From dusk till dawn: the residential mobility and location preferences of immigrants in France

Abstract

This paper provides an original analysis on residential mobility and locational choices of immigrants in France by using a unique database containing individual characteristics for 19 million French inhabitants in 2014. Residential mobility is studied at the level of the 304 metropolitan zones d’emploi, which is a very narrow spatial level of analysis taking into account worker’s commutes, local productive specializations and the spatial dimensions of the local labor markets. The paper shows, first, that when living in a zone with a high proportion of immigrants, an immigrant is less eager to move elsewhere. Second, the location decisions of settled immigrants who relocate do not exhibit path dependence: the immigrants do not move to zones where other immigrants cluster. The paper also provides evidence that newcomers tend to locate to areas with low real estate prices and a high percentage of immigrant households within the local population. This is an original study of the locational behavior of migrants based on ethnic origin; it shows that the presence of other immigrants with the same or different ethnic origin affects the location decision of newcomers.

Appendices

Appendix 1

KS test: Gaussian distribution of second-order residuals; p value: 0.7594.

Appendix 2

KS test: Gaussian distribution of second-order residuals; p value: 0.7071.

Appendix 3

KS test: Gaussian distribution of second-order residuals; p value: 0.6920.

Appendix 4

To test the quality of the logit model estimations, Rakotomalala (2011) suggests building a “confusion matrix,” which allows comparison of the observed values of the dependent variable Y with those predicted by the estimations (given by $\hat{Y}$). This method leads to perform specificity and sensitivity tests for each model. The specificity index is the percentage of individuals $Y = 1$ for whom $\hat{Y} = 1.$ The sensitivity index is the percentage of individuals $Y = 0$ for whom $\hat{Y} = 0.$

$\forall i$, $p_{i} = P\left( {Y_{i} = 1} \right)$ is estimated by the logit model, with $u_{i} \sim U_{{\left[ {0,1} \right]}}$ indicating a uniform distribution between 0 and 1.

$$\hat{Y}_{i} = \left\{ {\begin{array}{*{20}l} {1\;{\text{si}}\quad u_{i} \le p_{i} } \hfill \\ {0\;{\text{si}}\quad u_{i} > p_{i} } \hfill \\ \end{array} } \right.$$

We compare the observed and estimated values. The following table features the specificity and sensitivity indexes for Models 1a, 1b and 2. The indexes show that the predictions of the logit models are correct for 65% of the observations.

Specificity and sensitivity of Models 1a and 1b

Prediction	Observation		Prediction	Observation
	MIG = 1	MIG = 0		MIG = 1	MIG = 0
Model 1a (in %)			Model 1b (in %)
MIG = 1	75.2	12.4	MIG = 1	73.7	13.4
MIG = 0	24.8	87.6	MIG = 0	26.3	86.6

Specificity and sensitivity of Model 2

Prediction	Observation
	IMM = 1	IMM = 0
in  %
IMM = 1	67.2	15.2
IMM = 0	32.8	84.8

We follow a similar procedure for the multinomial logit model. $\forall i,$$p_{i,k} = P\left( {Y_{i} = k} \right)$ with $k \in \left\{ {0, \ldots ,K} \right\}$ and $u_{i} \sim U_{{\left[ {0,1} \right]}}$ a uniform distribution between 0 and 1.

$$\hat{Y}_{i} = k \Leftrightarrow \sum\limits_{m = 0}^{k - 1} { p_{i,m} } < u_{i} \le \sum\limits_{m = 0}^{k} { p_{i,m} } \left( {{\text{with}}\;\sum\limits_{m = 0}^{ - 1} { p_{i,m} } = 0} \right)$$

The results for Model 3 are displayed in the following table, which shows that the predictions of the model are correct in 60% of all cases.

Specificity of Model 3

Observation	Prevision
	Correct	Wrong
in %
Rest of the World	61.8	38.2
Europe	75.2	24.8
Algeria	68.6	31.4
Morocco	69.7	30.3
Tunisia	62.2	37.8
Rest of Africa	64.6	35.4

Appendix 5

The instrumental variable REPATR shows the geographical distribution of almost one million French citizens and immigrants previously living in Northern Africa who moved to France in the 1960s. We calculate the percentage of immigrants with a foreign nationality in each ZE in 2014 with the following regression:

$${\text{FOREIGNERS}}\% = \pi + \pi_{\text{ZE}} \cdot X_{\text{ZE}} + \pi_{\text{end}} \cdot {\text{REPATR}} + \varepsilon_{\text{ZE}}$$

with ${\text{FOREIGNERS}}\%$ representing the percentage of immigrants, $X_{\text{ZE}}$ the territorial variables used in Models 1a, 1b and 2 and $\varepsilon_{\text{ZE}}$ the Gaussian residuals. The estimation of ${\text{FOREIGNERS}}\%$ and the regression parameters are calculated using the OLS method. The adjusted R² of the regression is 0.19 without the instrumental variable and 0.55 when using it. The following figure illustrates the distribution of the Gaussian residuals from the regression. It shows that the estimation of ${\text{FOREIGNERS}}\%$ is non-biased and of a small variance.

