In bwv988/eusocialr: Load, filter, plot, and analyze unemployment data from the Eurostat database

Introduction

Eurostat -- the European central statistics office -- collates, stores, analyzes, and makes available a large range of statistical measures collected by the individual EU member countries' statistical authorities.

As such, it plays a key role in building, as well as maintaining historical records, achieving comparability of statistical results across countries, and in providing a sound basis for increasingly data-driven policy decisions made at a European level.

Most data is made accessible online via the Eurostat website. An R package exists which provides some basic functions to retrieve and work with data from Eurostat: eurostat.

Statistics provided by Eurostat are grouped in a number of different themes, e.g. Transport, Economy, Environment, etc.

The present package -- eusocialr -- works with socially themed data, particular data on unemployment. As such, it augments eurostat's functionality in several ways:

• Providing functions for simplified filtering of Eurostat data in order to be able to summarize by age group, gender, country, time
• Internal data handling and conversion
• Geo-spatial and time series plots
• ARIMA / ARFIMA time series modeling to forecast unemployment data

Data

Presently, eusocialr is largely designed to work with statistics that impact on society, more specifically data on unemployment in the EU.

For this purpose, the following two Eurostat tables are of deeper interest:

• lfst_r_lfu3rt -- Percentage of unemployment rates by sex, age, and NUTS 2013 regions -- yearly frequency.

• ei_lmhu_m -- Harmonized unemployment by 1000 people -- monthly frequency.

However, the package's functions are reasonably generic and could be extended to work with other, similarly structured tables.

Installation

The package may be installed as .tar.gz file, or, provided the devtools package is available, from GitHub directly:

install.packages("devtools")
devtools::install_github("bwv988/eusocialr")

1. Loading the package

Once the package is successfully installed, we can go ahead to retrieve and analyze unemployment data from Eurostat.

For the subsequent examples, let's take a look at percentage of unemployment broken down by NUTS 2013 administrative areas (see links section for a definition):

library(eusocialr)

# Install the below from the tidyverse, if needed.
library(magrittr)
library(dplyr)

# Load data.
load_eurostat_data()
get_eurostat_data() %>% head

2. Exploring Eurostat data

Example 1

We might be interested to learn what the average unemployment percentage figure was in the UK in 2016, for both genders. The answer can be found as shown below:

# All the NUTS areas.
all.nuts = get_eurostat_data() %>% select(geo)

# Filter out only UK areas.
uk.nuts = all.nuts[grepl("^UK*", all.nuts$geo), ]$geo

# Get the average.
uk.unemp.avg = round(get_eurostat_data() %>% 
      filter_eurostat_data(geo.s = as.array(uk.nuts)) %>% 
      filter(time > as.Date("2015-01-01")) %>%
      pull(values) %>%
      na.omit %>% 
      mean, 2)

cat("UK average unemployment (percentage) in 2015: ", uk.unemp.avg)

Example 2

Another question we could ask is: What were the top 5 regions UK with the highest percentage of unemployment males aged between 15 and 24 years, in 2016:

get_eurostat_data() %>% 
      filter_eurostat_data(age.a = "Y15-24", 
                           geo.s = as.array(uk.nuts), 
                           sex.a = "M") %>% 
      filter(time > as.Date("2015-01-01")) %>% 
      na.omit %>% 
      select(geo, values) %>%
      arrange(desc(values)) %>%
      head

Plotting data

Much more interesting than just looking at dull figures is to summarize the data graphically. The eusocialr package provides two visualization options: Coropleth map and time series plots. We'll give two examples in the next paragraphs.

Example 1: Coropleth map

Going back to the previous question, we now want to get a sense of the distribution of youth unemployment in 2016 in both genders (people aged between 15 and 25 years) at a European level, that is for all EU administrative areas.

For this purpose, we need to work with another table that contains raw dates, which is readily loaded from Eurostat in much the same manner as we did before.

The plot_nuts2013_coropleth() function then renders an interactive plot. Try hovering the mouse over different regions!

# Load the data using raw time format.
load_eurostat_data(time.format = "raw")

# Create an interactive map plot.
plot_nuts2013_coropleth(sex.a = "T", 
                        age.a = "Y15-24", 
                        time.s = 2016)

Example 2: Time series plot

In order to better understand the different trends of unemployment across countries, we can look at a time series plot that shows the evolution of unemployment over the years.

With eusocialr's filtering, we can produce such comparative plots with great ease.

Below is an example where we compare unemployment in Spain, Germany, and Ireland for people of both genders, aged 25 and above:

# Load the data.
load_eurostat_data()

# Select some countries.
selected.geos = c("ES", "DE", "IE")

# Filter the data.
compare.data = get_eurostat_data() %>% 
  filter_eurostat_data(geo.s = selected.geos)

# Render the time series plot.
plot_eu_ts(compare.data,
           title.s = "Comparison of unemployment in Europe\nfor people aged 25 and above")

Forecasting unemployment

We now wish to forecast a country's unemployment rate for the upcoming year, based on the historical Eurostat data. This can be done very simply with eusocialr's forecast_eu_unemp() function.

From a modeling perspective, we have a choice of applying ARIMA, or ARFIMA (fractionally integrated ARMA) models.

Both models have been used to predict unemployment -- see the link in the reference section -- however, since we don't have data with a very high time granularity, the model fits are generally not very useful for long-term predictions.

Model fitting

The below code shows how to fit the models and look at the residuals.

Later we can print the model summary to understand the $\mbox{AR(F)IMA}(p, d, q)$ structure.

# First we need to load the data.
load_eurostat_data(code = "ei_lmhu_m")
ireland.unemp.timeseries = get_eurostat_data(code = "ei_lmhu_m") %>%
    filter_ts_data(geo.s = "IE")

# Get results.
res.arima = forecast_eu_unemp(ireland.unemp.timeseries$values)
res.arfima = forecast_eu_unemp(ireland.unemp.timeseries$values, model.type = "arfima")

ARIMA results

res.arima

ARFIMA results

res.arfima

Plots

As mentioned before, the model fits also provide residuals. We can plot these, and it turns out the residuals are actually not too bad:

# Plot residuals.
par(mfrow = c(2, 1))

plot(res.arima$model$residuals,
    main = "ARIMA Residuals",
    ylab = "Residuals",
    col = "blue")

plot(res.arfima$model$residuals,
    main = "ARFIMA Residuals",
    ylab = "Residuals",
    col = "red")

References and further links

• Eurostat homepage: http://ec.europa.eu

• R Package eurostat: https://cran.r-project.org/web/packages/eurostat/index.html

• Devtools package: https://cran.r-project.org/web/packages/devtools/index.html

• NUTS local administrative units definition: http://ec.europa.eu/eurostat/web/nuts/local-administrative-units

• A paper on applying ARFIMA models to unemployment data: https://www.jstor.org/stable/20076284

bwv988/eusocialr documentation built on May 14, 2019, 11:15 a.m.