (1 − q 0 (1 − q 0 )(η + φ) } . (A.105)

Proof. First, we define k q 1 >

φ2

φ1 + φ2 (q 0 ) as the value of g

1 ∗ 0

such that q 1 (g 1 ∗ 0

, 1 − g 1 ∗ 0

) > φ2

φ1 + φ2 if and only if g 1 ∗ 0

> k q 1 >

φ2

φ1 + φ2 (q 0 ) .

The determination of k q 1 >

φ2

φ1 + φ2 (q 0 ) is represented in Fig. A.9 in the case where k

q 1 > φ2

φ1 + φ2 (q 0 ) ∈ [0 , 1] . Since q 1 (g 1 ∗0 , 1 −

g 1 ∗ 0

) is linearly increasing in g 1 ∗ 0

, there exists a unique threshold k q 1 >

φ2

φ1 + φ2 (q 0 ) such that q 1 (g

1 ∗ 0

, 1 − g 1 ∗ 0

) > φ2

φ1 + φ2 if and only

if g 1 ∗ 0

> k q 1 >

φ2

φ1 + φ2 (q 0 ) .

Using the expression of q 1 (g 1 ∗ 0

, 1 − g 1 ∗ 0

) in (A.2) , we deduce that k q 1 >

φ2

φ1 + φ2 (q 0 ) can be expressed as in (A.104) .

The main idea of the proof of the previous Lemma is represented in Fig. A.10 in the case where q 0 / ∈ [ q , q ] . In this case, given Proposition 1 , there is a unique SPE for any value of β. As represented in Fig. A.10 , if k

q 1 > φ2

φ1 + φ2 (q 0 ) < q 0 , then

26

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

Fig. A.9. Determination of k q 1 >

φ2

φ1 + φ2 (q 0 ) .

Fig. A.10. Proof of Lemma 6 when q 0 / ∈ [ q , q ] .

k q 1 >

φ2

φ1 + φ2 (q 0 ) < g

1 ∗ 0

in β = 0 , as g 1 ∗ 0

= q 0 when β = 0 . Hence, ∂g 1 ∗

0 ∂β

> 0 initially, and then k q 1 >

φ2

φ1 + φ2 (q 0 ) < g

1 ∗ 0

remains satis-

fied for β > 0 by monotonicity and continuity.

By contrast, if k q 1 >

φ2

φ1 + φ2 (q 0 ) > q 0 , then k

q 1 > φ2

φ1 + φ2 (q 0 ) > g

1 ∗ 0

in β = 0 , as g 1 ∗ 0

= q 0 when β = 0 . Hence, ∂g 1 ∗

0 ∂β

< 0 initially,

and then k q 1 >

φ2

φ1 + φ2 (q 0 ) > g

1 ∗ 0

remains satisfied for β > 0 by monotonicity and continuity.

When q 0 / ∈ [ q , q ] , the proof of the Lemma is illustrated in Fig. A.11 . When β < ˜ β, there is a unique SPE ( Proposition 1 ). By contrast, there is a bifurcation at β = ˜ β. The stable equilibrium when β < ˜ β becomes unstable, and two stable equilibria emerge on each side of the unstable equilibrium, as represented in Fig. A.11 . One equilibrium is necessarily such that

∂g 1 ∗ 0

∂β >

0 . The other can be such that ∂g 1 ∗

0 ∂β

< 0 , because it can be such that k q 1 >

φ2

φ1 + φ2 (q 0 ) > g

1 ∗ 0

, as represented.

We have demonstrated that ∂g 1 ∗

0 ∂β

> 0 for at least one SPE if q 0 > k q 1 >

φ2

φ1 + φ2 (q 0 ) , with k

q 1 > φ2

φ1 + φ2 (q 0 ) given by (A.104) . �

The second step of the proof consists in proving the following result:

Lemma 7. q 0 > k q 1 >

φ2

φ1 + φ2 (q 0 ) if and only if q 0 > ˜ q 0 , with ˜ q 0 ∈ (0 , 1) .

27

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

Fig. A.11. Proof of Lemma 6 when q 0 ∈ [ q , q ] .

Fig. A.12. Determination of ˜ q 0 .

