Measures of the integration of foreign migrants in Lombardia: some new experiences Marta Blangiardo 1 and Gianluca Baio 2 1 Department of Epidemiology, Public Health and Primary Care, Imperial College London (UK) Fondazione ISMU Milan (Italy) email: m.blangiardo@imperial.ac.uk 2 Department of Statistical Science, University College London (UK) Department of Statistics, University of Milan Bicocca (Italy) email: gianluca@stats.ucl.ac.uk BSPS Annual Conference, Canterbury 13 September, 2005 p. 1
Immigration in Italy 1.1.2001 1.1.2002 1.7.2003 1.7.2004 Italy Stay permits (thousands) 1380 1448 1503(a) 2194(a) Lombardia Estimated migrants from Less Developed or Eastern European Countries (thousands) Legal residents 292 293 366 473 Legal non residents 41 28 130 81 Total registered migrants 333 321 496 554 Illegal migrants (minimum) 72 111 52 79 Illegal migrants (maximum) 102 182 71 108 Total (mimimum) 405 432 548 633 Total (maximum) 435 503 567 662 (a) Datum at 1 January Source: Fondazione ISMU a) High growth nationwide (only 649k permits at 1 January 1991) b) High incidence of Lombardia (around 25% of total) p. 2
Data sources Although the problem is relevant, data sources are still poor in Italy p. 3
Data sources Although the problem is relevant, data sources are still poor in Italy Since 2001, Fondazione ISMU (Milan, Italy) started to collect data on migrants integration level for the Lombardia Region Probabilistic sampling based on Blangiardo G.C. (2004) from the universe of foreign people coming from Less Developed or Eastern Europe Countries and present in Lombardia by the time of the survey Annual sample size 8000 units Geographical stratification (12 territorial units); each year 350 different first stage units Survey on > 14 year old people, data collected on demographic, social, economic & health characteristics p. 3
Objective Outline of the work To estimate the level of integration of migrants, given some relevant observed characteristics Integration defined according legal status, stability, availability of basic needs and opportunities (house, job) To evaluate different profiles of migrants integration in terms of suitable post stratification variables p. 4
Objective Outline of the work To estimate the level of integration of migrants, given some relevant observed characteristics Integration defined according legal status, stability, availability of basic needs and opportunities (house, job) To evaluate different profiles of migrants integration in terms of suitable post stratification variables Problems Level of integration is unobservable Hence standard regression-like models not applicable p. 4
Objective Outline of the work To estimate the level of integration of migrants, given some relevant observed characteristics Integration defined according legal status, stability, availability of basic needs and opportunities (house, job) To evaluate different profiles of migrants integration in terms of suitable post stratification variables Problems Level of integration is unobservable Hence standard regression-like models not applicable A possible (good) solution Bayesian Latent Class (LC) models Useful to consider a graphical representation of the model p. 4
Graphical representation Integration Level High Medium Low Legal Status Population Registry Housing Arrangement Work Status A Stay permit B Stay permit Illegal Registered Not registered Private household Sharing Irregular Legal worker Inactive Illegal Unemployed p. 5
Graphical representation Integration Level High Medium Low Legal Status Population Registry Housing Arrangement Work Status A Stay permit B Stay permit Illegal Registered Not registered This graphical model encodes the following assumptions: Private household Sharing Irregular Legal worker Inactive Illegal Unemployed 1. The probability distribution of each of the four characteristics that we investigate depends on the probability distribution of the Integration Level A direct link from the variable Integration Level towards each characteristic p. 5
Graphical representation Integration Level High Medium Low Legal Status Population Registry Housing Arrangement Work Status A Stay permit B Stay permit Illegal Registered Not registered This graphical model encodes the following assumptions: Private household Sharing Irregular Legal worker Inactive Illegal Unemployed 1. The probability distribution of each of the four characteristics that we investigate depends on the probability distribution of the Integration Level A direct link from the variable Integration Level towards each characteristic 2. The four characteristics are independent on one another given the Integration Level No link connects the four characteristics directly; they are only connected through the variable Integration Level p. 5
