Studio dei flussi migratori in Italia mediante analisi di autocorrelazione spaziale, di Grazia Scardaccione, Francesco Scorza, Giuseppe Las Casas, Beniamino Murgante
University of Basilicata LISUT
Faculty of Engineering Laboratorio di Ingegneria dei Sistemi Urbani e
Department of Architecture, Planning and Transport
Infrastructure Laboratory of Urban and Regional Systems Engineering
Studio dei flussi migratori in Italia
mediante analisi di correlazione spaziale.
Grazia SCARDACCIONE, Francesco SCORZA, Giuseppe LAS
CASAS, Beniamino MURGANTE
Università della Basilicata – Laoratorio di Ingegneria dei Sistemi Urbani e Territoriali (LISUT)
University of Basilicata – Laboratory of Urban and Regional Systems Engineering (LISUT),
Viale dell’Ateneo Lucano 10, 85100, Potenza - Italy
INPUT 2010 – 13, 15 Settembre 2010 - Potenza Ing. Francesco Scorza
1. Migration analysis
2. Migrants Distribution Analysis in Italy:
3. Migrants Distribution Analysis in Italy: Spatial
Interdisciplinary research field
Relevant for the interpretation of socio economic
dynamics (multi-scale interpretation)
• “migrations are forms of human capital” (Sjaastad,
L. - 1962)
• “search for better economic conditions” (wealth
maximization) (Mincer, J. - 1978)
Italy: from origin to destination of migration flows
Structural aspects of the approach:
• Main statistical unit: the Municipality
• Data time series (from 1991 to 2007)
• “simple” data and elaborations
• High transferable approach
Efficacy index of migration (Ie).
• index of dissimilarity (D)
• and location quotient (LQ).
To assess levels of territorial differentiation of a group (the foreigners)
compared to resident population.
To evaluate possible ghetto or ‘ethnic islands’ effect depending on
social segregation connected with high concentration of a single
immigrant group compared to local residents.
Efficacy index of migration
(I − D )
Ie = 100
(I + D )
I = Members (people who have moved their residence to a specific
municipalities), D = Deleted (people who have cancelled their residence from
a specific municipality), (I-D) represents “net migration”.
Values close to zero -> migration exchange produces not significant change in
values close to 100 -> the incoming flows are greater than outgoing ones;
values close to -100 -> emigration flows are prevailing
Efficacy index of migration
1. Heterogeneous behaviour of the
2. No relevant cluster identified
3. Mountain municipalities have a
marked tendency to generate
Efficacy Index of Migrations calculated for migrants in Italy
in 2007 (our elaboration on ISTAT data).
"Location Quotient" (LQ) provides an estimation of specialization degree of the
each statistical unit to accept foreign population.
(xi yi )
(X Y )
xi represents the number of residents of a national group in area unit i (in our
case the municipality),
X the number of residents in the entire study area (in our case the Country),
yi the foreign population in area unit i
Y the foreign overall population in the study region.
LQ = 1 -> the analyzed group holds in the area unit I the same characteristics of
the whole study region;
LQ > 1 -> the analyzed group is overrepresented in area unit i,
LQ < 1 -> the analyzed group is underrepresented in area unit i,
1. Greater specialization is
localized in central and north-
eastern areas of the country
Location Quotient calculated for resident immigrants in Italy in 2007
(our elaboration on ISTAT data).
Dissimilarity Index (Duncan and Duncan – 1955) provides an estimation of
the segregation degree of two groups of population in the study area. It
describes a spatial concentration of population groups.
D= i =1
x i − z i 100
xi is the ratio between the number of residents in the area i and total
population in the whole study area;
Zi represents a ratio similar to x, for another group;
k is the number of territorial parts in which we divide the study area.
D varies between 0 and 100.
Values close to 0 -> low dissimilarity.
High values of D -> coexistence of the two groups in the same areas is
1. The index of dissimilarity
allowed to measure the
heterogeneity of the structure of
2. D allows a direct comparison of
different areas, but it is not
spatially embedded and it does
not explain internal aspects of
3. Segregation indices do not
provide guidance on the spatial
distribution of the phenomenon,
in particular they do not allow to
develop assessment of
segregation degree within the
Spatial Analysis Techniques
• Moran Index (I),
• Moran scatter plots
• Local Indicator of Spatial Association (LISA).
