Are Commuters in the EU Better Educated than Non-commuters but Worse than Migrants?

Abstract

This paper analyses the skill and age structure of commuters in 14 EU countries. Theory implies that commuters can be either more or less able than stayers, but are always less able and older than migrants. Empirically, all types of commuters are younger and have higher education than region stayers, but older and less educated than migrants. Internal commuters are better educated and younger than cross-border commuters, education decreases while age increases with distance commuted and recent migrants are younger but also more highly educated than commuters.

Keywords

commuting selectivity migration

1. Introduction

Increased geographical mobility of labour in the EU could have important repercussions on the skill distribution of the workforce residing and working in a region. This has long been recognised by the migration literature where the determinants of the skill structure of migrants have been a central concern of both empirical and theoretical research (for example, Chiswick, 1999; Hunt, 2004; Borjas, 1999). Similar research with respect to commuters, by contrast, has been rare. Empirical results for individual countries and regions (for example, Eliasson et al., 2003; van Ommeren et al., 1999; Rouwendal, 1999; Gottholmseder and Theurl, 2007; Paci et al., 2007; Huber and Nowotny, 2011) often find that commuters are more highly educated than non-commuters, but offer little theoretical explanation for this.

Sorek (2009) argues that this lack of theory implies that effects of infrastructure investments, reducing travelling times between regions, on settlement structures cannot be analysed. He therefore considers a general equilibrium model of two distant, disconnected geographical zones using different technologies to find that the least able live and work in the (sending) region with low returns to ability, while those with intermediate ability commute and the most able migrate to the (receiving) region with high returns to ability. This finding is slightly in contrast to results of migration theory, which predicts that the most able migrate from places with low to places with high returns to ability, while the least able migrate in the opposite direction (Borjas, 1987). The reason for this is that Sorek (2009) assumes equal wages for the least able in both regions, so that there are no incentives for them to migrate or commute.

This paper extends Sorek’s model in two directions and uses data from the European Labour Force Survey (ELFS) to test the predictions of the extended model. The theoretical analysis first of all allows for ability-independent wage components to differ across regions and thus accounts for potential incentives of low ability individuals to commute. Second, it considers selection of commuters with respect to age. I show that in this version of the model commuters can be either more or less able than stayers, but are always less able than migrants and that commuters are also always older than migrants.

The empirical analysis tests these hypotheses and differentiates between cross-border and within-country commuters as well as between commuters across different distances and commuters to capital cities and other regions. It finds that commuters in most of the 14 EU countries analysed are more highly educated and younger than region stayers. Deviations from this pattern arise only in the EU member-states which joined the EU after May 2004. In addition, internal commuters are more highly educated but slightly younger than cross-border commuters and persons commuting larger distances are less strongly positively self-selected on education but younger. Finally, cross-border and internal commuters are compared with recent cross-border and internal migrants. As predicted by theory, both cross-border and internal commuters are older but also less highly educated than migrants.

2. Theory

As a starting-point for the analysis, following Sorek (2009), I consider an economy consisting of two regions (denoted by f and n respectively) and focus on the decision of a resident of n to work and live in n, or to commute or migrate to f. Individuals differ with respect to ability (s > 0) and age (T > a > 0) with T the retirement age. Each individual commands one unit of time which is split between commuting $τ_{ij}$ from the region of work (j) to the region of residence (i) and working. Since I consider only two regions, I assume that time spent commuting is zero if the person lives and resides in the same region and $τ$ if the person resides in region f but lives in n (i.e. $τ_{ij} = 0$ if i = j and $τ_{ij} = τ$ if i ≠ j).

Aside from allowing for heterogeneity with respect to age, I also differ from Sorek (2009) by assuming that individuals working in region j receive income from an ability independent base wage rate ( $w_{j}$ ) (i.e. a subsistence wage level for the least able workers) which net of commuting time is higher in f than in n (i.e. $w_{f} (1 - r) > w_{1} n)$ . Therefore in this model, in the absence of commuting and migration costs, the least able prefer to work in f, while in Sorek (2009), who assumes $w_{f} = w_{n} = 0$ , the least able have no incentive for mobility. In addition wages as in Sorek (2009) also depend on an ability-dependent component ( $λ_{j} s$ ) with $λ_{j}$ a parameter measuring returns to ability, which differ among regions, to allow for differences in technology¹ as well as (in a cross-border context) for potential difficulties in transfering skills across borders. The expected lifetime earnings of an individual of age a and ability s working in region j is therefore $(T - a) (w_{j} + λ_{s}) (1 - τ_{ij})$ . In addition, migrants incur fixed migration costs of k while commuting is associated with financial cost c, which is, however, paid per period of commuting and where to assure viability of commuting k > cT.²

Individuals residing in region n are therefore faced with a choice between working and residing in region n (i.e. staying), which gives an income of

y^{s} = (T - a) (w_{n} + λ_{n} s)

(1)

residing in n and working in f (i.e. commuting), with income

y^{c} = (T - a) [(w_{f} + λ_{f} s) (1 - r) - c]

(2)

and working and residing in f (i.e. migrating), which yields income³

y^{M} = (T - a) (w_{f} + λ_{f} s) - k

(3)

Assuming that $λ_{f} > λ_{n}$ and (without loss of generality) normalising $λ_{f}$ to unity as well as letting $λ = \frac{λ_{n}}{λ_{f}}$ , income-maximising individuals are indifferent between migrating and commuting at ability

s^{CM} = \frac{k - c (T - a)}{(T - a) τ} - w_{f}

(4)

with the individual preferring to migrate if

s > s^{CM}

τ (s + w_{f}) = \frac{k - c (T - a)}{(T - a)}

Equation (4) defines the ability at which individuals of a given age are indifferent between commuting and migration and highlights the central trade-off driving the decision between migrating and commuting. If the difference between (annualised) costs of migration and commuting—i.e. $[k - c (T - a)] / (T - a)$ —is large relative to the foregone income arising from time spent commuting—i.e $τ (w_{f} + s)$ —individuals prefer commuting over migrating. In consequence, the most able, who have the highest opportunity costs for time spent commuting, migrate rather than commute and the critical ability at which migration is preferred to commuting increases with age.

