In zumbov2/swissparl: Interface to the Webservices of the Swiss Parliament

knitr::opts_chunk$set(echo = TRUE, dpi = 300, fig.width = 12, fig.height = 8, message = F)
pacman::p_load(dplyr, ggplot2, hrbrthemes, ts_box, lubridate, ggridges, quanteda, purrr)
biz <- readRDS("biz.rds")

The Federal Council's To-Do List

The Swiss Parliament has various instruments with which Council members can provide ideas for new legislation or request information and reports on a specific topic. These so-called procedural requests are usually addressed to the Federal Council and must be answered by it.

Let's find out if we can use the data from the Webservices to determine how long the Federal Council's to-do list normally is and how good the Swiss executive is at processing it.

We download the entire table Business using the function get_data. This may take a few minutes.

biz <- swissparl::get_data("Business", Language = "DE")

The dataset contains the 46,116 businesses that the Swiss Parliament has dealt with over the past 20 years. Let's find out how often the Federal Council has handled/answered procedural requests since 1999.

biz %>%
  filter(!is.na(FederalCouncilProposal)) %>% 
  group_by(BusinessTypeName) %>% 
  count() %>% 
  arrange(desc(n)) %>% 
  knitr::kable()

Quite a workload!

Now we are interested in how good the Federal Council was at working this pile off? For this purpose, we define a function that determines at any given time how many requests the Federal Council has not yet answered.

get_unanswered_requests <- function(date, x) {

  x %>% 
    mutate(open = ifelse(date >= SubmissionDate & date < FederalCouncilProposalDate, 1, 0)) %>% 
    group_by(open) %>% 
    count() %>% 
    ungroup() %>% 
    mutate(Date = date) %>%
    filter(open == 1) %>%
    select(Date, n)

} 

Length of the To-Do List

We now apply this function to our data in a weekly rhythm (always on Sundays). Unanswered requests are no longer taken into account after one year.

# All Sundays from 2000 to today
sundays <- seq.Date(lubridate::ymd("2000-01-02"), Sys.Date(), "week")

# Dataset 
requests <- biz %>%
  filter(BusinessTypeName %in% c("Interpellation", "Motion", "Postulat")) %>% 
  mutate(
    FederalCouncilProposalDate = ifelse(
      is.na(FederalCouncilProposalDate),
      SubmissionDate + 365,
      FederalCouncilProposalDate
      )
    )

# Unanswered requests  
unanswered_requests <- purrr::map_dfr(sundays, get_unanswered_requests, x = requests)

# Plot
library(tsbox)
library(ggplot2)
library(hrbrthemes)

p <- unanswered_requests %>% 
  ts_ggplot() + 
  labs(
    title = "The Federal Council's To-Do List",
    subtitle = "Number of Unanswered Procedural Requests, Dec 1999 - Dec 2019",
    caption = "Data: Parliamentary Services of the Federal Assembly, Bern",
    y = "Unanswered Requests"
    ) +
  theme_ipsum_rc() +
  theme(panel.grid.minor = element_blank())

p

It's called oscillation! Let's see if there's a trend.

# Trend
trend <- ts_trend(unanswered_requests)

# Plot
p + geom_line(
  data = trend,
  aes(x = Date, y = n),
  size = 2,
  color = "red"
  ) +
  labs(subtitle = "Loess-Based Trend Estimation of Unanswered Procedural Requests")

Not surprisingly, there is a trend. After all, the number of requests has increased steadily over the last few years. Has this had an impact on the duration required by the Federal Council to respond to the councillors' demands?

