Theory and mechanisms

[rafaelchduran]

Contending hypothesis: There are two contending expectations on the effect of reelection incentives on decentralization:

1. Wash hands-off security provision because they will be held accountable.
2. Hold the bull by the horns.

My results point to 2>1.

The question is whether 2 implies:

1. Credit clame
2. Citizens security demands
3. Political isolation
4. Fiscal isolation

5. All of the above

   To prove credit claim has an effect:

• [X] Alignment test. Ley (2017) finds that citizens only hold mayors accountable if they are politically aligned with the governor or president.

But alignment also means they are receiving more perks from upper administrative levels [Dell (2015)]. It is easier to do clientelism. These perks are fiscal transfer perks but not security ones because they are resigning from those by not signing security cooperation agreements.

So alignment means two different stories:

1. Transfers from the top
2. Accountability from the bottom

These leads to different mechanisms:

On mechanisms: There are two potential mechanisms driving incumbents behavior:

• [ ] Substitution between having federal forces but receiving alignment transfers --> fiscally isolates mayors from security provision
• [ ] Changes in the likelihood of winning office, i.e. incumbency advantage --> electorally isolates mayors from security provision

Answers point to both occurring at the same time.

On incumbency advantage: Three potential mechanisms:

• [ ] Clientelistic machinery-based incumbency advantage a la Fergusson et al (2019). Reelection incentives lead mayors to use the whole apparatus. If we had weak parties this makes a lot of sense because with reelection incentives you suddenly want to support your party (Klasnja and Titiunik 2018). With strong parties there is discipline, this is not a big difference. So then it maybe the case that you are replying to citizens. To test I need one thing:

  1. [] Compare parties with and without a clientelistic competitive advantage

• [ ] Quality-based incumbency advantage a la Eggers (2017). Reelection incentives lead mayors to use the whole apparatus. To test I need two things: -1. [X] Compare term-limited and non-term limited incumbents -2. [] Compare term-limited incumbent and challenger and non-term-limited incumbent and challenger. I could use those that drop off even though they could reelect.

• [ ] Information-based incumbency advantage a la BDM et a. (2020). Reelection incentives lead mayors to use the whole apparatus. To test I need one thing: --Survey data on what citizens know about security provision and how do actors behave. I can test this using the ENVIPE surveys.

Data to get

• [ ] Candidates lists from INE.

[rafaelchduran]

Short term to-dos:

• [x] check how acuerdo and acuerdo2 differ
• [x] check how governor_alignment was constructed
• [ ] run new interactions with new alignment variables
• [ ] run regressions controlling for envipe stuff pre treatment. To control for asigning blame.
• [ ] read Chaisemartin to understand the placebos
• [ ] read Sandra Ley (2017) paper on how citizens hold mayors accountable, and what happens with alignment
• [ ] request candidate lists from INE

[rafaelchduran]

On acuerdo vs acuerdo 2

• The difference comes from using either acuerdo_gobestatal or acuerdo_gobestatal2 for the years different from 2015, 2017 and 2018.
• acuerdo_gobestatal is a dummy that comes from the question CUEN_CON that is a dummy of "Clasificación de los municipios de acuerdo con la existencia de convenios o acuerdos para que el Gobierno del Estado se haga cargo de manera temporal o permanente de la prestación de servicios de seguridad pública, al momento de la aplicación del cuestionario (abril-octubre 2011)."
• acuerdo_gobestatal2 is a dummy that comes from the rowtotal of motives of whether the municipio stated that they have a security cooperation agreeement. The question is MOT_CON and refers to: "Clasificación de los municipios de acuerdo con las razones que motivaron la celebración de convenios o acuerdos para que el Gobierno del Estado se haga cargo de manera temporal o permanente de la prestación de servicios de seguridad pública, al momento de la aplicación del cuestionario (abril-octubre 2011)." Here we consider no aplica as =0 and the existence of at least another motive to have the dummy acuerdo_gobestatal2 =1.
• The difference between acuerdo and and acuerdo2 is that the latter uses acuerdo_gobestatal2 for every year except 2015, 2017 and 2018, while the former uses acuerdo_gobestatal. The correlation between both variables is 0.97.

In terms of results, acuerdo looks better than acuerdo2 for most regressions. I use acuerdo for main estimates and can use acuerdo2 as a robustness test.

[rafaelchduran]

On governor_alignment variable:

• alignment_executive_strong is a dummy of whether the mayor's incumbent is aligned with the President.
• president_alignment is a dummy of whether the municipality is priista and so is the President.
• governor_alignment is a dummy variable of whether the governor was from the PRI. So this variable is measuring something different from what I expected. Is not aligned with the mayor but whether the governor was in the PRI.
• alignment_governor_strong is a dummy of whether the mayor's party is the same as the governor's party
• alignment_governor_weak is wrong. Currently its whether the mayor's party is the same as the governor's party and if the mayor's party is the same as PRI, PVEM, PRD, PAR, and if the governor is from the PRI. This one doesn't make sense.
• double_alignment if there is alignment with the president and the governor. This one is right.

