A comparative classification of labour market characteristics of British and Greek small islands

Introduction

This paper presents results of a comparative analysis of 60 British and 74 Greek small offshore islands. The principal purpose of the paper is to analyse the relationship between the labour market characteristics of the islands and two groups of variables identified by the research literature as being of particular importance in determining the structure of island economies: (a) industry sectoral specialisation and (b) a number of geographical characteristics in addition to ‘islandness’.

As shall be shown in the next section, previous research has revealed that most small islands are unable to develop diversified economies and must rely on niche market specialisation. The particular nature of the specialisation which emerges has a direct bearing not only on the labour market characteristics of an island (the particular focus of this paper), but also on its economic success. Geographical characteristics have also been revealed by previous research as having an important influence on the structure of island economies. ‘Islandness’ itself is, of course, a geographical characteristic in its own right. The literature, however, also identifies a number of other geographical characteristics which can influence island economic structures. European Union (EU) Cohesion Policy recognises three such variables as ‘specific geographical features’ worthy of special regional policy provision (European Commission, 2008; Monfort, 2009). These are geographical remoteness from the main EU market (through special Outermost Regions policy provision), sparsely populated areas and mountainous regions. Our analysis incorporates these ‘specific geographical features’ variables, but also includes measures of transport accessibility and island size (i.e. area and population), both of which are stressed in the existing research literature.

Separate cluster analyses are conducted on Greek and British island data sets to develop typologies of the two sets of islands. Three research questions are of particular importance:

By using cluster analysis to address the above questions we can contribute towards building an evidence base that could be used to support the analysis of the possible impacts of regional, national and European policies upon different types of islands. In addition, the cluster analysis results can also be used, in a more speculative manner, to inform debates about the types of new policies that may be needed, especially in the context of the current global financial crisis, the impacts of which are especially severe in Greece. In particular, by classifying islands into distinct clusters on the basis of labour market, sectoral specialisation and geographical data, we may gain more insights and have a better idea regarding the number of islands that might be affected by different aspects of the recession (e.g. remote and inaccessible islands that may be particularly vulnerable to transport subsidy cuts, whereas islands with a strong tourism sector would be negatively affected by weaker demand) and about the ways in which different island groups might be able to respond to the crisis. It is also noteworthy that by identifying the extent to which Greek islands may fall into the same kinds of clusters as the British islands (in order to address our third research question) we can highlight key differences as well as possible common challenges, contributing in this way to the on-going debates about how EU regional Cohesion Policy might affect island economies in both member states. In the concluding section of this paper we revisit these issues and we discuss what our results might imply for the islands of the two countries with regards to the eventual regional and local impacts of the global financial crisis.

It should also be noted that the work presented in this paper could provide the basis for further international comparisons of this kind in the EU in order to further enhance the evidence base for policy analysis. Therefore, a secondary purpose of the analysis, but one that is of great importance at the present time in Europe for the reasons outlined above, is to explore whether separate national data sets can be brought to bear in comparative analysis of the socio-economic characteristics of small islands. Comparative analysis of the socio-economic characteristics of islands across the member states of the EU has been greatly hampered by data limitations. Eurostat’s regional nomenclature (NUTS) makes it unusually difficult to conduct analysis even at fine levels of disaggregation, such as NUTS3, for three main reasons. First, small offshore islands are often grouped together with adjacent parts of the mainland littoral. Second, many NUTS3 regions comprise not a single island but rather island groups. Third, bigger islands are often sub-divided into more than one NUTS region.

Some indication of the scale of this problem can be seen from previous attempts to conduct comparative statistical analysis. Planistat Europe, Bradley Dunbar Associates (2003) identified 286 island territories within EU15, but was able to analyse only 19 of these. Monfort (2009) was able to assemble comparative data for 56 islands, but even this small number contained several cases (e.g. Crete, Sardinia) in which individual islands comprised more than one NUTS3 region. ESPON (2010) identified 362 EU27 islands with more than 50 inhabitants, but could analyse only 31 islands and island groups. Moncada et al. (2010) identified 5,116 inhabited and uninhabited EU25+3 islands (i.e. EU27, excluding Bulgaria and Romania, but including the EEA states of Iceland, Norway and Switzerland), but could analyse only 28 of these. It is instructive that 17 years after the pioneering Eurostat (1994) Portrait of the Islands study researchers are still having to plead for improved harmonised data: ‘A major finding of this survey … concerns the need to develop comparable data sets on these issues, which may be used to inform policy decisions at EU level’ (Moncada et al., 2010: 83).

The two countries, Greece and Britain, were selected partly because there are some close similarities (e.g. long distances from core EU markets), partly because they contain within them large numbers of islands (allowing statistical analysis to be conducted) and partly because both sets of islands exhibit diverse geographical characteristics. The paper is based largely upon data from the British and Greek population censuses.

The paper begins with a review of previous literature on island economies. The next section examines the data sets used. Following a discussion of the methodology and variables used, the results of a cluster analysis are set out. The conclusion reviews the findings of the paper and examines the limitations of using national data sets for comparative analysis.