Similarly, the combinations of ability and age at which individuals are indifferent between staying and commuting is given as

s^{sc} = \frac{w_{n} - w_{f} (1 - τ) + c}{(1 - τ - λ)}

(5)

with individuals preferring to stay if

s^{SC} \geq s if 1 - τ > λ

and if

s^{SC} \leq s if 1 - τ < λ^{4}

Equations (4) and (5) state that with respect to selection on ability, two possible situations can arise. The first, occurs when returns to ability are larger in f than in n and commuting time is not too large (i.e. $1 - τ > λ$ ). In this case if $w_{n} > w_{f} (1 - τ) - c$ , individuals with an ability of $s^{SM} \geq s$ prefer commuting over migration and individuals with $s^{SC} < s$ prefer commuting over staying.⁵ Thus for commuting to be preferred over both migration and staying $s^{SM} \geq s > s^{SC}$ must apply and commuters are more able than stayers, but less able than migrants. The reason for this is that when $1 - τ > λ$ returns to education abroad are higher both for migrants and commuters than at home. This creates an incentive for the able either to migrate or commute to f. In deciding between migration and commuting, however, those with the highest ability have most to gain from higher returns to ability in f but also have the highest costs due to foregone earnings during time spent commuting. For them therefore migration is most attractive. For the medium-ability groups, which also gain from working abroad, but for whom foregone earnings due to time spent commuting are lower than for the high-ability groups, commuting is more attractive. For the least able, incentives to commute or migrate are lowest, since they do not profit from high returns to ability in the receiving region.

The second case occurs when returns to ability are low in f relative to n or commuting time is large (i.e. when $1 - τ < λ$ ). In this case comuting is optimal for individuals with the lowest ability (i.e. with $s < min [s^{SC}, s^{CM}]$ ). The reason for this is that if $1 - τ < λ$ returns to ability net of commuting time are lower in f than in n, although returns to education excluding commuting costs are higher. Therefore the most able lack incentives to commute, while the least able have such an incentive if base incomes net of commuting costs are higher in f than in n (i.e. $w_{n} < w_{f} (1 - τ) - c$ ).⁶

Furthermore, rearranging equations (4) and (5), we can derive the age at which individuals are indifferent between migration and commuting ( $a^{CM}$ ) as

a^{CM} = T - \frac{k}{(w_{f} + s) τ + c}

with the individual preferring to commute if $a \leq a^{CM}$ , while as can be seen from equation (5) the decision between commuting and staying is independent of age. Thus irrespective of whether $1 - τ < λ$ or $1 - τ > λ$ , staying or commuting is optimal for the older since for them higher returns to education abroad do not justify paying the fixed costs of migrating. For the youngest, migration is optimal since annualised fixed migration costs are low due to a long remaining working lifetime ( $T - a$ ).

Thus from equations (4) and (5) it follows that commuters and stayers are always older as well as less able than migrants. Depending on commuting time and relative returns to ability in the receiving and sending region, commuters may, however, be less or more able than stayers. A full description of the model, however, also has to consider the decision to stay or migrate. From equations (1) and (3), the level of ability at which individuals are indifferent between staying and migrating is given by

s^{SM} = \frac{k + (T - a) (w_{n} - w_{f})}{(1 - λ) (T - a)}

(6)

with the individual preferring to stay if $s^{SM} \geq s$ . This after rearranging gives a critical age at which individuals are indifferent between migrating and staying of

a^{SM} = \frac{k}{(1 - λ) s + w_{n} - w_{f}}

with the individual preferring to stay if $a^{SM} \leq a$ . So that for migration to be preferred to both commuting and staying, $s > max [s^{CM}, s^{SM}]$ and $a > min [a^{CM}, a^{SM}]$ must apply. Therefore for a given age migrants are always the most able and for a given ability they are the youngest.⁷

In sum, theory predicts that commuters can be either more or less able than stayers, but are always less able than migrants and that commuters and stayers are always older than migrants. For a given ability therefore the probability of a person to commute should be highest for the older age groups, while for a given age commuters should always be less able than migrants but more able than stayers if $1 - τ > λ$ , which means that commuting time is not too large (i.e. small $τ$ ) and relative returns to ability in f relative to n are large (i.e. small $λ$ ). By contrast if $1 - τ < λ$ , which means large commuting times but low returns to ability in f relative to n (i.e. large $λ$ ), commuters should be less able than stayers. Therefore all else equal commuters to regions that are closer together or commuters to regions with higher returns to education should be more able than stayers, at a given age, while commuters over longer distances or to regions with lower returns to ability should be less strongly positively self-selected. Furthermore from the definition of $a^{CM}$ it is easy to see that a change in $λ$ has no effect on the age of commuters. By contrast, an increase in $τ$ , which implies commuting over greater distances, increases $a^{CM}$ so that the impact of increasing commuting distances on commuter age is positive. Finally, also note that this model could be extended in a number of interesting ways by allowing for wages or commuting time to depend on age. While this would substantially complicate the analysis, since it would imply that no explicit solution could be derived for $a^{CM}$ and $a^{SM}$ , it could—depending on the relative returns of wages at home and abroad with respect to age or the slope of commuting time with respect to age—lead to commuters either being older or younger than stayers.