Data preprocessing For each individual we can observe a profile P = [LS = i, PR = j, HA = k, WS = h] In theory, we could observe in total 3 2 3 4 = 72 different profiles p. 6
For each individual we can observe a profile P = [LS = i, PR = j, HA = k, WS = h] Data preprocessing In theory, we could observe in total 3 2 3 4 = 72 different profiles However, some are not admissible, i.e.: P na = [LS = Illegal, PR = Registered, HA = Sharing, WS = Inactive] (an illegal migrant cannot be registered in the Population Registry) Consequently, the total number of admissible profiles that can be observed is 37 p. 6
Estimation procedure We are interested in the integration level of each profile: Pr(IL = q P), q = low, medium, high Using Bayes theorem we have that: Pr(IL = q P) Pr(P IL = q) Pr(IL = q) p. 7
Estimation procedure We are interested in the integration level of each profile: Pr(IL = q P), q = low, medium, high Using Bayes theorem we have that: Pr(IL = q P) Pr(P IL = q) Pr(IL = q) Hence we need to: 1. Provide a prior distribution for the integration level: Pr(IL = q) 2. Estimate the likelihood of the single profiles, given the value of the Integration level: Pr(P IL = q) p. 7
Estimation procedure (cont d) From the graphical structure we imposed, we have that: Pr(P IL = q) = Pr(LS = i, PR = j, HA = k, WS = h IL = q) = Pr(LS = i IL = q) Pr(PR = j IL = q) Pr(HA = k IL = q) Pr(WS = h IL = q) Each of these conditional distribution is given a prior Dirichlet distribution, and is estimated from the observed data on the profiles P, via MCMC p. 8
Estimation procedure (cont d) From the graphical structure we imposed, we have that: Pr(P IL = q) = Pr(LS = i, PR = j, HA = k, WS = h IL = q) = Pr(LS = i IL = q) Pr(PR = j IL = q) Pr(HA = k IL = q) Pr(WS = h IL = q) Each of these conditional distribution is given a prior Dirichlet distribution, and is estimated from the observed data on the profiles P, via MCMC As for the prior distributions on the Integration levels, we used: 1. A non informative distribution for the first year of observation (2001): Pr(IL = low) = Pr(IL = medium) = Pr(IL = high) = 1 3 2. The posterior distribution obtained is then used as a prior for the following year, to incorporate time correlation for 2002-2004 p. 8
In synthesis... Pr(IL = q P) 2001 Pr(P IL = q) 2001 Pr(IL = q) 2001 p. 9
In synthesis... Pr(IL = q P) 2001 Pr(P IL = q) 2001 Pr(IL = q) 2001 Pr(IL = q P) 2002 Pr(P IL = q) 2002 Pr(IL = q) 2002 p. 9
In synthesis... Pr(IL = q P) 2001 Pr(P IL = q) 2001 Pr(IL = q) 2001 Pr(IL = q P) 2002 Pr(P IL = q) 2002 Pr(IL = q) 2002 Pr(IL = q P) 2003 Pr(P IL = q) 2003 Pr(IL = q) 2003 p. 9
In synthesis... Pr(IL = q P) 2001 Pr(P IL = q) 2001 Pr(IL = q) 2001 Pr(IL = q P) 2002 Pr(P IL = q) 2002 Pr(IL = q) 2002 Pr(IL = q P) 2003 Pr(P IL = q) 2003 Pr(IL = q) 2003 Pr(IL = q P) 2004 Pr(P IL = q) 2004 Pr(IL = q) 2004 Today s posterior is tomorrow s prior p. 9
Evidence propagation After we have estimated Pr(IL = q P) we can associate each observed individual with their integration level probability distribution p. 10
Evidence propagation After we have estimated Pr(IL = q P) we can associate each observed individual with their integration level probability distribution Integration Level High =? Medium =? Low =? Legal Status Population Registry Housing Arrangement Work Status A Stay permit B Stay permit Illegal Registered Not registered Private household Sharing Irregular Legal worker Inactive Illegal Unemployed First we instantiate (set) the observed factors, according to the profile P p. 10
Evidence propagation After we have estimated Pr(IL = q P) we can associate each observed individual with their integration level probability distribution Integration Level High = 0.9535 Medium = 0.0005 Low = 0.0460 Legal Status Population Registry Housing Arrangement Work Status A Stay permit B Stay permit Illegal Registered Not registered Private household Sharing Irregular Legal worker Inactive Illegal Unemployed First we instantiate (set) the observed factors Then, we propagate the evidence, in order to update the probability distribution of the unobserved latent variable, by means of Bayes theorem p. 10