Island economies

As research on island economies has gradually grown, it has become increasingly apparent that being a small island could not possibly be a universal handicap. There is now considerable evidence from research on global small states (the majority of which are islands) that many small island states have been able to produce economic performances at least as good as, and often better than, those of their larger, non-island counterparts (e.g. Armstrong and Read, 2000, 2003; Bertram and Karagedikli, 2004). These findings are mirrored by research on island regions within the EU. Monfort (2009: 7), for example, found that, while average GDP per capita in island regions was lower than the EU average and unemployment higher, ‘economic performance is … extremely diverse. In 2006, the richest island region was Åland (Finland) with a GDP per capita corresponding to 147% of the EU-27 average while Medio Campidano in Sardinia had a level of GDP per head of 54% of the EU-27 average’. Other studies of EU islands have invariably found the same sort of diversity, with island regions being found among both the richest and poorest of EU regions, demonstrating that islandness cannot be always a systematic handicap (ESPON, 2010).

Previous research on small islands has produced five sets of key findings that bear directly on the labour market characteristics which are the focus of this paper. The first of these is niche specialisation. Industrial diversification is a characteristic to which only the largest islands can aspire. We would therefore anticipate that the employment structure of the Greek and British offshore islands in our data set would reflect how specialised an island is, and other key labour market characteristics will be influenced by the particular niches developed. Niche specialisation is the result of the interaction of three forces: a small domestic market, limited factor supplies and high transport costs. It has long been known that small population size, coupled with higher transport costs, means that island domestic markets may be too small for local businesses to attain minimum efficient scale (MES) (Kuznets, 1960). In the context of modern economic theory, it is more difficult for small islands to develop vibrant industrial clusters. This is partly because of restricted local demand, one of the four Porter (1990) ‘diamond’ factors underpinning successful clusters. It is also, however, because the other main industrial cluster requirements, such as developing a critical mass of competing firms, are rarely fulfilled on small islands (Baldacchino and Fairbairn, 2006). Moreover, high transport costs (and other costs associated with sea transport such as higher packaging and insurance costs; Armstrong et al., 1993) mean that to be successful the niche sectors ideally need to be high value-to-weight ones.

The second key finding concerns the types of niche sectors typically found in small islands. The particular nature of an island’s niche specialisation is probably the most important factor influencing its labour market characteristics. An island which has retained a rural economy specialising in agriculture, for example, will have very different labour market characteristics from one specialising in tourism. Agriculture-based regions in the EU, particularly those in more remote areas, tend to have a rather adverse set of labour market characteristics, with slow growth or declining employment, an ageing workforce, higher unemployment, lower activity rates (especially female activity rates) and more seasonal unemployment but also higher self-employment rates (Copus et al., 2006). The global evidence overwhelmingly shows that specialisation in agriculture is associated with lower incomes and growth among small island economies (Armstrong and Read, 2000, 2003). Moreover, the very worst-performing islands at the global level (so-called MIRAB islands – those dependent on migrant remittances, aid and bureaucracy; Bertram, 2006; Bertram and Watters, 1985) typically have agriculture as their dominant production sector. However, within the EU it is likely that some islands, especially those with good climate and soils, may well be more successful. Of all small islands that retain agriculture as a successful niche sector, the most successful have been those that have adapted to high transport costs by developing high value-to-weight exports (e.g. food and drink products or textiles using local farm products such as Harris tweed on the Scottish Outer Hebrides).

The research literature, however, shows that the most successful small islands of all tend to have moved on from agricultural niches to one or more of four other types of niche specialisation: natural resources, tourism, financial services and high-value-added manufacturing. Of these, financial services is of least interest for the purposes of this paper for two reasons. First, while offshore financial centres are important for many successful global small island states (Hampton and Abbott, 1999), conditions do not exist within the EU for these to develop. It is noteworthy in this respect that none of the offshore financial centres identified on official lists is a within-EU island, and European offshore finance centres, such as the Channel Islands and Isle of Man, are not EU members (Rose and Spiegel, 2007) Second, only a very few within-EU islands have been able to develop successful niche financial services sectors, and where this has happened it is because of unusual historical factors or governance arrangements (e.g. Åland Islands, Cyprus). None of the Greek and British islands that are the focus of this paper have significant financial services sectors.

Export niches based on resource endowments are somewhat more common, and two are of particular importance for within-EU small islands: oil and gas, and fish. High-value resource endowments, such as oil, can at a stroke transform the economic prospects of a small island. In practice, within the EU only a very few small islands enjoy this advantage. Only the Shetland Islands within our Greek and British islands data set have significant oil and gas assets. The Shetland Islands are interesting in that they have become a classic case of how small islands can gain access to resource revenues and how the use of heritage funds can help to prevent inflation and tight labour markets whilst simultaneously protecting more traditional industries such as farming (Butler and Nelson, 1994; Coull, 2006). Fishing is the other resource-based niche sector of interest for this paper since all EU islands have access to fish stocks of some sort. Fishing as an export niche market is particularly important for some of the British offshore small islands and has the added advantages for small island economies of offering opportunities to (a) develop high-value/low-weight exports (e.g. live shellfish exports from the Outer Hebrides) and (b) move forward into the fast-growing fish-farming sector (Scottish Government, 2009a, 2009b). In reality, however, declining fish stocks and increasing national and EU quota and other restrictions on fishing have meant that the sector is now rarely as important as it once was for EU small islands.