Tobler's First Law of Geography “All things are related, but nearby things
are more related than distant things” (1970)
Positive Negative No Autocorrelation
(O’Sullivan and Unwin, 2002)
Moran’s I statistic
∑ ∑ (x − x )(x − x )w
i j ij
0 ∑ (x − x )2
Xi is the variable observed in n spatial partitions and x is the variable average;
Wij is the generic element of contiguity matrix;
n is the sum of all matrix elements defined as contiguous
S 0 = ∑ w ij
according to the distance between points-event.
In the case of spatial contiguity matrix, the sum is equal to
the number of non-null links.
Moran’s I statistic
The generalized matrix of W weight expresses the concept of contiguity
W is usually symmetrical, representing the pattern of connections or ties and
W is a dichotomic matrix of contiguity where wij = 1 if the i area touches the
boundary of j area; and wij = 0 is otherwise.
Index values may fall outside the range (-1, +1). Moreover, in case of no
autocorrelation the value is not 0 but is -1/(n-1). So if:
I < -1/(n-1) = Negative Autocorrelation,
I = -1/(n-1) = No Autocorrelation,
I > -1/(n-1) = Positive Autocorrelation.
Moran Scatter plot
GEODA allows to build Moran Scatter plot.
The graph represents the distribution of the statistical unit of analysis.
Moran Scatter plot shows the horizontal axis in the normalized variable x,
and on the normalized ordinate spatial delay of that variable (Wx).
In this representation the I° and III°
quadrants represent areas with positive
correlations (high-high, low-low) while
the II° and IV° quadrants represent
areas with negative correlation.
x = a + α Wx
Moran Scatter plot allow to generate spatial clusters of statistical units
but it doesn’t provide information on the significance of spatial
Moran’s I statistic
Moran Scatter plot for the variable Foreigners/Residents in 1999(a), 2002(b), 2004(c), 2007(d) (our elaboration with GeoDa on ISTAT data).
Moran’s I statistic
Moran Scatter plot distribution a) in 1999 and b) in 2007 (our elaboration with GeoDa on ISTAT data)
Local indicators of spatial association
∑w ij ( y i − y )( y j − y )
∑ Ii = γ .I
i − y) 2
LISA allows for each statistical unit to assess the similarity of each observation
with that of its surroundings.
Five scenarios emerge:
• Locations with high values of the phenomenon and high level of similarity with
its surroundings (high - high defined as HOT SPOTS;
• Locations with low values of the phenomenon and high level of similarity with
its surroundings (low - low), defined as COLD SPOTS;
• Locations with high values of the phenomenon and low level of similarity with
its surroundings (high - low), defined as Potential "Spatial outliers";
• Locations with low values of the phenomenon and low level of similarity with its
surroundings (low - high), defined as Potential "Spatial Outliers";
• Location devoid of significant autocorrelations.
“LISA cluster map” 1999 (our elaboration with GeoDa on ISTAT data)
“LISA cluster map” 2002 (our elaboration with GeoDa on ISTAT data)
“LISA cluster map” 2004 (our elaboration with GeoDa on ISTAT data)
“LISA cluster map” 2007 (our elaboration with GeoDa on ISTAT data)
three agglomerations emerged:
1. The first cluster included values for positive autocorrelation type high-high
increasing over the years, geographically concentrated in north-eastern
areas. Such areas are characterized by increasing levels of welfare and
therefore they express strong attraction for foreigners linked with
2. The second cluster, always of high-high type affected the central part of the
national territory and it could be explained with high levels of income and
3. The third cluster, Low-Low type, included the towns of Southern Italy and
islands, notoriously characterized by low incomes and few employment
The comparison of LISA cluster maps at different dates highlight the trend of the
Migration phenomena is one of the key issues of political and social debates.
Clustering as crutial step for effective policy development: Clusters could become
target areas for specific policies
Overcoming the traditional representation in macro regional aggregation
Uncertainty linked to the illegal component of the migration flows in Italy
to the whole study.
Regional disparities of migration could be linked with the performance of each
area: areas characterized by the same performance (high presence of foreigners
or low presence of foreigners) tend to aggregate and to expand including
Thanks for your attention