Response Time

library(lubridate)
library(ggridges)

# Data prep
biz2 <- biz %>% 
  mutate(ResponseTime = as.numeric(FederalCouncilProposalDate - SubmissionDate)) %>% 

  # Filtering out unrealistic values
  filter(ResponseTime > 0) %>% 
  filter(ResponseTime < 365) %>% 

  # Focusing on phase with full coverage...
  filter(year(SubmissionDate) > 1999) %>% 

  # ...and the most common request types
  mutate(BusinessTypeName = ifelse(BusinessTypeName == "Einfache Anfrage", "Anfrage", BusinessTypeName)) %>% 
  filter(BusinessTypeName %in% c("Motion", "Postulat", "Interpellation", "Anfrage")) %>% 
  mutate(BusinessTypeName = factor(BusinessTypeName, levels = c("Anfrage", "Interpellation", "Postulat", "Motion"))) 

# Plot
biz2 %>% 
  ggplot(aes(x = ResponseTime, y = as.factor(SubmissionLegislativePeriod), fill = BusinessTypeName)) +
  stat_density_ridges(quantile_lines = TRUE, quantiles = 2, alpha = 0.7) +
  scale_x_continuous(limits = c(0, 180), breaks = seq(0, 180, by = 30)) +
  facet_wrap(.~BusinessTypeName) +
  labs(
    title = "The Federal Council's To-Do List",
    subtitle = "Response Time for Different Business Types, 1999-2019",
    caption = "Data: Parliamentary Services of the Federal Assembly, Bern",
    x = "Response Time in Days",
    y= "Term"
  ) +
  scale_fill_ipsum() +
  theme_ipsum_rc() +
  theme(
    legend.position = "none",
    panel.grid.minor = element_blank()
    )

The response time does not seem to have increased despite the greater workload. On the contrary, it seems that the Federal Council has become more efficient in the last two decades. What influence did this have on the length and the complexity of the answers?

Response Length

library(tokenizers)

# Data prep
biz2 %>% 
  filter(!is.na(FederalCouncilResponseText)) %>% 
  mutate(AnswerLength = lengths(tokenize_words(swissparl::clean_text(FederalCouncilResponseText)))) %>% 

  # Plot
  ggplot(aes(x = AnswerLength, y = as.factor(SubmissionLegislativePeriod), fill = BusinessTypeName)) +
  stat_density_ridges(quantile_lines = TRUE, quantiles = 2, alpha = 0.7) +
  scale_x_continuous(limits = c(0, 2000), breaks = seq(0, 2000, by = 500)) +
  facet_wrap(.~BusinessTypeName) +
  labs(
    title = "The Federal Council's To-Do List",
    subtitle = "Response Length for Different Business Types, 1999-2019",
    caption = "Data: Parliamentary Services of the Federal Assembly, Bern",
    x = "Response Length in Words",
    y = "Term"
  ) +
  scale_fill_ipsum() +
  theme_ipsum_rc() +
  theme(
    legend.position = "none",
    panel.grid.minor = element_blank()
    )

On average, the Federal Council's responses appear to have been somewhat shorter for 3 out of 4 types. However, the difference between the legislative periods is marginal.

Response Complexity

library(quanteda)

# Lexical diversity
biz2$LexicalDiversity <- biz2 %>%
  pull(FederalCouncilResponseText) %>% 
  swissparl::clean_text() %>%
  tokens("word") %>%
  textstat_lexdiv("CTTR") %>%
  pull(CTTR)

# Plot
biz2 %>% 
  ggplot(aes(x = LexicalDiversity, y = as.factor(SubmissionLegislativePeriod), fill = BusinessTypeName)) +
  stat_density_ridges(quantile_lines = TRUE, quantiles = 2, alpha = 0.7) +
  facet_wrap(.~BusinessTypeName) +
  labs(
    title = "The Federal Council's To-Do List",
    subtitle = "Lexical Diversity of Responses for Different Business Types, 1999-2019",
    caption = "Data: Parliamentary Services of the Federal Assembly, Bern",
    x = "Carroll's Corrected TTR",
    y = "Term"
    ) +
  scale_fill_ipsum() +
  theme_ipsum_rc() +
  theme(
    legend.position = "none",
    panel.grid.minor = element_blank()
    )

Same picture with the complexity of the responses. Hardly any change over the years. An in-depth analysis would now be necessary to get to the bottom of the matter.

But for now, we can say that the Federal Council seems to be dealing well with the increased number of requests and businesses.

zumbov2/swissparl documentation built on July 28, 2022, 11:34 a.m.