This leaves tourism and high-value manufacturing. Tourism is associated with distinctive labour market characteristics such as high rates of self-employment and seasonal unemployment (McElroy and Hamma, 2010). It is also of particular importance both globally for small islands and within the EU, for three main reasons: (a) tourism is an unusual industry in that the customers pay their own transport costs, an important feature for an island economy, (b) it contains within it a number of fast-growth sub-sectors (e.g. cruise tourism, activity tourism) and (c) it allows small islands to exploit two of their main assets – natural environment and cultural distinctiveness. It comes as no surprise that the service sector is much larger in island regions than other EU regions, ‘reflecting the importance of the tourism industry for island regions … [and] its weight in the activity of island regions induces an almost mono-activity structure of the productive base’ (Monfort, 2009: 6). This is certainly true for some of the Greek and British islands, which are the focus of this paper. To the marine and coastal landscape advantages of all of the islands can be added climatic advantages for ‘sun and sand’ mass tourism of many of the Greek islands and the cultural and landscape distinctiveness of individual islands one from another (Andriotis, 2003; Coccossis, 2001; Gössling and Wall, 2007; Tsartas, 2003).

Finally, there are the various manufacturing niches. High transport costs, for both exports and the import of components, greatly limit the volume of manufacturing exports and hence its employment potential on small islands. However, where islands have been able to develop high value-to-weight manufacturing its contribution to wealth and GDP greatly outweighs its employment impact. The most successful island manufacturing niches tend to build upon a traditional industry by moving up the value chain, or else exploit the environment and cultural distinctiveness to produce unique food and drink products or culturally based craft and fashion products (Baldacchino, 2010).

The third set of key findings from the research literature highlights how the unusual demography of small islands, and in particular migration patterns, can affect labour market characteristics. Small population not only limits the size of the local market, but also places strict limits on the local labour supply. This can, of course be supplemented by in-migration, and many islands do seek to attract both seasonal and permanent migrants, but in practice there are limits on how far this can be taken. Moreover, two other types of migration flows often serve to exacerbate labour supply problems and create unusual labour market characteristics. First, out-migration from islands in search of better employment opportunities or lifestyle is frequently disproportionately made up of younger and better-educated persons, leaving many islands with an ageing population. In extreme cases this can lead to a downward spiral of economic activity and depopulation. Second, and to some degree offsetting out-migration from small islands, is the attractiveness of islands (a) for retirees, (b) for so-called ‘lifestyle’ migrants and (c) for second home owners. Categories (a) and (c) exacerbate labour supply constraints, while (b) can have the opposite effect since many ‘lifestyle’ migrants are both young and economically active.

The fourth set of key findings in the research literature is that islandness is rarely the sole geographical challenge faced by islands. It is typically a combination of islandness with other geographical characteristics that is important. Remoteness from the main EU markets, for example, may combine with islandness to make it harder to develop successful niche products. Being an archipelago rather than a single main island results in so-called ‘double insularity’ (or ‘islands off islands’; Bardolet and Sheldon, 2008), further breaking up the small local market. It is for this reason that the research in this paper incorporates a range of other geographical characteristics in addition to islandness.

Finally, the research literature has shown that small islands may be unusually vulnerable to sudden changes in economic circumstances. The lack of diversification in small island economies is not in itself a bad thing since a good living can be made by specialising, but specialisation does make an economy more vulnerable to sudden shifts in external economic factors. Islands are also frequently not only dependent on the export earnings from a small number of products, but also disproportionately dependent on a single overseas market (e.g. the mainland of the member state of which they are part, or at the global level the former colonial power). Economic vulnerability is, moreover, only one of a series of vulnerabilities which can affect the economic performance of islands (Atkins et al., 2000; Briguglio and Galea, 2003).

Data sets utilised

An initial decision that all researchers studying islands must make is how to define an ‘island’. This is surprisingly difficult. Between 1994 and 2003 the EU gradually honed its definition of an island (Eurostat, 1994; Planistat Europe, Bradley Dunbar Associates, 2003) to incorporate only those entities with:

Even this definition has proved very difficult to maintain since, while most researchers were happy to see very big island states such as Great Britain and Ireland excluded (using the ‘no capital city’ criterion) to allow attention to be concentrated on the more representative group of smaller islands, the accession of the small sovereign island states of Cyprus and Malta in 2004 has led to the rather clumsy definition now embodied in Article 52 of the Structural Funds and Cohesion Fund regulation of ‘island member states eligible under the Cohesion Fund, and other islands except those on which the capital of a member state is situated or which have a fixed link to the mainland’ (Monfort, 2009: 4).

In this study a somewhat wider definition of an ‘island’ than the Planistat Europe one is used, one that uses only three of the five criteria: a land area of at least 1 km², a population of at least 50 and no EU capital. There is clear logic in excluding uninhabited islands and those with very small population islands (on the grounds of small numbers and confidentiality problems). There is also clear logic in excluding islands which house an EU capital, since capital cities have major economic functions to which the vast majority of islands cannot aspire. There seems, however, little logic in imposing a minimum of 1 km distance from the continent, since the nature of transport costs is such that any separation, however short, will trigger the important trans-shipment cost elements. On the other hand, there is logic in excluding islands with fixed links to the mainland. This logic is, however, not as clear as it may appear at first sight. To begin with, many of the fixed links (both bridges and tunnels) are subject to tolls and can be subject to closure as a result of weather conditions, both of which act as barriers to economic integration. Finally, many islands with fixed links stubbornly retain distinctive ‘islandness’ characteristics long after the link is completed (Baldacchino, 2007). Based on the three chosen criteria and small area data availability pertaining to these criteria, some 60 British offshore islands and 74 Greek islands were incorporated into the data sets. The full lists of islands used in the research can be seen in Figures 1 and 2.

OPEN IN VIEWER

Figure 1.