3. Data and Method

To test these predictions empirically, using education as a proxy for ability, I estimate logit models of the choice between residing and working in a region and commuting for different types of commuters. I differentiate between cross-border and internal commuters, since these may differ with respect to travelling times, differences in returns to education and difficulties in transfering human capital across regions. This could lead to cross-border commuters being more highly educated and younger than internal commuters if differences in returns to education are larger for cross-border commuters and older but less educated if highly educated commuters face greater problems of skill-transfer when commuting across borders. In addition, among internal commuters, commuters to capital cities and other regions are considered separately, since the little evidence available on regional differences in returns to education (for example, Cabral-Vieira et al., 2006; Hazans, 2003a) suggests that these are higher in capitals than elsewhere.⁸ This should make commuters to capital cities more able than those to other regions. Finally, commuters are also differentiated by distance covered in commuting, since theory suggests that commuters over larger distances are less able and older than commuters over shorter distances.

The data are taken from the ELFS for the year 2006. They contain information on the region of work and residence (where the lowest regional disaggregation is NUTS1 for Austria, Germany and the UK and NUTS2 for all other countries) as well as on the region of residence one year ago and a number of demographic and workplace characteristics of persons in paid employment in 14 EU countries (Austria, Belgium, Bulgaria, Czech Republic, Finland, France, Germany, Hungary, Italy, Poland, Slovakia, Spain, Romania, UK).⁹ Therefore commuters can be defined as persons living in one region and working in another, with cross-border commuters working in another country than they live in, and internal commuters working in a different region than they live in, but in the same country. Furthermore, by comparing the current region of residence with the region of residence one year ago, it is also possible to define both recent internal and cross-border migrants as persons who have moved region of residence in the past year, and to compare these with commuters as well as to stayers (i.e. persons that neither migrated nor commuted).

Table 1 provides descriptive statistics for all groups considered (i.e. stayers, commuters, migrants, cross-border and internal migrants, and commuters and commuters to capital cities and other regions). According to these data, around 0.6 per cent of the employed commuted across borders and 4.2 per cent commuted across regions within their respective countries in 2006. Similarly, the share of internal migrants was 0.6 per cent, while cross-border migrants accounted for 0.1 per cent of the employed. Commuters differ most significantly from stayers by a high share of males and a larger share of persons aged 20–39. Cross-border commuters often have intermediate education and work in manufacturing (including construction). Internal commuters are more often highly educated and often work in market services. Relative to migrants, however, both cross-border and internal commuters more often have intermediary education and are also older. Finally, migrants are more often single than either commuters or stayers, while differences between these groups with respect to having children are small.

Table 1.

Descriptive statistics for different types of commuters and migrants (percentages)

	Stayers	Commuters					Migrants			Total
		Overall commuters	Cross-border commuters	Internal Commuters			Overall migrants	Cross-border migrants	Internal migrants
		Overall commuters	Cross-border commuters	Total	To capital	To other regions	Overall migrants	Cross-border migrants	Internal migrants
Total (percentage of resident population)	94.2	5.2	0.6	4.6	3.7	0.9	0.7	0.1	0.6	100.0
	Age
15–19 years	2.5	2.3	1.2	2.4	2.8	1.1	3.1	2.5	3.1	2.5
20–29 years	19.2	22.9	26.5	22.4	21.8	25.0	51.5	54.3	51.5	19.3
30–39 years	26.7	27.8	26.9	27.9	27.9	28.0	25.7	30.4	25.7	26.8
40–49 years	27.3	26.6	27.3	26.5	26.2	27.6	14.6	6.6	14.6	27.2
50–59 years	19.6	17.6	15.5	17.8	18.2	16.5	4.5	6.2	4.5	19.5
60 or more years	4.7	2.9	2.6	2.9	3.2	1.8	0.6	0.1	0.6	4.6
	Education
Low education^a	24.0	16.5	15.4	16.6	17.2	14.0	12.3	12.3	12.3	23.6
Medium education^b	50.7	48.7	58.0	47.5	48.0	45.7	42.4	38.6	42.4	50.6
High education^a	25.2	34.8	26.7	35.9	34.8	40.2	45.3	49.1	45.3	25.7
	Gender
Female	45.3	33.4	28.1	34.1	33.7	35.7	43.2	44.6	43.2	44.6
Married	58.8	53.7	53.8	53.7	54.0	52.6	30.1	32.3	30.1	58.5
Child	45.4	44.4	46.4	44.1	42.0	52.6	41.9	31.7	41.9	45.3
	Sector
Agriculture	5.9	1.6	2.4	1.5	1.5	1.2	0.9	7.5	0.9	5.7
Manufacturing^a	27.6	31.3	43.6	29.7	30.9	24.9	22.4	23.4	22.4	27.8
Market services	36.6	41.4	36.2	42.1	41.4	45.2	42.1	45.1	42.1	36.8
Non-market services	29.9	25.7	17.7	26.7	26.2	28.7	34.7	24.0	34.7	29.7

ISCED 2 or less.

