Quick analysis of the Italian referendum results

The 2016 Italian referendum torpedoed the constitutional reform presented by the government presided by Matteo Renzi (41). According to the final count, which includes 1.2 million votes cast overseas, the reform was rejected by almost 60% of the voters.

Three parties played a predominant role during the electoral campaign: the ruling Democraric Party (PD), leaded by the chief of government Renzi, the Five Star Movement (M5S), founded and leaded by Beppe Grillo (68), and the Lega Nord (LN), leaded by Matteo Salvini (43). The fourth Italian party, Forza Italia, for different reasons - including the health of Silvio Berlusconi (80) - played a minor role.

Geographically, opposition to the reform was widespread with the exception of some provinces in Tuscany, Emilia-Romagna and the autonomous and germanophone province of Bolzano/Bozen. Turnout was strong for a referendum, 68.5%, but with a stronger participation in the North of the country. Notably, a higher turnout seems associated with more ‘Yes’ votes, but again turnout is not homogeneously distributed across the country so it is possible that other geographically-dependent variables might have an influence.

Source: Ministero Interno (elezioni.interno.it)

Support for the PD in the 2013 general election seems to explain a lot of the referendum results. This is not surprising since the PD was the only major party to campaign in support of the reform. According to opinion polls, the PD would receive about 30% of the preferences in a general election. The remaining 10% could be explained with voters of small centrists parties who decided to vote for the constitutional reform. Nevertheless, the PD underperformed in the southern provinces (red dots in the picture).

Source: Ministero Interno (elezioni.interno.it)

More interesting is the role played by the electorates of the two major opposition parties: the M5S and the LN. The two electorates do partially overlap. Also, the two parties seem to share a similar anti-establishment rhetorical style and have few programmatic issues in common (e.g. radical opposition to EU institutions). From the 2013 electoral map it appears that the two parties play a win-lose game competing for the same pool of votes: in the northern regions, where the LN is stronger the M5S is weaker.

Source: Ministero Interno (elezioni.interno.it)

A more focused comparison on the electoral results of the two parties at the provincial level in the eight northern regions with the exclusion of the province of Bolzano/Bozen shows indeed a negative association between the results of the two parties.

Source: Ministero Interno (elezioni.interno.it)

A multivariate regression analysis on the results of the referendum can help assess the contribution of different variables. We saw that the geographic dimension is important, so I introduce as control variable the unemployment rate (source: istat.it). I’m not suggesting that the unemployment rate might have any direct influence on the referendum results but the unemployment rate (along with per capita GDP) captures effectively the difference between northern and southern regions. Additionally to the unemployment rate, I also control for turnout and for latitude of the centroid of each province.

In the first regression bloc I use data from all regions. First, turnout - but not the unemployment rate - stops being relevant as soon as I control for the votes PD received in 2013. As expected, PD and M5S votes in the 2013 general election are both significant and respectively positively and negatively associated with the ‘Yes’ vote in the referendum.

	Dependent variable:
	All regions - Yes vote in 2016 referendum (%)
	(1)	(2)	(3)	(4)	(5)
turnout (%)	0.811***	0.563***	0.568***	0.051	0.174
	(0.071)	(0.161)	(0.172)	(0.162)	(0.159)
unemployment rate (%)		-0.350*	-0.360	-0.394**	-0.401**
		(0.205)	(0.237)	(0.197)	(0.189)
latitude			-0.0004	0.007*	0.002
			(0.004)	(0.004)	(0.004)
PD votes (%, 2013)				0.563***	0.529***
				(0.083)	(0.080)
M5S votes (%, 2013)					-0.303***
					(0.093)
Constant	-0.156***	0.058	0.072	-0.038	0.199
	(0.049)	(0.134)	(0.213)	(0.178)	(0.185)
Observations	106	106	106	106	106
R2	0.560	0.572	0.572	0.706	0.734
Adjusted R2	0.556	0.564	0.559	0.694	0.721
Note:	p<0.1; p<0.05; p<0.01

By controlling for the Northern regions only, we can introduce the variable quantifying the support for the LN. We note first that the unemployment rate stops being relevant (in the northern regions unemployment varies less but still does varies between 4 and 12%) suggesting that indeed is not unemployment per se to strengthen the No vote. Second, votes to the three major parties in the 2013 general elections are all significant predictors of the referendum results.

	Dependent variable:
	Northern regions - Yes vote in 2016 referendum (%)
	(1)	(2)	(3)	(4)	(5)
turnout (%)	0.604***	0.609**	0.272*	0.145	0.212
	(0.208)	(0.229)	(0.156)	(0.148)	(0.141)
unemployment rate (%)		0.020	-0.267	-0.096	-0.194
		(0.384)	(0.254)	(0.238)	(0.227)
PD votes (%, 2013)			0.552***	0.579***	0.419***
			(0.073)	(0.066)	(0.089)
M5S votes (%, 2013)				-0.302***	-0.529***
				(0.096)	(0.127)
LN votes (%, 2013)					-0.325**
					(0.128)
Constant	-0.013	-0.019	0.106	0.252**	0.334***
	(0.153)	(0.183)	(0.121)	(0.119)	(0.117)
Observations	46	46	46	46	46
R2	0.160	0.160	0.647	0.716	0.755
Adjusted R2	0.141	0.121	0.622	0.688	0.724
Note:	p<0.1; p<0.05; p<0.01

The M5S and the LN have set the same goal to their referendum campaigns: to get rid of the current government. They succeeded also because they effectively managed to mobilise those who voted for them in 2013. The PD failed because the referendum was largely played along party lines and although it seems it did mobilise its electorate (but less successfully in the South), it could not reach beyond it.