Dendrogram of British island clusters.

OPEN IN VIEWER

Figure 2.

Dendrogram of Greek island clusters.

Even when one works with rich data sets such as the national population censuses, there are many problems encountered in constructing comparable data for small islands. Despite these limitations, it was possible to obtain data on a series of relevant variables from the national population censuses of Britain and Greece, as well as EU peripherality indexes from Copus (1999). Peripherality indexes seek to produce measures of the accessibility from any one region to all other EU regions.

Taking the British offshore islands first, the most recent census in Britain for which data are currently available was on 29 April 2001. Data output is available at six different spatial levels from the 2001 census: government office regions, unitary authorities, counties, districts, wards and output areas. Output areas were designed specifically for statistical purposes on the basis of 2001 census data and are built from postcode units. They are the lowest geographical level at which data may be retrieved from the British census.

Most of the islands in the British data set are Scottish islands (54 of the 60). There were in fact 96 inhabited individual islands in Scotland in 2001. However, only 54 islands and island groups are represented in the data set. Hence, even using census data, with all its richness, there remain problems in that the very smallest output units can still comprise groups of islands rather than individual islands. An island group may contain an individual island or a main island and other islands, which are so small that they have been merged in order to form an output area (Fleming, 2003). Of the 54 Scottish islands, 18 are island groups comprising more than one island. In England and Wales, Tresco is the sole island group.

Turning to the Greek offshore islands, the most recent Greek census for which data are available was conducted on 18 March 2001. Data may be retrieved from the national census of Greece at four spatial levels: regions, prefectures, municipalities and communities. ¹ Greece has 13 regions, four of which (Ionian Islands, Crete, North Aegean and South Aegean) are composed entirely of islands. Of the 55 prefectures, 10 are individual islands or island groups. Many larger islands are grouped with smaller adjacent islands to form a prefecture or municipality. Hence, data on the very smallest islands in the Greek data set are attainable only at the municipality or community level. In such cases, island groups are disaggregated at the next lower level. For example, at the time when the analysis for this paper was carried out the island Chios was a prefecture in the North Aegean region, whilst the islands Psara and Inousses were municipalities within the prefecture of Chios. For the purposes of our analysis the data for situations such as Chios were disaggregated to the municipality level to obtain individual data for islands such as Psara and Inousses.

Methodology and variables

Typologies have been developed using cluster analysis. Cluster analysis is a classificatory technique particularly valuable for constructing typologies with data sets comprising many variables and ‘cases’ (islands in this study). The analysis uses Ward’s method, an agglomerative, hierarchical technique, with squared Euclidean distance as the (dis)similarity measure. Ward’s method has long been the most popular of the cluster analysis algorithms (Everitt, 1993) and for this reason it is also the default method in the main software packages. It also has the advantage in that ‘it will generally find tight minimum variance spherical clusters’ (Wishart, 1987: 91).

As noted in the introduction, three sets of variables have been developed for use in the analysis: labour market characteristics, geographical characteristics and sectoral specialisation variables. There were three reasons for the focus on labour market characteristics. First, population censuses provide rich information sets on labour market characteristics. Second, labour market characteristics are important measures in their own right, and in addition as noted earlier the research literature suggests that islands often have very distinctive and differing labour markets. Finally, definitions of key labour market characteristics are much more similar, though not identical, across countries than for many other economic, social and demographic variables.

Although extensive in their coverage of labour market characteristics, certain desirable statistics, such as the extent of the informal economy and multi-tasking (both thought to be unusually extensive in island economies) were unfortunately not available from the national censuses. The variables used are therefore the more traditional types (e.g. activity rates, unemployment rates). This is an advantage in that definitions are more comparable across the two countries than for more unusual variables, but disadvantageous in that a few important other labour market characteristics have had to be excluded.

As noted earlier, in addition to the labour market variables, a number of variables measuring geographical characteristics of the islands were also included in the analysis. As the previous section has shown, the research literature confirms that it is dangerous to consider islandness in isolation from other geographical characteristics. Most islands exhibit a combination of geographical characteristics, and this combination of characteristics can pose as great a challenge as insularity itself.

As Table 1 shows, it proved possible to obtain useable data for 13 geographical variables to be included in the cluster analyses. The 13 variables contain almost all of the geographical variables identified by previous research as being potential determinants of island economic characteristics. Since the data sets are, by definition, all islands, there is no need for a separate islandness variable. Nor, for obvious reasons, is there a need for two other geographical variables identified by global economic growth analysis (Ahlfeld et al., 2005; Sachs, 2003): tropical climate and whether land-locked. The first two variables in Table 1 (land area and population) seek to measure size. Population density has been included since low population densities are identified by the EU Cohesion Policy as an important geographical characteristic.

OPEN IN VIEWER

Table 1.

Description of variables used in the cluster analysis.