ISCED 3–4.

ISCED 5 or more.

including construction.

Note: standard deviations of dummy variables are given by $\sqrt{s^{*} (1 - s)}$ where s is the share of persons with the respective attribute.

Source: ELFS (2006).

Both internal and cross-border commuting are, however, also highly dependent on a country’s geography (see Figure 1). High rates of outbound cross-border commuting primarily occur in regions close to borders and in small countries (such as Belgium and Austria), where most regions are located close to borders. High rates of outbound internal commuting are found primarily in the vicinity of large urban agglomerations (London, Madrid, Prague and Bratislava) and countries with smaller regions. In addition, cross-border commuting is most prevalent at borders of countries which either share a common language (for example, France and Belgium or Austria, Germany and Switzerland) or have been a single country until recently (i.e. the Czech Republic and Slovakia) but also at the Austrian–Hungarian and Czech–German border where wage differences are large.

Figure 1.

Out-commuting in the EU27 by NUTS2 regions, 2006.

In the logit analysis, I therefore include a set of dummy variables for each of the 158 regions considered, to capture any effects of differences in size, geographical position and economic development between regions on commuting. In addition based on the results of the empirical literature, which finds that commuters are often male and also establishes an impact of marital status and having children on the probabilty of commuting in some cases (for example, Hazans, 2003b; Benito and Oswald, 2000; Paci et al., 2007; and White, 1986), controls for gender, marital status and presence of children in the household and dummy variables for the sector of employment (agriculture, manufacturing and private or public services) are included. Finally, two dummy variables measuring whether a person has completed intermediate (ISCED 3 or 4) or a high (ISCED 5 or 6) education respectively, with compulsory education (ISCED 2 or less) as the excluded base group, and five dummies for the age of respondents (measuring whether individuals are 20–29, 30–39, 40–49, 50–59 and 60 or more years old, with 15–19 year olds as a base category) are included. These are the variables of interest, with education dummies proxying ability and age dummies accounting for potential non-linearities of the impact of age on commuting. A positive coefficient of these variables signifies that commuters are positively self-selected from this group relative to stayers and a negative coefficient indicates negative self-selection of commuters.

In addition, I also separately compare cross-border and internal commuters with cross-border and internal migrants and stayers by means of a multinomial logit analysis of the choice between migrating, commuting and staying, controlling for the same explanatory variables as before. For cross-border migrants and commuters, this is, however, only possible at the national level, since these groups are not asked about their region, but only about their country of previous residence in the ELFS. When considering cross-border migrants and commuters, therefore, all those living and working in the same country (irrespective of whether they commute within the country or not) are defined as stayers and I can only control for country dummies (rather than region dummies) as explanatory variables.

4. Results

Table 2 shows the results for all commuters, cross-border commuters, internal commuters, commuters to capital city regions and to other regions respectively, by presenting odds-ratios of the estimates.¹⁰ It provides strong evidence of positive self-selection of commuters relative to stayers on education, irrespective of commuting type. In all of the estimates, the coefficients on both the dummy variable for intermediary as well as for high education are highly significantly different from 1 and increase with educational attainment.¹¹

Table 2.

Regression results for different types of commuting

	Overall commuters	Cross-border commuters	Internal commuters	Internal commuters
	Overall commuters	Cross-border commuters	Internal commuters	To capital	To other region
Age 20–29 years	1.40***	1.34***	2.19***	1.72***	1.28***
Age 20–29 years	(0.10)	(0.11)	(0.38)	(0.42)	(0.11)
Age 30–39 years	1.26***	1.24***	1.71***	1.46**	1.21**
Age 30–39 years	(0.10)	(0.10)	(0.30)	(0.26)	(0.10)
Age 40–49 years	1.17***	1.13	1.80***	1.47	1.08
Age 40–49 years	(0.09)	(0.09)	(0.32)	(0.36)	(0.09)
Age 50–59 years	1.06	1.03	1.44	1.18	1.02
Age 50–59 years	(0.08)	(0.08)	(0.36)	(0.29)	(0.09)
age 60 or more years	0.75	0.71	1.43	0.80	0.69***
age 60 or more years	(0.07)	(0.07)	(0.35)	(0.23)	(0.07)
Medium education	1.27***	1.25***	1.42***	1.32***	1.23***
Medium education	(0.04)	(0.04)	(0.12)	(0.09)	(0.04)
High education	1.89***	1.51***	1.79***	2.60***	1.77***
High education	(0.06)	(0.06)	(0.16)	(0.18)	(0.07)
Female	0.60***	0.61***	0.53***	0.58***	0.62***
Female	(0.01)	(0.01)	(0.03)	(0.03)	(0.02)
Married	0.97	0.97	0.90	0.84***	1.01
Married	(0.02)	(0.03)	(0.06)	(0.05)	(0.03)
Child	0.93***	0.93***	0.93	0.95	0.92***
Child	(0.02)	(0.03)	(0.07)	(0.05)	(0.03)
Observations	1,043,689	1,033,444	994,456	752,145	1,018,415
Log-likelihood	−32,591.6	−29,411.8	−5401.5	−6207.8	−25,186.8
Pseudo R²	0.12	0.13	0.18	0.27	0.12

Notes: Table reports odds ratios for weighted logit regressions on the probability of commuting relative to the probability of staying (sample excludes recent migrants), values in brackets are cluster robust standard errors. ***, ** and * signify odds ratios significantly different from 1 at the 1 per cent, 5 per cent and 10 per cent levels respectively. Base categories for dummy variables are 15–19-year-old males with completed compulsory education. Results for fixed effects of region of residence and sector of employment not reported.