So the conclusion is the following:

1. alignment with the president by mayors = alignment_executive_strong (CREDIT CLAME & TRANSFERS)
2. alignment with the governor by mayors = alignment_governor_strong (CREDIT CLAME & TRANSFERS)
3. alignment with both president and governor by mayors = double_alignment (CREDIT CLAME & TRANSFERS)
4. subset of alignment of PRI mayors with PRI president = president_alignment (FEDERAL CLIENTELISTIC TRANSFERS vs non clientelistic parties (all other but not the PRI))
5. subset of alignment of PRI mayors with PRI president = governor_alignment2 (STATE CLIENTELISTIC TRANSFERS)
6. if governor is from the PRI = governor_alignment (NOT SURE) and now named governor_pri
7. governor_alignment_2: into three dummies to get alignment by pri and not.

Maybe I can do the difference between credit clame & transfers and transfers to get credit clame.

[rafaelchduran]

On governor_alignment variable:

• alignment_executive_strong is a dummy of whether the mayor's incumbent is aligned with the President.
• president_alignment is a dummy of whether the municipality is priista and so is the President.
• governor_alignment is a dummy variable of whether the governor was from the PRI. So this variable is measuring something different from what I expected. Is not aligned with the mayor but whether the governor was in the PRI.
• alignment_governor_strong is a dummy of whether the mayor's party is the same as the governor's party
• alignment_governor_weak is wrong. Currently its whether the mayor's party is the same as the governor's party and if the mayor's party is the same as PRI, PVEM, PRD, PAR, and if the governor is from the PRI. This one doesn't make sense.
• double_alignment if there is alignment with the president and the governor. This one is right.

So the conclusion is the following:

1. alignment with the president by mayors = alignment_executive_strong
2. alignment with the governor by mayors = alignment_governor_strong
3. alignment with both president and governor by mayors = double_alignment
4. subset of alignment of PRI mayors with PRI president = president_alignment
5. if governor is from the PRI = governor_alignment

[rafaelchduran]

After seeing that governor pri has a negative effect, it wold be good to separate acuerdos between president and municipality and governor and municipality

[rafaelchduran]

On Ley (2017)

• RQ: can voters hold elected officials accountable for crime and insecurity?
• Results: they can but only in narrow circumstances:
  1. rising levels of organized crime against public figures (crime issue voting): divides safe and insecure places. electoral support for incumbent party candidates varies inversely with local organized crime-related violence, regardless of their partisan alignment.
  2. political alignment (responsible party model): as with government cohesion, alignment conveys information. When one party controls different branches of government, citizens administer sanctions more easily (Leyden and Borelli 1995). Its easier to blame one party. Thus, partisanship acts as a "cognitive shortcut".
  3. political alignment with hawkish party: when party attempts to "own" crime. The effect is larger. The issue Sandra is facing is that she only focuses on the Calderon time period.

Notes

• by alignment, it referes to alignment with the President.
• by controlling by the levels of violence I can only see the effect of partisanship as an information shortcut.
• however, alignment also carries a story of perks from the top (particularistic goods)
• see variables in Envipe of who do citizens assign reponsibility or attribution for the security situation in Mexico
• perceptions of insecurity do mediate the results.

[rafaelchduran]

Notes on empirics

• [ ] Condition on vote share to rule out the incumbency advantage story.
• [ ] Use the log on treatment and acuerdo2. Things look better with acuerdo than acuerdo2, and with IHS rather than logs. I will use logs because the interpretation is straight forward and the IHS transformation shouldn´t be used the min of the variable is lower than 10.

[rafaelchduran]

On Decentralization literature:

1. Besley and Coate (2003)

   • standard approach to decentralized public goods comes from Oates (1972): Oates decentralization theorem: without spillovers, a decentralized system will be preferred. If not there is a tradeoff that depends on the heterogeneity of tastes and the degree of spillovers.
   • In this model centralized systems choose a uniform level of public spending for each district
   • the drawback in decentralized systems is that local governments will neglect benefits going to other districts and thus local public goods will be underprovided in the presence of spillovers
   • the drawback of a centralized system is that it produces a one size fits all outcome.
   • but uniformity in centralized systems is a bad assumption: empirically it doesn't happen (when in my case I condition on violence it may). Moreover, the government can differentiate levels of public goods according to district heterogeneity in tastes.
   • this paper relax such assumption
   • argue that the costs of local public spending are shared in a centralized system, a conflict of interests between citizens in different districts arises. This will be played out in a legislature
   • I'm not assessing whether centralization is better or worse than decentralization: its obvious that the former is better. I can see this in homicides.
   • Crime creates many spillovers and there can be wide variation in preference heterogeneity.
   • Higher spillovers always makes centralization better.
   • If districts (say my muns) are not identical, then a decentralized system produces a higher level of surplus when spillovers are sufficiently small, while centralization produces a higher level if spillovers are large.
   • by making my results conditional on violence and preferences then something else is left.
   • in Mexico's case, the allocation of public security forces depends entirely on the president not a legislature. So there is no minimum winning coalition of representatives creating a misallocation problem. But still suffers from an information problem of what happens on the ground so could allocate things wrongly.
   • I resemble more the book on Proxy Wars then. But in that case the decision is not from the top but from the bottom.