Variable	Definition
Geographic characteristics	British census	Greek census
Land area	Surface area (km2)	Surface area (km2)
Population	Number of permanent residents	Number of permanent residents
Population density	Residents per km2	Residents per km2
Access to national capital city	Great circle distance in km to London	Great circle distance in km to Athens
Access to EU	Great circle distance in km to Brussels	Great circle distance in km to Brussels
Perindex-ECU	Copus’s (1999) peripherality index for the NUTS3 region of which the island is a part, using GDP as the mass variable, in current prices (euros)	Copus’s (1999) peripherality index for the NUTS3 region of which the island is a part, using GDP as the mass variable, in current prices (euros)
% mountainous	Binary variable. Over 50% of land area mountainous, as defined by UNEP World Conservation Centre (over 300 m altitude, slopes ≥ 2°, elevation range > 300 m	Binary variable. Over 50% of land area mountainous, as defined by UNEP World Conservation Centre (over 300 m altitude, slopes ≥ 2°, elevation range > 300 m
Air national capital city	Binary variable. Direct air link (at least one scheduled service per week) between island and any one of the four London airports	Binary variable. Direct air link (at least one scheduled service per week) between island and Athens airport
Air mainland	Binary variable. Direct air link (at least one scheduled service per week) between island and a mainland airport other than the capital city	Binary variable. Direct air link (at least one scheduled service per week) between island and a mainland airport other than the capital city
Air international	Binary variable. Direct air link (at least one scheduled service per week) to at least one airport in another country (not necessarily EU state)	Binary variable. Direct air link (at least one scheduled service per week) to at least one airport in another country (not necessarily EU state)
Sea to national capital city	Binary variable. Direct sea link (at least one scheduled service per week) between island and London	Binary variable. Direct sea link (at least one scheduled service per week) between island and Piraeus or Rafina
Sea mainland	Binary variable. Direct sea link (at least one scheduled service per week) between island and a mainland port other than the capital city	Binary variable. Direct sea link (at least one scheduled service per week) between island and a mainland port other than Piraeus or Rafina
Sea international	Binary variable. Direct sea link (at least one scheduled service per week) to at least one port in another country (not necessarily EU state)	Binary variable. Direct sea link (at least one scheduled service per week) to at least one port in another country (not necessarily EU state)
Labour market characteristics
Active population	Total economically active population (age 16–74) as a percentage of total population	Total economically active population (age 10 and above) as a percentage of total population
Active male	Total economically active males (age 16–74) as percentage of total active population	Total economically active males (age 10 and above) as percentage of total active population
Active female	Total economically active females (age 16–74) as percentage of total active population	Total economically active females (age 10 and above) as percentage of total active population
Working age population	Population aged 15–64 as percentage of total population	Population aged 15–64 as percentage of total population
Employed	All persons aged 16–74 in employment as a percentage of total active population	All persons aged 10 and above in employment as a percentage of total active population
Self-employed	All persons aged 16–74 self-employed as a percentage of total employed	All persons aged 10 and above self-employed as a percentage of total employed
Unemployed	All persons aged 16–74 unemployed as a percentage of total employed plus unemployed	All persons aged 10 and above unemployed as a percentage of total employed plus unemployed
Sectoral specialisation variables
Agriculture	All persons aged 16–74 in employment working in agriculture, hunting, forestry and fishing as a percentage of total active population	All persons aged 10 and above employed in agriculture, animal breeding, forestry and fishing as a percentage of total active population
Manufacturing	All persons aged 16–74 in employment working in manufacturing as a percentage of total active population	All persons aged 10 and above employed in manufacturing as a percentage of total active population
Services	All persons aged 16–74 in employment working in services (retail and distribution, hotel and restaurants, transport, storage and communication, financial intermediation and real estate, public administration and defence, education, and health and social work) as a percentage of total active population	All persons aged 10 and above employed in services (retail and distribution, hotel and restaurants, transport, storage and communication, financial intermediation and real estate, public administration and defence, education, and health and social work) as a percentage of total active population
Other variables
Occupancy	Total occupied dwellings as a percentage of total dwellings	Total occupied dwellings as a percentage of total dwellings

Nine geographical variables were developed to capture different aspects of ‘remoteness’ and ‘accessibility’. The former are distance measures (using great circle distances) from main national and EU markets, whereas the latter seek to measure accessibility through transport networks (not necessarily the same remoteness since a highly peripheral EU island may nevertheless have well-developed sea or air transport links). Table 1 shows that it has been possible to incorporate three different remoteness variables (distance to national capital – the main national market in both Greece and Britain), distance to Brussels (Brussels being a surrogate for the centre of the EU market) and Copus’s peripherality index for the NUTS3 region of which the island is a part. Finally, six accessibility variables were incorporated, seeking to pick up some of the important nuances of transport network links. These are all binary variables and indicate the presence/absence of regular direct air and sea links with, respectively, the national capital city, the mainland and international destinations.

Turning next to the labour market characteristics of the islands, it proved possible to obtain reasonably comparable data for 10 variables. As Table 1 shows, these include most of the popular traditional labour market variables: activity rates, male and female economic activity proportions, working-age populations, employment rates, unemployment rates and levels of self-employment.

The third and final set of variables it proved possible to include in the analysis comprised measures of sectoral specialisation. Ideally one would wish to work with very fine levels of disaggregation so that particular very narrow niche specialisation situations could be identified. Unfortunately the available data do not permit this, particularly for the very important tourism sector categories. The sectoral specialisation variables therefore comprise broad employment structure variables (i.e. agriculture, manufacturing and services).