There are, however, differences among commuter types. The coefficients imply that internal commuters are more positively self-selected on education than cross-border commuters. Persons with intermediary education have, by a factor of 1.9, a higher probability of commuting across borders relative to staying than persons with compulsory education. The probability for internal commuting relative to the probability of staying is, however, only by a factor of 1.3 higher for persons with intermediary education than for persons with compulsory education. Similarly, persons with tertiary education have, by a factor of 1.8, higher odds of commuting within a country but only by a factor of 1.5 a higher probability of commuting across borders. In terms of the theoretical model presented earlier, this could be explained by larger problems of cross-border commuters in transfering education across borders (for example, due to language problems) or by the longer travelling times in cross-border commuting leading to a weaker self-selection of cross-border commuters.

Furthermore—consistent with the theoretical model and the assumption that returns to ability are highest in large cities—among internal commuters, those commuting to capital cities are more positively selected on education than those commuting elsewhere. A person with intermediary education is by a factor of 1.3 more likely to commute to the capital city (relative to staying) than a person with at most completed compulsory education. For persons with completed tertiary education, the relative probability increases by a factor of 2.6. For internal commuters to other regions, these gains are 1.2 and 1.8 respectively.

Table A1 in the Appendix augments these results, by estimates for all commuters on a country-by-country basis. It suggests that these patterns apply in almost all countries of the EU. The odds ratios on the education variables are significantly larger than one in all countries except for the secondary educated in the new member-states (NMS) of the EU, which joined the EU after May 2004 (i.e. the Czech Republic, Hungary, Poland, Slovakia as well as Bulgaria and Romania, where the coefficient for tertiary education is also smaller than one).¹² Thus the education structure of commuters differs between the NMS and the pre-existing member-states. This may be a consequence of the substantial regional restructuring in the NMS in past decades (see Huber, 2007, and Ferragina and Pastore, 2008, for surveys).

Highly significant coefficients are also found for age. For all commuting types, the commuting probability (relative to the staying probability) attains a maximum for the 20–29-year-olds, with odds ratios suggesting a 2.2 times higher relative commuting probability for this age group than for the 15–19-year-olds among internal commuters and a 1.3 times higher probability among cross-border commuters. By contrast, coefficients for the age groups older than 50 years remain insignificant. Commuters are therefore younger than stayers, which—as discussed earlier—can be consistent with theory if either commuting costs increase with age or if there are differences in returns to age in the wage function. Once more, these results also hold for most EU countries except for the NMS (where results suggest that commuters are mostly 15–19 years old or do not differ significantly in age from stayers) considered in Table A1. Furthermore, the longer travel times implied by cross-border commuting lead to cross-boder commuters being older than internal commuters, while commuters to capital cities are slightly older (with commuting odds being higher relative to the base group of the very young for each age group for commuters to capital cities) than region stayers.

Aside from providing strong evidence for positive self-selection of commuters on education and a negative one on age, Table 2 also suggests that commuters are significantly less often female than male, with the coefficients, however, varying only marginally for different types of commuter. Once more, this result is highly consistent across countries (see Table A1). Although in our model gender differences are not modelled, this is consistent with the conjecture of White (1986) that, due to higher opportunity costs of time spent commuting for women (which may result from the traditional role of women in household production or alternatively lower wages in market production due to discrimination), women commute less.

Finally, results also suggest that having children significantly reduces the probability of cross-border commuting, while married persons less often commute to capital cities than singles. With respect to these variables, however, results vary somewhat across countries. This rather non-robust impact of these variables on commuting behaviour is consistent with the literature. For instance, Paci et al. (2007) in a comparative study of eight countries find that marital status has a significant impact on the probability of commuting in only three countries and according to Hazans (2003b) having children has a significant impact on the probability of commuting in only one of three Baltic countries.

4.1 Commuting across Different Distances

Table 3 takes this analysis one step further by considering the probability of commuting across different distances (i.e. flows where the capital cities of the sending and receiving regions are less than 50 kilometres, 50–100 kilometres, 100–150 kilometres and more than 150 kilometres apart).¹³ This is interesting, because theory suggests that commuters over longer distances should be less able and also older than short distance commuters.

Table 3.

Regression results for different commuting distances of internal commuters

	Distance travelled
	To 50 km	50–100 km	100–150 km	150+ km
Age 20–29 years	1.93***	1.90***	1.79***	1.49**
Age 20–29 years	(0.40)	(0.36)	(0.18)	(0.25)
Age 30–39 years	2.21***	2.17***	1.92***	1.20
Age 30–39 years	(0.42)	(0.46)	(0.16)	(0.20)
Age 40–49 years	1.95***	1.79***	0.91	1.13
Age 40–49 years	(0.41)	(0.34)	(0.16)	(0.19)
Age 50–59 years	1.66**	1.64*	0.77	1.07
Age 50–59 years	(0.35)	(0.32)	(0.13)	(0.18)
Age 60 or more years	0.91	0.87*	0.69*	0.69*
Age 60 or more years	(0.22)	(0.19)	(0.14)	(0.14)
Medium education	1.37***	1.32***	1.27***	1.21***
Medium education	(0.08)	(0.08)	(0.09)	(0.08)
High education	2.15***	2.02***	2.00***	1.97***
High education	(0.14)	(0.13)	(0.15)	(0.14)
Female	0.61***	0.59***	0.55***	0.53***
Female	(0.03)	(0.03)	(0.03)	(0.03)
Married	1.07	1.01	1.03	0.94
Married	(0.06)	(0.06)	(0.06)	(0.05)
Child	0.96	0.94	0.93	0.84***
Child	(0.06)	(0.05)	(0.05)	(0.05)
Observations	240,952	543,699	861,599	877,202
Log-likelihood	−5013.61	−7396.13	−8230.70	−9813.73
Pseudo R²	0.22	0.18	0.12	0.11