Proof. This result is based on the fact that the function k q 1 >

φ2

φ1 + φ2 (. ) is decreasing in q 0 . Indeed, q 1 (g

1 ∗ 0

, 1 − g 1 ∗ 0

) is increasing

in q 0 . Hence, when q 0 increases, the linear black curve representing q 1 (g 1 ∗ 0

, 1 − g 1 ∗ 0

) in Fig. A.9 is shifted upward. Hence, it

is direct that the threshold k q 1 >

φ2

φ1 + φ2 (q 0 ) decreases.

Given that lim q 0 → 0 p(q 0 ) = + ∞ and lim q 0 → 1 p(q 0 ) = −∞ , k q 1 > φ2 φ1 + φ2

(0) = 1 and k q 1 >

φ2

φ1 + φ2 (1) = 0 . Furthermore, as

k q 1 >

φ2

φ1 + φ2 (q 0 ) is a decreasing function of q 0 , we can represent it as in Fig. A.12 .

From Fig. A.12 , it is direct that there exists a threshold ˜ q 0 ∈ (0 , 1) such that q 0 > k q 1 >

φ2

φ1 + φ2 (q 0 ) if and only if q 0 > ˜ q 0 . �

At β = 0 ,

Z(0) = q 0 − q 0 + (φ2 − φ1 )(1 − q 0 )

η , (A.106)

so

Z(0) < 0 (A.107)

when φ2 > φ1 and η ∈ [0 , 1 / 4] . Since Z(. ) is increasing with β when q 0 > ˜ q 0 , lim β→∞

Z(β ) = 1 − g 0 ≥ 0 . (A.108)

28

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

Hence, there exists a unique threshold value ˜ β1 > 0 such that if β > ˜ β1 , then Z(β ) > 0 . Importantly, from the definition of a cultural revival, it must be that q 0 < ˜ q

DPE . Hence, we have demonstrated that when ˜ q DPE > q 0 > ˜ q 0 , the conditions for a

revival are fulfilled.

Last but not least, a revival occurs for a positive measure of parameters if ˜ q DPE > ˜ q 0 . This inequality is satisfied when β is sufficiently high. Indeed, ˜ q DPE is non-decreasing in β ( Proposition 2 ) and ˜ q 0 is independent from β. Hence, there exists some threshold ˜ β2 such that ˜ q

DPE > ˜ q 0 holds iif β > ˜ β2 . We have demonstrated that there exists a threshold ˜ q 0 ∈ (0 , 1) , and a threshold β = max ( ̃ β1 , ˜ β2 ) > 0 such that if ˜ q DPE >

q 0 > ˜ q 0 and β > β, then Z(β ) > 0 . Given that φ 2 > φ1 , there is a cultural revival favoring type 1 in the SPE. We have proven

that ˜ q DPE > q 0 > ˜ q 0 and β > ˜ β are sufficient conditions for cultural revivals favoring type 1.

References

Acemoglu, D. , 2003. Why not a political Coase theorem? Social conflict, commitment, and politics. J. Comp. Econ. 31 (4), 620–652 .

Acemoglu, D. , Robinson, J.A. , 2006. Economic backwardness in political perspective. Am. Polit. Sci. Rev. 100 (1), 115–131 . Acemoglu, D. , Robinson, J.A. , 2012. Why Nations Fail: the Origins of Power, Prosperity, and Poverty. Crown, New York .

Alesina, A. , Giuliano, P. , 2015. Culture and institutions. J. Econ. Lit. 53 (4), 898–944 . Algan, Y. , Cahuc, P. , 2014. Trust, growth, and well-being: new evidence and policy implications. In: Handbook of Economic Growth, 2. Elsevier, pp. 49–120 .

Almagro, M., Andrés-Cerezo, D., 2020. The construction of national identities. Theor. Econ. 15 (2), 763–810. doi: 10.3982/TE3040 .

Bénabou, R., Ticchi, D., Vindigni, A., 2015. Religion and innovation. Am. Econ. Rev. 105 (5), 346–351. doi: 10.1257/aer.p20151032 . Bénabou, R. , Ticchi, D. , Vindigni, A. , 2021. Forbidden fruits: rhe political economy of science, religion and growth. Rev. Econ. Stud. . forthcoming

Bénabou, R. , Tirole, J. , 2006. Belief in a just world and redistributive politics. Q. J. Econ. 121 (2), 699–746 . Bisin, A. , Rubin, J. , Seror, A. , Verdier, T. , 2021. Culture, Institutions, and the Long Divergence. NBER Working Paper No. 28488 .