The Hausman test confirms the presence of endogeneity. The estimations of the parameters related to ${\text{FOREIGNERS}}\%$ are significantly different for Models 1a, 1b and 2 with the use of the instrument (p value < 0.0001 in all cases).

KS test: Gaussian distribution of second-order residuals; p value: 0.9580.

Appendix 6

Descriptive statistics for Models 1a and 1b.

		Statistics for Models 1a and 1b
		Total			in  %
		MIG = 0	MIG = 1	Total	MIG = 0	MIG = 1	Total
Total		41,217,196	1305,261	42,522,457
Nationality	Natives	35,284,536	1,141,124	36,425,660	85.61	87.42	85.66
	Immigrants: French nationality	3,318,970	74,183	3,393,153	8.05	5.68	7.98
	Immigrants: foreign nationality	2,613,690	89,954	2,703,644	6.34	6.89	6.36
Gender	Men	19,150,861	634,984	19,785,845	46.46	48.65	46.53
	Women	22,066,335	670,277	22,736,612	53.54	51.35	53.47
Owner	No	14,858,221	974,225	15,832,446	36.05	74.64	37.23
	Yes	26,358,975	331,036	26,690,011	63.95	25.36	62.77
Education (university)	No	29,377,766	709,628	30,087,394	71.28	54.37	70.76
	Yes	11,869,430	595,633	12,465,063	28.80	45.63	29.31
Living alone	No	32,035,877	892,101	32,927,978	77.72	68.35	77.44
	Yes	9,181,319	413,160	9,594,479	22.28	31.65	22.56
Working	No	16,061,239	313,008	16,374,247	38.97	23.98	38.51
	Yes	25,155,957	992,253	26,148,210	61.03	76.02	61.49

1. Residents over 20 years old in metropolitan France

Descriptive statistics for Model 2.

		Statistics for Model 2
		Total			Total
		IMM = 0	IMM = 1	Ensemble	IMM = 0	IMM = 1	Ensemble
Total		2,703,644	94,953	2,798,597
Gender	Men	1,329,967	41,660	1,371,627	49.19	43.87	49.01
	Women	1,373,677	53,293	1,426,970	50.81	56.13	50.99
Owner	No	1,773,510	77,342	1,850,852	65.60	81.45	66.13
	Yes	930,134	17,611	947,745	34.40	18.55	33.87
Education (university)	No	2,040,731	46,692	2,087,423	75.48	49.17	74.59
	Yes	662,913	48,261	711,174	24.52	50.83	25.41
Living alone	No	2,245,117	71,128	2,316,245	83.04	74.91	82.76
	Yes	458,527	23,825	482,352	16.96	25.09	17.24
Working	No	1,028,284	38,827	1,067,111	38.03	40.89	38.13
	Yes	1,675,360	56,126	1,731,486	61.97	59.11	61.87

1. Immigrants over 20 years old in metropolitan France

Appendix 7

		Statistics for Model 3
		Rest of world	Europe	Algeria	Morocco	Tunisia	Africa other
		Country = 0	Country = 1	Country = 2	Country = 3	Country = 4	Country = 5
Total		37,583	58,875	7090	6,403	3,666	9,811
Gender	Men	17,080	27,685	2667	2,673	1,660	4,483
	Women	20,503	31,190	4423	3,730	2,006	5,328
Owner	No	32,112	50,582	6379	5652	3197	7881
	Yes	5471	8293	711	751	469	1930
Education (university)	No	10,230	23,600	4364	3504	1855	4800
	Yes	27,353	35,275	2726	2898	1811	5011
Living alone	No	24,281	38,667	6384	4887	2922	7590
	Yes	13,302	20,208	706	1516	744	2221
Working	No	15,381	17,873	3059	2868	1357	3225
	Yes	22,202	41,002	4031	3534	2309	6586
in %
Gender	Men	45.45	47.02	37.62	41.75	45.28	45.69
	Women	54.55	52.98	62.38	58.25	54.72	54.31
Owner	No	85.44	85.91	89.97	88.27	87.21	80.33
	Yes	14.56	14.09	10.03	11.73	12.79	19.67
Education (university)	No	27.22	40.08	61.55	54.72	50.60	48.92
	Yes	72.78	59.92	38.45	45.26	49.40	51.08
Living alone	No	64.61	65.68	90.04	76.32	79.71	77.36
	Yes	35.39	34.32	9.96	23.68	20.29	22.64
Working	No	40.93	30.36	43.15	44.79	37.02	32.87
	Yes	59.07	69.64	56.85	55.19	62.98	67.13

1. Immigrants over 20 years old who arrived in metropolitan France in 2013
2. Source: INSEE, Recensement de la population

Appendix 8

The convexity of the curve for the locational preferences of newcomers according to the proportion of foreigners in a zone clearly shows that up to a certain threshold, the probability for a new migrant to settle in a given zone decreases as the number of immigrants already living in the zone increases. Past this threshold, the incoming mobility of newcomers strongly increases.