Notes: Table reports odds ratios for weighted logit regressions on the probability of commuting (sample excludes recent migrants), values in brackets are cluster robust standard errors. ***, ** and * signify odds ratios significantly different from 1 at the 1 per cent, 5 per cent and 10 per cent levels respectively. Base categories for dummy variables are 15–19-year-old males with completed compulsory education. Results for fixed effects of region of residence and sector of employment not reported.

In these regressions, in accordance with theory, short-distance commuters are younger than long-distance commuters, since for each age dummy the impact on the probability of commuting decreases for each consecutive distance category. Similarly, the impact of the educational variables on the probability of commuting decreases for each consecutive distance category. For instance, the 30–39-year-olds have, by a factor of 2.2, a higher probability of commuting less than 50 km (relative to staying) than the 15–19-year-olds, while for those commuting in excess of 150 km this effect is only 1.2 and statistically insignificant. Similarly, persons with intermediary education have, by a factor of 1.4, a higher relative probability of commuting across a distance of 50 or less km but, by a factor of 1.2, a higher relative probability of commuting more than 150 km than persons with a low education. For persons with a high education, the odds ratio is 2.2 for commuting distances below 50 km but 2.0. for commuting in excess of 150 km.

Furthermore, gender differences in commuting increase slightly with distance. Females have, by a factor of 0.6, a lower commuting probability than males for moves below a distance of 50 km but, by a factor of 0.5, a lower probability of commuting more than 150 km. This is once more consistent with the results of White (1986), since women’s higher opportunity costs of commuting would lead women to be particulary reluctant to commute over long distances. In addition, children in the household significantly reduce the probabilty of commuting more than 150 km and being married remains insignificant throughout.

4.2 Selection of Commuters and Migrants

Finally, the analysis can be extended to consider the selection of migrants relative to commuters. Table 4 reports results of multinomial logit regressions on the decision to migrate, commute and stay for both internal as well as cross-border migrants and commuters. These suggest that migrants are usually younger and more highly educated than commuters. The odds ratios imply that a completed tertiary education increases the probability of being a migrant relative to staying by a factor of 3.3 over that of a person with compulsory education for cross-border migrants and by factor of 2.7 for internal migrants.¹⁴ The respective odds ratios are 1.6 for cross-border and 1.9 for internal commuters. In addition, for cross-border commuters, odds ratios are higher than for cross-border migrants for intermediary education, but lower for tertiary education. This does not, however, apply to internal migrants and commuters. Cross-border commuters are therefore predominantly drawn from medium education levels, while internal commuters have a lower education than migrants (but a higher one than stayers) throughout.

Table 4.

Multinomial logit regression results for the choice of commuting, migrating and staying for cross-border and internal commuters

	Cross-border^a		Internal^b
	Commute vs stay	Migrate vs stay	Commute vs stay	Migrate vs stay
	Coefficient	Coefficient	Coefficient	Coefficient
Age 20-29 years	2.04***	1.97	1.34***	1.63**
Age 20-29 years	(0.35)	(1.46)	(0.11)	(0.40)
Age 30–39 years	1.64***	1.05	1.24***	0.61*
Age 30–39 years	(0.29)	(0.80)	(0.10)	(0.16)
Age 40–49 years	1.73***	0.26*	1.14	0.37***
Age 40–49 years	(0.31)	(0.20)	(0.09)	(0.10)
Age 50–59 years	1.39*	0.31	1.03	0.16***
Age 50–59 years	(0.25)	(0.26)	(0.09)	(0.05)
Age 60 or more years	1.39	0.02***	0.68***	0.11***
Age 60 or more years	(0.35)	(0.02)	(0.07)	(0.05)
Medium education	1.42***	1.00	1.24***	1.37***
Medium education	(0.12)	(0.24)	(0.04)	(0.16)
High education	1.59***	3.27***	1.90***	2.68***
High education	(0.15)	(0.78)	(0.06)	(0.31)
Female	0.56***	0.95	0.60***	0.75***
Female	(0.03)	(0.18)	(0.02)	(0.06)
Married	0.91	0.69	0.99	0.62***
Married	(0.06)	(0.18)	(0.03)	(0.06)
Child	0.95	0.62*	0.90***	1.16
Child	(0.07)	(0.16)	(0.03)	(0.12)
Observations	1,041,296		1,027,617
Log-likelihood	−6954.19		−33,105.29
Pseudo R2	0.35		0.13

Results for fixed effects of country of residence and sector of employment not reported.

Results for fixed effects of region of residence and sector of employment not reported.

Notes: Table reports odds ratios for weighted multinomial logit regression on the probability of commuting relative to the probability of staying (sample excludes recent migrants), values in brackets are cluster robust standard errors. ***, ** and * signify odds ratios significantly different from 1 at the 1 per cent, 5 per cent and 10 per cent levels respectively. Base categories for dummy variables are 15–19-year-old males with completed compulsory education. Results for fixed effects of region of residence and sector of employment not reported.