2. Reinikka and Svensson (2004). not useful.

3. Treisman (2002).

   • decentralization: makes government more honest and efficient by bringing officlas "closer to the people", forcing them to compet for mobile capital, and facilitating the satisfaction of diverse local tastes
   • but create coordination problems, and exacerbate incentives to predate and shirk on public good provision
   • there are 5 types of decentralization. Two of those are relevant for my case study: structural decentralization (number of tiers of government) and decision decentralization (focuses on the scope of issues on which subnational governments can decide autonomously). Mexico holds structural and decision decentralization in the provision of public security. At the end, the municipality is the one in charge even if higher coorporations exist. That is why, the constitutional reform to create 32 state-level security agencies was banned as unconstitutional.
   • This generates, in the words of Triesman, a "vertical competition within the State". Here there are two arguments

4. Vertical competition should decrease the quality of government: level of bribes extracted in return for regulatory leniency will be higher than if just one government had the authority. There is "overgrazing" of the bribe base, and higher corruption (Shleifer and Vishny, 1993). This is my case maybe. Second, if independent levels share the responsibility for providing a particular good, the level of proviison will be lower than if a single government (or coordinated group of them) had this responsibility. (see the model description and similarity to competitive taxation) -Overall, the level of corruption will be higher and the level or quality of provision of public services will be lower in countries with a large number of tiers of government, and especially those which have relatively more autonomous subnational governments with regulatory authority.... "Some parts of the state might be particularly vulnerable to overgrazing problems. For instance, if both a national police force and subnational police forces have authority to enforce laws, both will have opportunities to extort bribes from the same citizen. In some countries -for instance, Canada, Switzerland, the UK and Japan- separate police forces exist under the control o state or local governments."... contradictions between the laws of the different levels of state create opportunities for corrupt officials to extract bribes.

   • so by choosing decentralization you allow your police force and the military to extract rents. Particularistic good to its bureaucracy.

5. Vertical competition should increase the quality of government: If two levels of government produce the same public good or service, voters can use the performance of each as a benchmark to judge the efficiency of the other (Salmon 1987; Breton 1996). A level of government that provides the good or service less efficiently will be punished by the voters. This would suggest that -so long as governments at all levels are subject to electoral accountability and the particular contributions of each government to public provision are clear to the voters [this is clearer when there is alignment]- the effectiveness and honesty of government should be greater when more than one level of government has responsibility for providing the same public good.

• Local governments are more corrupt and prone to capture: local governments and officials may be more susceptible to corruption than their central counterparts (Tanzi 1995; Prud'homme 1995). At more decentralizaed levels, the intimacy and frequency of interaction between private individuals and officials are likely to be greater, reducing the risk of corrupt collusion. Similar to the findings in italy on infrastructure collusion. The local press are less professional and easier to buy than the national press. Interest groups might be more cohesive at the local level, leading to greater state capture and the substitution of private transfers for public services (Bardhan and Mookherjee 2000).

So my story

• In a setting with vertical competition within the state: not clear if decentralization is better to tackle security provision because of the two options of vertical competition above. Moreover, we don't know how reelection incentives affects this decision making.
• On the first question, I find that decentralization led to public good underprovision. Is this evidence of overgrazing? Not there yetl
• On the second question, I want to see how reelection incentives modify this.
• There are overgrazing problems and this is tied to the reelection story.
• my story is this: reelection increases electoral accountability; but it also leads to create particularistic things to deviate and benefit certain clienteles or forbear. Forbearance as redistribution. -I have (1) with the addition that I'm analyzing what happens when reelection incentives are introduced. Who buys politicians in the form of bribes or threats? DTOs.
• As Treisman (2002) I can provide evidence of the overgrazing theory: higher bribes and corruption when facing subnational police forces or higher decentralization.
• be being decentralized there is potential for overgrazing benefitting (1) criminal groups (capture) and (2) police forces (redistribution by forbearance). These are the two stories now. How to separate them? See if places with DTO presence are less likely to sign this agreements. Places with DTO presence decrease the likelihood of signing even more.
• the big question is the following: why does reelection incentives lead to greater capture? Those with longer term horizon fall?