Bisin, A. , Seror, A. , Verdier, T. , 2019. Religious legitimacy and the joint evolution of culture and institutions. In: Carvalho, J.-P., Iyer, S., Rubin, J. (Eds.), Advances in the Economics of Religion. Palgrave MacmilLan .

Bisin, A. , Verdier, T. , 20 0 0. Beyond the melting pot: cultural transmission, marriage, and the evolution of ethnic and religious traits. Q. J. Econ. 115 (3),

955–988 . Bisin, A. , Verdier, T. , 20 0 0. A model of cultural transmission, voting and political ideology. Eur. J. Polit. Econ. 16 (1), 5–29 .

Bisin, A. , Verdier, T. , 2001. The economics of cultural transmission and the dynamics of preferences. J. Econ. Theory 97 (2), 298–319 . Bisin, A. , Verdier, T. , 2011. Chapter 9 – the Economics of Cultural Transmission and Socialization. In: Handbook of Social Economics, 1. North-Holland,

pp. 339–416 . Bisin, A. , Verdier, T. , 2017. On the Joint Evolution of Culture and Institutions. NBER Working Paper No. 23375 .

Bozça ̆ga, T. , Christia, F. , 2020. Imams and Businessmen: Islamist Service Provision in Turkey. Working Paper .

Carvalho, J.-P. , Koyama, M. , 2016. Jewish emancipation and schism: economic development and religious change. J. Comp. Econ. 44 (3), 562–584 . Carvalho, J.-P. , Koyama, M. , Sacks, M. , 2017. Education, identity, and community: lessons from jewish emancipation. Public Choice 171 (1–2), 119–143 .

Carvalho, J.-P., Sacks, M., 2021. The Economics of Religious Communities: Social Integration, Discrimination and Radicalization. Working Paper doi: 10.2139/ ssrn.3297267 .

Chaney, E. , 2016. Religion and the Rise and Fall of Islamic Science. Working Paper . Chaudhary, L. , Rubin, J. , 2016. Religious identity and the provision of public goods: evidence from the indian princely states. J. Comp. Econ. 44 (3), 461–483 .

Chaudhry, A. , Garner, P. , 2006. Political competition between countries and economic growth. Rev. Dev. Econ. 10 (4), 666–682 . Chaudhry, A. , Garner, P. , 2007. Do governments suppress growth? institutions, rent-seeking, and innovation blocking in a model of schumpeterian growth.

Econ. Polit. 19 (1), 35–52 .

Co ̧s gel, M.M. , Miceli, T.J. , 2009. State and religion. J. Comp. Econ. 37 (3), 402–416 . Feldman, G. , 2004. The Disfranchisement Myth: Poor Whites and Suffrage Restriction in Alabama. University of Georgia Press, Athens, GA .

Ferraz, C. , Finan, F. , Martinez-Bravo, M. , 2020. Political Power, Elite Control, and Long-Run Development: Evidence from Brazil. NBER Working Paper No. 27456 .

Forret, J. , 2006. Race Relations at the Margins: Slaves and Poor Whites in the Antebellum Southern Countryside. LSU Press, Baton Rouge, LA . Fouka, V. , 2020. Backlash: the unintended effects of language prohibition in US schools after world war I. Rev. Econ. Stud. 87 (1), 204–239 .

Frazier, E.F. , 1949. The Negro in the United States. Macmillan Company, New York .

George, S.E. , Ponattu, D. , 2020. Like Father, Like Son? The Effect of Political Dynasties on Economic Development. Working Paper . Gill, A. , 1998. Rendering Unto Caesar: the Catholic Church and the State in Latin America. University of Chicago Press, Chicago .

Guiso, L. , Sapienza, P. , Zingales, L. , 2006. Does culture affect economic outcomes? J. Econ. Perspect. 20 (2), 23–48 . Hauk, E. , Mueller, H. , 2015. Cultural leaders and the clash of civilizations. J. Conflict Resol. 59 (3), 367–400 .