Similarly, the probability of migrating as well as commuting is largest for the age group of the 20–29-year-olds for both internal and cross-border migrants and commuters (although insignificantly so for cross-border migrants), but for the age groups older than 40 the probability of migration relative to staying is already significantly lower than 1 for both cross-border and internal migrants, while it is still larger than 1 (although insignificantly so for internal commuters) for cross-border and internal commuters. Thus, consistent with theory, both internal and cross-border commuters are older than migrants.

In addition, both cross-border and internal commuters are less often female than stayers. This does not, however, apply to cross-border migrants. By contrast, being married significantly reduces only the probability of migrating (relative to staying) internally while having children reduces only the relative probability of commuting internally and migrating across borders (although the latter coefficient is only on the margin of significance).

5. Conclusions

Increased geographical mobility of labour may have important repercussions on the skill distribution of the workforce residing and working in a region. Aside from migration, commuting is another mechanism by which this population sorting may be encouraged. This paper analyses the education and age structure of commuters in 14 EU countries. Theory implies that commuters are always older as well as less able than migrants. Depending on the commuting time between regions and relative returns to education, they may, however, be less or more able than stayers.

The empirical results indicate that all types of commuters in most countries are younger and have higher education than residents of the same region who do not commute. Deviations from this pattern only occur in the EU member-states which joined the EU in May 2004, where in particular workers with completed secondary or vocational education tend to have a lower probability of commuting, and internal commuters are often younger than in the other EU countries. In addition internal commuters are more strongly positively self-selected on education (in particular, when commuting to capital city regions) and younger than cross-border commuters; persons commuting larger distances are usually less highly educated and older; and recent migrants are younger but also more highly educated than commuters. In addition, commuters are often young and male, with gender differences being largest for shorter-distance commuting.

From a policy perspective, this implies that measures to reduce travel times between regions (such as investments in transport infrastructure or in the European context integration of cross-border labour markets) aside from leading to increased commuting, will also lead to a larger share of highly educated commuting and could thus impact on population sorting. Relative to policies focusing on migration such policies are, however, also likely to affect disproportionately the medium-skilled.

Footnotes

Appendix

Table A1.

Regression results for overall commuting by country

	Austria		Belgium		Bulgaria		Czech Republic		Germany		Spain		Finland
	Coefficient	S.E.	Coefficient	S.E.	Coefficient	S.E.	Coefficient	S.E.	Coefficient	S.E.	Coefficient	S.E.	Coefficient	S.E.
Age 20–29 years	2.12***	0.27	1.97***	0.31	1.16	0.47	1.32*	0.22	1.42***	0.13	2.75***	0.85	2.25***	0.70
Age 30–39 years	2.01***	0.27	2.04***	0.32	1.07	0.43	0.90	0.15	1.39***	0.13	1.93**	0.61	1.26	0.40
Age 40–49 years	1.81***	0.25	2.01***	0.32	1.12	0.45	0.63***	0.11	1.22	0.19	1.98**	0.64	1.70*	0.53
Age 50–59 years	1.81***	0.26	1.87***	0.30	0.87	0.35	0.58***	0.10	1.00	0.24	1.77	0.62	1.34	0.42
Age 60 or more years	1.22	0.38	1.06	0.19	0.58	0.26	0.61***	0.11	0.89	0.24	1.24	0.52	1.52	0.56
Medium education	1.30***	0.10	1.20***	0.05	0.79**	0.08	0.96	0.07	1.72***	0.19	1.45***	0.18	1.34**	0.14
High education	1.69***	0.12	2.14***	0.07	0.61***	0.08	2.30***	0.18	2.31***	0.28	2.17***	0.22	1.98***	0.15
Female	0.73***	0.03	0.66***	0.02	0.45***	0.04	0.57***	0.02	0.64***	0.04	0.53***	0.05	0.43***	0.05
Married	1.03	0.05	0.93	0.03	1.02	0.10	0.80***	0.03	0.99	0.07	0.83	0.11	0.63***	0.07
Child	0.74***	0.04	0.95***	0.04	1.02	0.09	0.88***	0.04	0.90	0.06	0.89	0.10	1.17	0.22
Observations	−586.20		−2015.11		−245.15		−880.88		−8091.71		−1930.41		−296.86
Log-likelihood	93,823		47,494		52,349		112,871		22,429		41,540		17,668
Pseudo R²	0.02		0.10		0.04		0.12		0.09		0.08		0.06
	France Coefficient	S.E.	Hungary Coefficient	S.E.	Italy Coefficient	S.E.	Poland Coefficient	S.E.	Romania Coefficient	S.E.	Slovakia Coefficient	S.E.	UK Coefficient	S.E.
Age 20–29 years	1.14***	0.06	1.09	0.16	1.49**	0.23	1.36	0.41	0.72*	0.14	0.70***	0.09	1.22**	0.10
Age 30–39 years	1.13***	0.06	0.76*	0.11	1.06	0.17	1.14	0.35	0.61**	0.12	0.39***	0.05	1.25**	0.10
Age 40–49 years	1.09*	0.05	0.62***	0.09	0.86	0.14	0.94	0.29	0.57***	0.11	0.31***	0.04	1.03	0.10
Age 50–59 years	1.19	0.17	0.52***	0.08	0.65***	0.10	0.71	0.22	0.31***	0.06	0.28***	0.04	0.91	0.09
Age 60 or more years	0.64	0.22	0.52***	0.10	0.65**	0.12	0.71	0.26	0.05***	0.03	0.20***	0.04	0.55***	0.07
Medium education	1.15**	0.08	0.86***	0.04	1.31***	0.05	1.19	0.15	0.38***	0.04	0.83**	0.07	1.30***	0.06
High education	1.46***	0.12	1.42***	0.08	2.56***	0.12	1.91***	0.26	0.30***	0.04	1.17**	0.0.8	2.06***	0.11
Female	0.66	0.04	0.51***	0.02	0.48***	0.02	0.40***	0.03	0.41***	0.03	0.62***	0.02	0.60***	0.02
Married	0.96	0.07	0.81***	0.03	0.83***	0.04	1.15*	0.09	0.78***	0.06	0.52***	0.02	1.18***	0.05
Child	0.96	0.07	0.82***	0.03	0.82***	0.04	1.06	0.09	0.90	0.07	0.99	0.04	1.00***	0.04
Observations	−5862.60		−640.66		−2480.85		−1191.45		−401.57		−720.72		−7007.43
Log-likelihood	22,229		113,219		235,107		78,483		105,560		45,412		55,505
Pseudo R²	0.06		0.09		0.05		0.07		0.20		0.10		0.08