There is one final variable shown in Table 1 that is not a labour market, geographical or sectoral characteristic: domestic property occupation rates. Although virtually no comparative research exists, there are strong reasons to suspect that second home ownership may be much more extensive among Greek islands than in Britain. ² This is for two reasons: (a) the national capital (Athens) is very close to a large number of Greek islands, and (b) the family home is particularly important in Greece (Birdwell-Pheasant and Lawrence-Zúñiga, 1999; Kenna, 1976). In addition, there is a relatively high number of unoccupied dwellings in Greece that are not actually occupied as primary residences but are privately owned (Earley, 2004), which might be explained by a tradition for families to retain ownership of the family home and use it for vacation and recreational purposes. Extensive second-home ownership is known to generate a number of major economic impacts on host communities, impacts so severe that special policies are in place to ameliorate them in many EU countries (Casado-Diaz, 1999; Gallent, 2007; Gartner, 1987; Girard and Gartner, 1993). We have taken the opportunity provided by available comparative data in the two censuses to explore this issue further.

Results of the cluster analysis

The results of the cluster analysis using Ward’s method are set out as dendrograms in Figures 1 and 2, whereas Figures 3 to 7 present maps of the islands by cluster membership Two separate cluster analyses were conducted. The British data set comprised 24 variables and 60 islands whilst the Greek data set comprised the same 24 variables, but this time for 74 Greek islands.

OPEN IN VIEWER

Figure 3.

Map of British Islands – Shetland and Orkney.

OPEN IN VIEWER

Figure 4.

Map of British Islands – west of Scotland.

OPEN IN VIEWER

Figure 5.

Maps of British islands – England and Wales.

OPEN IN VIEWER

Figure 6.

Map of Greek islands in the Aegean sea.

OPEN IN VIEWER

Figure 7.

Map of Greek islands in the Ionian sea.

Taking the British offshore islands first, it should be noted that there is no single ‘best’ solution to the choice of the number of clusters. Indeed, in cluster analysis one can work with several different levels of disaggregation if one wishes. It is most sensible to seek the number of clusters that maximises within-cluster homogeneity and between-cluster differences (across the 24 clustering variables). Examination of Figure 1 suggests that a five-cluster solution is the most appropriate. The degree of difference between any pair of clusters is shown by the length of the dendrogram horizontal bars (the horizontal axis measures the dissimilarity coefficient). The cluster analysis identifies which islands fall within each of the five clusters; this is shown by the numbers alongside each island name on Figure 1 (e.g. the 29 islands running down the figure from North Uist at the top to Eigg are the members of cluster 1, while the 17 islands running from Bressay to Fetlar on Figure 1 form the members of cluster 2, and so on).

In order to establish the characteristics of each cluster, the average values for each of the 24 variables are calculated across the islands in each cluster in turn. These values are set out in columns 2–6 of Table 2. The final column of Table 2 presents the overall mean value for each variable, this time calculated across all 60 islands taken as a whole. Cluster mean values that are greater than the overall mean are picked out in Table 2 in bold numbers.

OPEN IN VIEWER

Table 2.

Mean values, by variable, for the five British island clusters.

Variables	Cluster 1: Sparsely populated, accessible to mainland and dependent on services	Cluster 2: Small, remote, inaccessible and agriculture dependent	Cluster 3: Larger, remote, but accessible and diversified	Cluster 4: Very small, relatively close to London and the EU and highly dependent on services	Cluster 5: Large, least remote, accessible to mainland and diversified	Overall mean values
1. Geographical characteristics
Land area	265.6	49.6	254.2	4.2	371.0	186.8
Population	2049.4	354.2	5666.8	455.6	70,316.0	5211.4
Population density	17.8	12.0	26.0	104.9	439.6	45.3
Access to London	725.7	915.8	948.7	509.3	275.5	761.3
Access to EU	970.0	1062.9	1085.3	794.5	558.4	972.6
Perindex-ECU	85.2	94.4	95.9	77.2	67.1	87.3
% mountainous	21	0	0	0	0	10
% air London	0	0	50	0	0	5
% air mainland	55	0	100	40	0	40
% air international	0	0	0	0	0	0
% sea London	0	0	0	0	0	0
% sea mainland	90	0	100	20	100	60
% sea international	0	0	83	0	33	10
2. Labour market
Active population	47.3	51.7	54.2	69.8	43.2	50.8
Active male	55.5	58.2	57.3	51.4	54.9	56.1
Active female	44.5	41.8	42.7	48.6	45.1	43.9
Working age population	61.9	64.2	66.3	72.1	62.3	63.9
Employed	92.4	94.1	97.8	98.0	93.0	93.9
Unemployed	7.5	5.8	2.0	1.1	6.6	5.9
Self-employed	29.8	38.0	18.9	35.9	13.8	30.8
3. Sectoral specialisation
Agriculture	15.0	25.8	8.6	8.2	2.1	16.5
Manufacturing	7.3	7.9	11.1	4.1	17.4	8.5
Services	55.6	47.3	62.6	70.4	60.2	55.4
4. Other variables
Occupancy	74.8	79.6	91.9	65.1	93.9	77.7
N	29	17	6	5	3	60