Questions on the form of decentralization: who initiaties? am I observing that fewer agreements are offered or demanded? Seems to be the demand since governors were moving forward in creating the 32 state-level security apparatus.

[rafaelchduran]

Issues to solve

1. have cartel presence effects as well as alignment ones. How to reconcile?
2. theoretically, why do reelection incentives increase capture? Is it because of a story similar to the one in Italy? Corruption and capture? (so overgrazing and capture)

[rafaelchduran]

Motivation:

1. Centralized vs decentralized systems to fight crime
2. Both have benefits and consequences
3. Naive notion that centralization counteracts violence.
4. How are these affected by reelection incentives?
5. Reelection incentives led local incumbents to choose decentralization to fight crime.
6. Why? Allows for rents and capture: the SQ. 6.1 Het effects by cartel presence (conditional and not conditional on violence) 6.2 Higher police detentions but no military activity (the former cannot tackle crime, only the latter; so detentions are a measure of overgrazing. E.g. barbies. Forbearance as redistribution)
7. Not explained by: 7.1 het preferences 7.2 spillover effects 7.3 electoral advantages
8. Welfare consequences: higher violence and a captured state.
9. Electoral consequences: incumbency advantage (this strategy pays off or explains the behavior at hand)
10. Conclusion -evidence of the overgrazing (bribes and corruption) theory in decentralized systems (similar to the cross-country analysis by Tresiman 2002, or Coviello et al 2017) -reelection did not yield citizens preferences but those of the dominant party -reelection increased state capture -reelection increased violence -local electoral incentives modify the strategies to fight crime when they have a big say on policy decision making.

Mexico´s scope conditions

1. Interesting case with first time reelection, staggered way that we exploit
2. Vertical competition within the state to provide public security (Treisman 2002)
3. Decentralization decision is made by local actors.
4. Ongoing criminal wars with spatial and temporal variation.

Big issues to solve: -[ ] differences by political alignment. Separate alignment in general with alignment with the PRI. -[ ] addition of pretreatment covariates

Notes Decentralization allows for rents and local capture. Everyone (cops and DTOs) wins except the citizen. So reelection reform led to this. and this is party-led. My measure of rents could be detenidos. Or if there is no effect, I can use it to tease out whether signing a security cooperation agreement is due to capture. So reelection incentives led to go back to their SQ of rents and no homicides. It doesn't affect homicides surprisingly. But it would be good if it increased it.

[rafaelchduran]

NOTES FROM STATA: Main Messages

1. Sec. Coop. Agreements decrease violence

2. Reform decreases the likelihood of signing sec. coop. agreements. Things look better with acuerdo than acuerdo2, and with IHS rather than logs. I will use logs because the interpretation is straight forward and the IHS transformation shouldn´t be used the min of the variable is lower than 10. This is true even when not conditioning on crime pretreatment. -Wash hands-off < hold the bull by the horns -Why? Four actors to consider: (i) citizens; (ii) incumbent; (iii) parties; (iv) DTOs.

3. The fall in likelihood is stronger when there is political alignment. But there are two stories: 3.1 Transfers due to the alignment (Alignment literature, Dell (2015)). = capacity isolation 3.2 Blame/credit clame from/to citizens (Sandra Ley 2017 on Mexico). = accountability

4. story of capture! interact by pretreatment DTO presence

TO TEST

1. condition on vote share. things look the same.

2. condition on alignment. things look the same.

3. condition on citizens security demands: not a story of accountability or credit blame. So left the capacity isolation hypothesis.

4. If the effect is only seen with PRI alignment, then is going to the old model of local capture by DTOs and allow rents for police forces.

5. Need to rule out the effect of the incumbency advantage.

6. condition on citizens preferences: this can help me separate from the transfers and credit blame

7. get data on deaths of PRI politicians. This ones should sign less.

ANOTHER MECHANISM IS INCUMBENCY ADVANTAGE: If incumbency disadvantage then this would be negative. But I find there is an inc. advantage. Now why is there an inc. advantage: 3.2.a. Clientelistic machinery-based incumbency advantage a la Fergusson et al (2019). Reelection incentives lead mayors to use the whole apparatus. 3.2.b. Quality-based incumbency advantage a la Eggers (2017) 3.2.c. Information-based incumbency advantage a la BDM et a. (2020).

**IT'S WRONG FROM HERE ON:

4. Het effects of alignment with governor PRI (negative and significant) and president ALGINED (positive and significant)

5. Check type of agreements: only has this for control municipalities which is weird

Notes

1. check how acuerdo and acuerdo2 differ - done, use acuerdo
2. check how governor_alignment was constructed
3. read Chaisemartin to understand the placebos
4. read Sandra Ley paper on how citizens hold mayor accountables, and what happens with alignment