Local participation and not unemployment explains the M5S result in the South

The abundance of economic data and the scarcity of social data with a comparable level of granularity is a problem for the quantitative analysis of social phenomena. I argue that this fundamental problem has misguided the analysis of the electoral results of the Five Star Movement (M5S) and its interpretation. In this article, I provide statistical evidence suggesting that — in the South — unemployment is not associated with the exceptional increase in the M5S support and that local participation is a stronger predictor of support than most of the demographics.

What happened

The 2018 Italian general elections (elections, since both the Chamber of Deputies and the Senate, were renewed) saw

  1. a significant increase in the number of votes for two parties, the Five Start Movement (M5S) and the League (formerly Northern League),


  1. an increase in the importance geography as an explanatory dimension for the distribution of votes.

The following two maps show where the M5S and the League have increased electoral support from 2013 to 2018. (Electoral data are always data for the election of the Chamber of Deputies).

Vote difference: 2018-2013 (a few communes have not reported all the results, notably Rome)


The geographic pattern is quite simple. The M5S has increased its support in the South and maintained its votes in the North, the League has significantly strengthened its support in the North but has also collected votes in the South, where it had virtually no support. The third and the fourth most voted parties, the Democratic Party (PD) and Berlusconi’s Forza Italia (FI), have lost votes almost everywhere. If we map the results of the four parties side-by-side with the same scale, the PD and FI almost faded into the background.

Votes in the 2018 General elections

Yet, major metropolitan areas do not always follow the national trend. If Naples unambiguously voted M5S, Turin, Milan and Rome did saw the Democratic Party as the most voted party in the wealthiest districts.

Votes in the 2018 General elections (Clock-wise from top-left: Turin, Milan, Naples, Rome)

The density of the distribution of results at the commune and sub-commune level in the macro regions indicates that if the M5S electorally dominates in the South and in the two major islands, the League is the most popular party in the North.

Distribution of votes at commune or sub-commune level

The territoriality of the results, especially along the North-South dimension, makes the analysis especially complicated. This because the strong result of the League in the North and of the M5S in the South might simplistically suggest that immigration (which is much stronger in the North) explains the League’s result in the North and unemployment and poverty (stronger in the South) explain the M5S’s result in the South. This reading is especially attractive since immigration and the M5S proposal to introduce a guaranteed minim income have dominated the campaign.


Tuesday, 20 March 2018

2018 Italian general election: Details on my simulation

This article describes the simulation behind the app that you find here

This simulation of the results for the 2018 general election is based on the results from the last two national elections (the Italian parliament election in 2013 and the European Parliament election 2014) and national polls conducted until 16 February 2018. The simulation is based on one assumption, which is reasonable but not necessarily realistic: the relative territorial strength of parties is stable. From this assumption derives that if the national support for a party (as measured by national voting intention polls) varies, it varies consistently and proportionally everywhere. A rising tide lifts all boats and vice versa. The assumption has some empirical justification. If we compare the difference from the national support (in percentage) for each district in 2013 and 2014 we see a significant correlation, especially in the major parties.

Votes to party in the 2018 Chamber districts


Tuesday, 27 February 2018


Twitter: frbailo



RSS r-bloggers.com

  • Introducing scale model in greybox
    At the end of June 2021, I released the greybox package version 1.0.0. This was a major release, introducing new functionality, but I did not have time to write a separate post about it because of the teaching and lack of free time. Finally, Christmas has arrived, and I could spend several ... Continue reading: […]
  • Plotting Bee Colony Observations and Distributions using {ggbeeswarm} and {geomtextpath}
    Setup Loading the R libraries and data set. # Loading libraries library(geomtextpath) # For adding text to ggplot2 curves library(tidytuesdayR) # For loading data set library(ggbeeswarm) # For creating a beeswarm plot library(tidyverse) # For the gg... Continue reading: Plotting Bee Colony Observations and Distributions using {ggbeeswarm} and {geomtextpath}
  • Non-linear model of serial dilutions with Stan
    In chapter 17 “Parametric nonlinear models” of Bayesian Data Analysis1 by Gelman et al., the authors present an example of fitting a curve to a serial dilution standard curve and using it to estimate unknown concentrations. Below, I build t... Continue reading: Non-linear model of serial dilutions with Stan
  • Predicting future recessions
    Even if this sounds incredible, yes, we can predict future recessions using a couple of time series, some simple econometric models, and … R !  The basic idea is that the slope of the yield curve is somewhat linked to the probability of future recessions. In other words, the difference between the ... Continue reading: […]
  • Detecting multicollinearity — it’s not that easy sometimes
    By Huey Fern Tay with Greg Page When are two variables too related to one another to be used together in a linear regression model? Should the maximum acceptable correlation be 0.7? Or should the rule of thumb be 0.8? There is actually no single, ‘one-size-fits-all’ answer to this question. As an ... Continue reading: Detecting multicollinearity […]

RSS Simply Statistics

RSS Statistical Modeling, Causal Inference, and Social Science