BeamerPresentation_MalteJonas.Rmd

---
title: "Political Economy of Public Banks"
author: "Jonas Markgraf & Malte Berneaud-Kötz"
date: |
  | Hertie School of Governance
  | 02 May 2016
output:
  beamer_presentation:
    colortheme: dolphin
    fonttheme: structuresmallcapsserif
    slide_level: 2
    theme: Szeged
  ioslides_presentation: default
subtitle: Electoral Effects of Savings Banks in German Subnational Elections
---

```{r setting_chunk_options, include=FALSE}
############### NOTE ###############
# In order for the RMarkdown file to render correctly, you need to place the source data sets into the data/ folder in the root directory of this repo. 

setwd("~/Git/PresentationAssignment")  # Adjust this in your file to match the directory where you cloned the repo. 

knitr::opts_chunk$set(fig.height=3, fig.width = 5, fig.align = "center", fig.path='graphs/',
                      include = FALSE, warning=FALSE, message=FALSE, cache=TRUE)
# Supressed output for any code chunk in the document, so all graphs, tables,
# etc which need to have the outputs shown should use include = TRUE as an option
```

```{r loading_preparing_data}
# Loading of libraries, cleaning and subsetting code is done in the external R-script
source("standalone_R_code.R")
```


# Background
## Motivation

- financial crisis 2007-8: increased interest in bank-politics interface
- politicians' participation in public banks understudied
- public banks are central element of Europe's banking sector

## Institutional Setting

- large number of public savings banks (*Sparkassen*) in Germany
- *Sparkassen* important providers of credit and public goods
- regional outlook of savings banks & close ties to local politics

## Sparkassen business areas

![](SavingsBank_map.png)


# Hypothesis
## Hypothesis

- mayors with board seat in savings bank extract political rents from board membership

 > **Hypothesis:**

 > *Mayors with a board seat in a Sparkasse are more likely to be re-elected than mayors without a boardseat, holding all other explanatory variables constant.*

## Storyline

How does membership in a *Sparkasse* affect electoral success?

* direct channel: influence on strategic decisions of bank
     + credit volume
     + timing of consolidation and mergers
     + donations & payouts

* indirect channel: competence signal to constituency

# Empirical Strategy & Findings | Model & Controls

## Data & Subset
### Data
  1. Data on all Bavarian mayoral elections conducted 1948-2014 were obtained from the Bavarian Statistical Office

  2. Hand collected dataset on board members in Bavaria's *Sparkassen* for 2006-2015

    
Data is matched on names to find mayors who were members of a *Sparkassen* governing board  

### Subset

  + 2006 to 2014
  + competitive elections
  + no run-offs


## Statistical Model

Logistic regression to deal with binary dependent variable

  + DV: Re-election of mayor yes / no
  + main IV: *Sparkassen* board membership in year prior to election
  + control IVs:
    + vote share received by the last winning mayor
    + gender
    + municipality size proxy: log(number of eligible voters)

## Regression Results
```{r regression models}
fullmodel <- glm(Reelection~IncumbentSparkassenMember + L.VoteShareWinner + L.Geschlecht1 + log(NumberEligVoter),
  data = MayorElection,
  family = "binomial")
covarnames <- c("Sparkassen board member", "Previous vote share", "Female", "Number eligible voters")
```

```{r tests, include=TRUE, echo=FALSE, results="asis"}
stargazer(fullmodel, title = "Logistic regression results w/ odds ratios", dep.var.labels = "Re-election of mayor", 
          covariate.labels = covarnames, header = FALSE, font.size = "scriptsize", single.row = TRUE,
          apply.coef = exp, t.auto = F, p.auto = F, report = "vcp*", digits = 2, keep.stat = c("n", "ll", "rsq"))
## Note: coefficients are reported as odds ratios, but t-scores and p-values are calculated using the underlying logit function, coefficients, and standard errors
```

## A closer look at previous vote share
```{r simulation_for_vote_share}
# Creating dummy from lagged sex variable
MayorElection$L.DummySex <- ifelse(MayorElection$L.Geschlecht1 == "m", 0, 1)

# Find range of L.VoteShare1 in data
vs_range <- 20:97
# extract coefficients and variance-covariance matrix from model
fm_coef <- matrix(coef(fullmodel))
fm_vcov <- vcov(fullmodel)

# drawing simulations of the point estimates
drawn <- data.frame(mvrnorm(n = 1000, mu = fm_coef, Sigma = fm_vcov))

# matching simulations to all possible values in L.VoteShare1
draw_sim <- merge(drawn, vs_range)
draw_sim <- dplyr::rename(draw_sim, fitted.vs = y)

# Calculating predicted probabilites with the data at hand
# I am using the averages of all covariates except vote share
draw_sim$pred.prob <- 1 / (1 + exp(-(draw_sim[, 1] + draw_sim[, 2] + draw_sim[, 3] * draw_sim[, 6] + draw_sim[, 4] * mean(MayorElection$L.DummySex) + draw_sim[, 5] * mean(log(MayorElection$NumberEligVoter), na.rm = TRUE))))
```

```{r graphing_simulation, include=TRUE, echo=FALSE}
ggplot(draw_sim, aes(x = fitted.vs, y = pred.prob)) + geom_point(alpha = 0.1) +
  stat_smooth(se = FALSE) + theme_bw() + ggtitle("Probability of Re-election across Previous Vote Shares") + 
  ylab("Probability of Re-election") + xlab("Vote Share in Previous Election")
```

----

Thank you for your attention!

https://github.com/jmarkgraf/PresentationAssignment