Hiller, V., Touré, N., 2021. Endogenous gender power: the two facets of empowerment. J. Dev. Econ. 149, 102596. doi: 10.1016/j.jdeveco.2020.102596 . Hoffman, P.T. , 2015. Why did Europe Conquer the World?. Princeton University Press, Princeton, NJ .

Iyigun, M. , Rubin, J. , 2017. The Ideological Roots of Institutional Change. Working Paper .

Iyigun, M. , Rubin, J. , Seror, A. , 2019. A Theory of Cultural Revivals. AMSE Working Papers . Jones, B.F., Olken, B.A., 2005. Do leaders matter? National leadership and growth since world war II. Q. J. Econ. 120 (3), 835–864. doi: 10.1093/qje/120.3.835 .

Kuran, T. , 2011. The Long Divergence: how Islamic Law Held Back the Middle East. Princeton University Press, Princeton, NJ . Logan, T.D. , 2020. Do black politicians matter? Evidence from reconstruction. J. Econ. Hist. 80 (1), 1–37 .

Marshall, R. , 1961. Industrialisation and race relations in the southern United States. In: Hunter, G. (Ed.), Industrialisation and Race Relations: a Symposium. Oxford University Press, London .

Matthews, D. , 2020. Turkey’s coup: the Gülen movement, explained. Vox .

Merritt, K.L. , 2017. Masterless Men: Poor Whites and Slavery in the Antebellum South. Cambridge University Press, New York . North, D.C. , 1981. Structure and Change in Economic History. Norton, New York .

North, D.C. , 1990. Institutions, Institutional Change and Economic Performance. Cambridge University Press, New York . Nunn, N. , 2012. Culture and the historical process. Econ. Hist. Dev. Regions 27 (sup1), S108–S126 .

Ottinger, S. , Winkler, M. , 2020. Political threat and propaganda: Evidence from the US south. Working Paper . Passarelli, F., Tabellini, G., 2017. Emotions and political unrest. J. Polit. Econ. 125 (3), 903–946. doi: 10.1086/691700 .

Prummer, A. , 2019. Religious & cultural leaders. In: Carvalho, J.-P., Iyer, S., Rubin, J. (Eds.), Advances in the Economics of Religion. Palgrave Macmillan .

Prummer, A. , Siedlarek, J.-P. , 2017. Community leaders and the preservation of cultural traits. J. Econ. Theory 168, 143–176 . Rubin, J. , 2017. Rulers, Religion, and Riches: why the West Got Rich and the Middle East Did Not. Cambridge University Press, New York .

Spolaore, E. , Wacziarg, R. , 2013. How deep are the roots of economic development? J. Econ. Lit. 51 (2), 325–369 . Squicciarini, M. , 2020. Devotion and development: Religiosity, education, and economic progress in 19th-century france. Am. Econ. Rev. 110 (11), 3454–3491 .

Tabellini, G. , 2008. The scope of cooperation: values and incentives. Q. J. Econ. 123 (3), 905–950 . Tee, C. , 2016. The Gülen Movement in Turkey: the Politics of Islam and Modernity. I. B. Tauris Publishing, London .

29

M. Iyigun, J. Rubin and A. Seror European Economic Review 135 (2021) 103734

Ticchi, D. , Verdier, T. , Vindigni, A. , 2013. Democracy, Dictatorship and the Cultural Transmission of Political Values. IZA Discussion Papers. Institute of Labor Economics (IZA) .

Tilly, C. , 1990. Coercion, Capital, and European States, AD 990–1990. Blackwell, Oxford . Verdier, T. , Zenou, Y. , 2015. The role of cultural leaders in the transmission of preferences. Econ. Lett. 136 (C), 158–161 .

Verdier, T. , Zenou, Y. , 2018. Cultural leader and the dynamics of assimilation. J. Econ. Theory 175, 374–414 . Wilson, W.J. , 1976. Class conflict and jim crow segregation in the postbellum south. Pac. Sociol. Rev. 19 (4), 431–446 .

Woodward, C.V. , 1974. The Strange Career of Jim Crow, third ed. Oxford University Press, New York .

Yao, Y. , Zhang, M. , 2015. Subnational leaders and economic growth: Evidence from chinese cities. J. Econ. Growth 20 (4), 405–436 .

30