Notes: Table reports odds ratios for weighted logit regression on the probability of commuting relative to the probability of staying (sample excludes recent migrants), S.E. = cluster robust standard errors, ***, ** and * signify odds ratios significantly different from 1 at the 1 per cent, 5 per cent and 10 per cent levels respectively. Base categories for dummy variables are 15–19-year-old males with completed compulsory education. Results for fixed effects of region of residence and sector of employment not reported.

Acknowledgements

The author would like to thank Klaus Nowotny, Liv Oswald, participants of the ERSA Congress 2010 and two anonymous referees for helpful comments

Funding

The author is grateful to the Austrian National Bank (Jubiläumfondsprojekt 13804) for financial support.

Notes

References

Benito

Oswald

A. J.

(2000) Commuting in Great Britain in the 1990’s. Department of Economics, University of Warwick.

Borjas

(1987) Self-selection and the earnings of immigrants, American Economic Review, 77, pp. 531–553.

Borjas

(1999) The economic analysis of immigration, in: Ashenfelter

Card

(Eds) Handbook of Labor Economics, Vol. 3A, pp. 1697–1760. Amsterdam: Elsevier.

Brücker

Trübswetter

(2007) Do the best go west? An analysis of the self-selection of employed east-west migrants in Germany, Empirica, 34(4), pp. 371–395.

Cabral-Vieira

J. A.

Couto

J. P. A.

Borges Tiago

M. T.

(2006) Regional differences in returns to education in Portugal. Paper presented at the 46th Congress of the Regional Science Association, Thessaly.

Chiswick

B. R.

(1999) Are immigrants positively self-selected?, American Economic Review, 89, pp. 181–185.

Chiswick

B. R.

(2000) Are immigrants favourably self-selected? An economic analysis. Discussion Paper No. 131, IZA, Bonn.

Eliasson

Lindgren

Westerlund

(2003) Geographical labour mobility: migration or commuting?, Regional Studies, 37, pp. 827–837.

Ferragina

A. M.

Pastore

(2008) Mind the gap: unemployment in the new EU regions, Journal of Economic Surveys, 22, pp. 73–113.

10.

Gottholmseder

Theurl

(2007) Determinants of cross-border commuting: do cross-border commuters within the household matter?, Journal of Borderland Studies, 22, pp. 97–112.

11.

Gries

Kraft

Pieck

(2011) Interregional migration, self-selection and the returns to education in Brazil, The Annals of Regional Science, 46, pp. 707–732.

12.

Hazans

(2003a) Returns to education in the Baltic countries. Manuscript, University of Latvia.

13.

Hazans

(2003b) Commuting in the Baltic states: patterns, determinants and gains. Working Paper No. B02/2003, Centre for European Integration Studies, University of Bonn.

14.

Huber

(2007) Regional labour market developments in transition: a survey of the empirical literature, European Journal of Comparative Economics, 4, pp. 263–298.

15.

Huber

(2011) The self-selection of commuters. Working Paper No. 2011-6, Working Papers in Economics and Finance, University of Salzburg.

16.

Huber

Nowotny

(2011) Moving across borders: who is willing to migrate or to commute?, Regional Studies, DOI: 10.1080/00343404.2011.624509.

17.

Hunt

(2004) Are migrants more skilled than non-migrants? Repeat, return and same-employer migrants, Canadian Journal of Economics, 37, pp. 830–849.

18.

Ommeren

J. van

Rietveld

Nijkamp

(1999) Job moving, residential moving, and commuting: a search perspective, Journal of Urban Economics, 46, pp. 230–253.

19.

Paci

Tiongson

E. R.

Walewski

Liwinski

Stoilkova

M. M.

(2007) Internal labour mobility in central Europe and the Baltic region. The World Bank, Washington, DC.

20.

Rouwendal

(1999) Spatial job search and commuting distances, Regional Science and Urban Economics, 29(4), pp. 491–517.

21.

Sorek

(2009) Migration costs, commuting costs and intercity population sorting, Regional Science and Urban Economics, 66, pp. 335–357.

22.

White

(1986) Sex differences in urban commuting patterns, American Economic Review, 76, pp. 368–372.