3.1 Hidrología
3.1.4 Acumulador de Flujo
As stressed previously, past migration affects the actual likelihood of moving and not simply on the basis of the results in terms of labour market outcomes and education. In fact, empirical work has argued that previous migration is relevant to the second generation of immigrants and the results in terms of wage and educational outcomes are related to changes in concentration of immigrants in the chosen location (Borjas, 1992; Goodwin-White, 2012).
One of the main channels by which past migration influences newer migration is the local labour market information flow that facilitates the catching up process, reducing the disadvan- tage deriving from being “non-native” (Cutillo and Ceccarelli, 2012). Taking into account that one of the determinants in location choice is the percentage of an individual’s ethnic group that is working in the same area (Bartel, 1989), I have assumed that relocation decisions after grad- uation are influenced by the percentage of people who relocate and the fact that some of these come from the same region of origin43.
For these reasons I have taken the internal migration rate between regions in 2002, the first year available. This data are provided by ISTAT and refers to registration and deregistration (of Italian and foreign individuals) numbered in local authority birth records44.
The index (“Past Relocation in 2002”) used as instrument is structured as follows:
P.R.j 2002=
Regionij
T otal number of invidual transf erred f rom regioni
i=1,...,20 (Origin45) j=1,...,20 (Destination)
The index is expressed in percentage terms and given by the ratio between of total number of individuals coming from the region i and transferred to region j on the total number of indi- viduals transferred from region i in 2002.
Table B.1 (appendix) shows the value of the index, by region of birth and destination. The higher value corresponds to the cell where region of birth and destination are the same, suggesting that the lion’s share of mobility occurs within the same region46.
I have sought to disentangle this “composition effect” using a different instrumental variable derived from data already used in a previous work (Piras, 2005).
The instrument proposed has the same structure as its predecessor but is based on the graduate migration flow (between regions), from 1990 to 1999.
43Bartel (1989) has suggested that economic conditions have relatively limited effects on destination choice
and immigrants are mainly attracted by a large concentration of earlier immigrants.
44Documents and data are downloadable from the following website:
http://dati.istat.it/Index.aspx?DataSetCode=DCIS MIGRAZIONILang=
The alternative instrument (“Past Graduates relocation”) is made up as follows:
P.G.R.ij t=
Regionij t
T otal number of individuals transf erred f rom regionit
i=1,...,20 (Origin47) j=1,...,20 (Destination) t=¯t=1990,...,1999.
The index is expressed in percentages and show how many graduates coming from region i went to region j (transfer residence).
The two instrument presented are very similar except in the compostion: in the second there are only graduates (with no distinction between M.A.s and B.A. or specification reasons for move- ment). Testing a further instrument helped me to understand the composition effect better, even though the motivation behind it is the same as explained above. Using alernative instrument, the results were very similar both for selection and outcome equations but, as shown in table B.8 (appendix), the correlation between past relocation and current workplace mobility becomes non significant and I decided to opt for the first instrument presented.
Scholarly arguments on the right variable to use in interregional mobility vary and often conflict. With a linear probability model, Devillanova (2013) tried to estimate the effect of mi- gration on over-education and used ‘housing tenure’ as instrumental variable, a variable which should affect the likelihood of moving but not the independent variable (over-education)48. Dev- illanova also discussed the problems potentially coming from using regional variables as tools for migration, supporting the idea that local variables cannot capture the individual unobservable characteristics which are at the origin of the endogeneity problem.
In a very similar work in which she tried to rule out determinants of study and work migration through a bivariate Probit model, Capuano (2009) also supported the thesis by which regional variables are not suitable as exclusion restrictions in both equations.
Internationally speaking, Venhorst and Corvers (2015) investigated the impact of inter- regional mobility on job-match quality using two dummies as instrumental variables: one indi- cating whether a graduate lives in the central economic region of the Netherlands at age sixteen, and one indicating whether the graduate had one or more parents born outside the Netherlands. The latter is an individual characteristic while the second is based on motivation closely related to the one argued for here. In fact, living in a rich region at the age of 16 affects search behav- ior, positively if we assume that there is a correlation between coming from a rich region and available information, negatively if we assume that there are better employment opportunities in rich regions.
Furthermore they also introduced a third variable, “Spatial mobility before the onset of study”, which is a similar variable to that used in this exercise. Every individual’s migration path (in terms of distance) is compared to the same migration path of very similar individuals (same graduation cohort, same home region, same field of study)49.
The example cited above and the absence in the dataset of an individual variable suitable to overcoming endogeneity, led me to choose the instrument described above.
48However, housing tenure can be correlated with income and with actual profession and this casts doubt on
this choice.
49This is the ratio of the distance moved by graduate i to the average distance traversed by his or her peer