The average values set out in Table 2 facilitate the labelling of the clusters. For instance, cluster 1 of the British islands has been labelled as ‘Sparsely populated, accessible to mainland and dependent on services’. As can be seen from Table 2, this cluster exhibits a distinctive, and rather unfavourable, combination of labour market characteristics. Male economic activity (economically active males as a percentage of total males plus females economically active) is low (55.5%), although the size of the service sector is associated with a high female activity percentage (44.5%). However, the overall employment rate is low (92.4%), the unemployment rate is the highest of all five clusters (7.5%) and the population is an ageing one as indicated by the low percentage of population of working age (61.9%), suggesting out-migration. The most unusual labour market characteristic for this cluster is the low self-employment rate (29.8%). One would have expected the dominance of the service sector to have been associated with high self-employment. This suggests either that the sector is dominated by public sector services, or that tourism on these islands may not be as supportive of self-employment as is normal in tourism economies (e.g. external ownership, large hotels), or some combination of the two. Cluster 1 is the single largest cluster of British islands, comprising 29 of the 60 islands. Figure 1 gives the names of the islands in each cluster. Virtually all of the islands within cluster 1 lie off the Scottish west coast. Interestingly, property occupancy rates at 74.8% are lower than average. Despite the distances most of these islands are from the main British population centres, second-home ownership would appear to be an issue of some importance. It should be noted in this context that the occupancy rates for all the British islands are much higher than those we find for the Greek islands. This is almost certainly a reflection of two factors. First, as noted earlier, many Greek families have a tradition of retaining the traditional extended family home, especially those on the islands, which is then used for summer vacation purposes. At any point in time (as at the census date, for example), occupancy rates are therefore lower than in Britain. Second, many more Greek than British offshore islands have highly developed mass ‘sun and sand’ tourism industries, again contributing to lower occupancy rates at census dates in the spring.

Figure 1 lists the 17 British islands that comprise cluster 2, labelled as ‘Small, remote, inaccessible and agriculture dependent’. Agriculture accounts, on average, for a significant 25.8% of all employment for these small islands. The islands in this cluster exhibit the labour market characteristics of traditional agricultural economies: high male activity (58.2%) and levels of self-employment (38.0%), but low female activity (41.8%). Unemployment rates are relatively low (5.8%). The islands are remote, both from London and also from the wider EU, and also have virtually no direct air or sea links anywhere (having access to the wider world only via other islands). From examination of the list of islands falling within cluster 2 shown on Figure 1, it can be seen that many are the smaller members of the Orkney and Shetland island groups to the north of the Scottish mainland. Other members (e.g. Eigg, Coll) are Scottish west coast islands, but again lying offshore from other, larger islands. The combination of small size, remoteness and lack of direct transport links presumably accounts for the continued importance of the agricultural economy.

Cluster 3 comprises only six islands, which we have characterised as ‘Larger, remote, but accessible, and diversified’. These islands have very different labour market characteristics from clusters 1 and 2. They exhibit high employment rates (97.8%), large working-age populations (66.3%), high male activity (57.3%) and low unemployment (a mere 2.0%). The islands have diversified from agriculture into both manufacturing (11.1%) and services (62.6%). Only female activity levels (at 42.7%) and self-employment rates (18.9%) are relatively low. Whilst relatively remote, both from London and from the wider EU, these islands have excellent direct air access to both London and the mainland, and excellent sea access links to the mainland. As can be seen from Figure 1, this cluster is dominated by the larger islands of Orkney and Shetland. There is clearly an oil industry effect here, since Shetland and Orkney are the two Scottish NUTS3 regions which have benefited most from offshore oil and gas. Orkney also has a thriving tourism industry based on cultural and historical assets. As shall be shown later, there is no Greek equivalent to this cluster.

The final two clusters (clusters 4 and 5) are both small (containing five and three islands respectively) and also are clearly very distinctive from the other three clusters given their free-standing position on Figure 1. Taking cluster 4 first, Figure 1 shows that the five islands in this cluster are all, with the sole exception of Iona, in England. St Agnes, St Martin’s, St Mary’s and Tresco make up the English contingent and are all in the Scilly Islands off the coast of Cornwall. They are highly successful tourism islands. Iona is a world-famous religious heritage island off the west coast of Scotland with huge visitor numbers each year. It is interesting that of all the many Scottish islands the cluster analysis has grouped this particular one with the Scilly Islands. These appear to be classic small island tourist economies (SITE; McElroy and Hamma, 2010). The cluster is dominated by the four Scilly Islands. The cluster exhibits classic labour market characteristics for tourism economies: low male activity levels (51.4%) but very high female activity (at 48.6% the highest of all clusters), and high rates of self-employment (35.9%), total employment (98.0%) and working-age population (72.1%). Most noteworthy of all is the very low unemployment rate (1.1%). Finally, the property occupancy rate for this cluster is the lowest of all five clusters at 65.1%, almost certainly a reflection of large-scale second-home ownership.

Finally, we come to cluster 5. This is the smallest of the five clusters, comprising only three islands, Anglesey (off North Wales), the Isle of Wight (off the south coast of England) and the Isle of Walney (off North West England). The results are dominated by the two very large islands of Anglesey and the Isle of Wight. As a group these are large islands (both in area and populations), with high population densities. In fact, they are the biggest and most densely populated of all the British offshore islands by a wide margin. They are also the least remote of all the islands from both London and the wider EU and have comprehensive sea links to the mainland. We have given this cluster the label ‘Large, least remote, accessible to mainland and diversified’. They are diversified because their agricultural employment shares are extremely low (2.1%) as a result of their having diversified into both manufacturing (17.4%, higher than any of the other clusters) and services (60.2%).

Turning to the labour market characteristics of cluster 5, we observe a most unusual pattern. Of all the indicators, only female activity shows above-average strength (45.1%). Male activity levels are very low (54.9%). The working-age population is also low (62.3%), suggesting a loss of young people over time, and the overall employment rate is also low (93.0%). Unemployment rates are very high (6.6%). This stands in stark contrast to the extremely low unemployment rate of the cluster 5 small tourism islands (1.1%). Finally, self-employment rates are also extremely low (13.8%), the lowest of all five clusters. Why the biggest and least remote of all the British offshore islands have such a poor array of labour market characteristics is an issue worthy of more research. Two possible explanations are (a) the long-term decline in traditional extended family holidays in Britain and (b) the ‘costa-del-dole’ phenomenon in traditional British seaside resorts, with disproportionate numbers of benefits claimants being attracted to these types of communities.