Post stratification The original dataset can be augumented with the integration levels probability Id... LS P R HA W S High Medium Low 1... A Permit Registered Sharing Legal 0.9715 0.0000 0.0285 2... A Permit Registered Private Legal 0.9980 0.0000 0.0020 3... B Permit Not registered Private Inactive 0.9890 0.0005 0.0105........................... 3218... A Permit N/A Private Legal 0.9950 0.0000 0.0050........................... 8000... B Permit Not registered Irregular Legal 0.0005 0.9185 0.0810 The procedure is able to manage missing data in the observed profiles (i.e. case 3218) The distribution of the integration level is estimated averaging over the missing values p. 11
Integration level trend in the last four years 1 Integration Level probability distribution 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 High Medium Low 0 2001 2002 2003 2004 General increase in the level of integration Possible effect of legalisations (year 2002) p. 12
Integration level & origin 0.5 Probability of low integration level 0.4 0.3 0.2 South America Eastern Europe Other Africa Northern Africa Asia 0.1 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 Probability of high integration level South Americans & Eastern Europeans arrived later, and show lowest integration levels p. 13
Level of integration & duration of staying 1 0.9 0.8 Probability of high integration level 0.7 0.6 0.5 0.4 0.3 2001 2002 2003 2004 0.2 0.1 0 2 4 6 8 10 12 14 16 18 20 Time from arrival in Italy 5/6 years sufficient to reach high levels of integration? p. 14
Duration of staying in different sub-populations 1 Probability of high integration level (average 2001 2004) 0.8 0.6 Males Females 0.4 0.2 0 0 2 4 6 8 10 12 14 16 18 20 Time from arrival in Italy (years) 1 0.8 0.6 Catholics Muslims 0.4 0.2 0 0 2 4 6 8 10 12 14 16 18 20 Time from arrival in Italy (years) Females have higher levels of integration (more likely to join already settled relatives?) p. 15
1 Origin & duration of staying Probability of high integration level (average 2001 2004) 0.9 0.8 0.7 0.6 0.5 0.4 0.3 Eastern Europe Asia South America Northern Africa Other Africa 0.2 0.1 0 1 2 3 4 5 6 Time from arrival in Italy (years) Asians & Northern Africans to reach higher levels of integration earlier? p. 16
Results: 2001 investigation Participation to social life 1 Probability of integration Do you consider important the participation to social life? 0.9 0.8 High Medium Low 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Yes No Don t know p. 17
Results: 2001 investigation Access to media 1 Probability of integration How do you keep yourself informed? 0.9 0.8 0.7 0.6 High Medium Low 0.5 0.4 0.3 0.2 0.1 0 not interested newspapers TV news radio other don t know newspaper, TV news radio p. 18
Results: 2002 investigation Influence of religious activity 1 Probability of high integration level 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 Non active Catholics Active Catholics Non active Muslims Active Muslims 0.1 0 0 1 2 3 4 5 6 7 8 9 10 Time from arrival in Italy (years) p. 19
Results: 2003 investigation Effect of legalisation Year 2003 Never taken advantage of legalisation Took advantage of at least one legalisation Probability of high integration level Total 0.7179 Have had a stay permit 0.9186 Total 0.7447 Legalisation 1986/87 0.9514 Legalisation 1990/91 0.9554 Legalisation 1995/96 0.9416 Legalisation 1998/99 0.9140 Legalisation 2002 0.3798 a) Always-legal-migrants tend to have a higher probability of being highly integrated, as compared to individuals taking advantage of at least one legalisation (0.91 vs 0.74) b) The longer since legalisation, the greater the probability of high integration p. 20
Possession/purchase of goods & services Results: 2004 investigation Goods % of possession Probability of high integration level Car 37.4 96.7 PC 16.5 94.6 Motorbike 16.6 89.3 Travel for pleasure 10.4 95.1 The consumption of durable goods or services can be associated with higher integration levels p. 21
Conclusions Integration seems to be something achieved with time, directly or through relatives already settled in Italy (e.g. Muslim females) At a global level, the higher integration is associated with the increasing establishment of the phenomenon Conversely, at the individual level the tendency is: the longer since arrival, the higher the integration level Different religions, and the active practice do not seem to discriminate Only in the very first period of immigration it tends to facilitate Catholics and to be a weakening factor for Muslims Individuals who benefit of legalisations take a longer time to increase their integration level Integration level is directly correlated to age of individual presence in Italy, but it is also correlated in the same sense with the average age of presence in Italy of the community or ethnic group p. 22