Turning to the Greek islands, the cluster analysis again identified five clusters (see Figure 2). Table 3 sets out the cluster mean values for the 24 variables that were again used. The most distinctive feature of Table 3 is cluster 5. This contains a single island, Crete. The singling out of Crete is, in some ways, reassuring. Crete is by far the biggest island both in terms of land area and population. It is a highly accessible island (both to Athens and internationally), with excellent airport connections for tourists from mainland Greece and northern Europe. Its size also means that it is highly diversified. These all single the island out as unique within the data set.

OPEN IN VIEWER

Table 3.

Mean values, by variable, for the five Greek island clusters.

Variables	Cluster 1: Small and agriculture dependent, good accessibility to mainland but not Athens	Cluster 2: Small, remote, inaccessible and agriculture dependent	Cluster 3: Closest to Athens, accessible to Athens (sea and air) and the mainland (sea) and diversified	Cluster 3a	Cluster 3b	Cluster 4: Large, rather remote but highly accessible (Athens and EU) and diversified	Cluster 5: Crete	Overall mean values (figures in brackets exclude Crete)
1. Geographical characteristics
Land area	85.3	61.5	306.3	604.6	175.8	801.5	8336.0	337.4 (227.9)
Population	2740.0	2141.9	16512.0	38421.6	6926.6	57396.9	601131.0	20878.3 (12929.6)
Population density	41.9	35.7	79.3	132.5	56.0	80.7	72.0	56.7 (56.5)
Access to Athens	236.5	323.5	126.6	57.0	157.0	304.1	317.0	230.7 (229.5)
Access to EU	2083.6	2370.5	2169.7	2098.2	2201.0	2113.7	2381.7	2183.5 (2180.8)
Perindex-ECU	93.5	99.0	94.0	89.3	96.1	96.7	97.8	95.3 (95.3)
% mountainous	60	59	74	29	94	63	100	65 (64)
% air Athens	8	18	44	0	63	100	100	32 (32)
% air mainland	0	0	9	0	13	75	100	12 (11)
% air international	0	6	17	0	25	100	100	19 (18)
% sea Athens	4	0	74	88	69	25	100	28 (27)
% sea mainland	32	0	48	88	31	38	100	31 (30)
% sea international	4	0	0	0	0	100	100	14 (12)
2. Labour market
Active population	36.0	36.2	39.8	37.3	40.8	40.6	44.0	37.8 (37.7)
Active male	70.1	75.5	68.9	70.3	68.3	64.6	62.0	70.3 (70.4)
Active female	29.9	24.5	31.1	29.7	31.7	35.5	38.0	29.7 (29.6)
Working age population	60.2	62.7	65.1	67.7	64.0	66.3	67.0	63.1 (63.0)
Employed	88.8	82.3	89.5	87.6	90.4	86.9	89.0	87.3 (87.3)
Unemployed	8.7	15.0	10.4	12.4	9.5	13.1	11.0	11.2 (11.2)
Self-employed	27.1	22.6	18.2	14.9	19.7	19.8	23.0	22.4 (22.4)
3. Sectoral specialisation
Agriculture	24.1	19.2	10.4	7.0	11.9	13.9	22.0	17.6 (17.5)
Manufacturing	3.2	3.5	6.3	7.7	5.6	5.2	6.0	4.5 (4.5)
Services	41.4	44.9	51.6	56.1	49.7	55.9	51.0	47.1 (47.0)
4. Other variables
Occupancy	44.5	49.2	42.1	44.6	41.1	58.9	65.0	46.7 (46.4)
N	25	17	23	7	16	8	1	74 (73)

Leaving Crete to one side, there are good reasons for taking clusters 1 and 2 together to begin with. Both are large groups (25 and 17 islands respectively). Cluster 2 is an interesting one since it has a close British equivalent. We have labelled cluster 2 as ‘Small, remote, inaccessible and agriculture dependent’. The closest equivalent to this cluster in Britain is cluster 2. In the Greek cluster 2, agriculture accounts for a high 19.2% of all employment. The labour market characteristics are typically agricultural: a high male activity level (75.5%), low female activity (24.5%) and high self employment (22.6%). Here, unlike British cluster 2, the Greek cluster 2 islands have a low working-age population (62.7%), low employment rate (82.3%) and exceptionally high unemployment (at 15.0% the highest of any Greek cluster). The Greek cluster 2 islands are doing less well than their British cluster 2 counterparts. Most of the Greek cluster 2 islands are in the Dodecanese region, located in the southeast Aegean Sea and off the southwest coast of Turkey. In other words, these are as remote from Athens as it is possible to be. Moreover, the British cluster 2 counterparts are Scottish islands facing onto the Atlantic Ocean, whereas the Greek cluster 2 islands are far from the Greek mainland and face onto Turkey, which is a non-EU country and in 2001 had very limited trade and other links with the Greek islands. The parallels are close ones.

When we turn to the Greek cluster 1, however, more significant differences between the Greek and British cases begin to emerge. Superficially, the Greek cluster 1 would also (like the Greek cluster 2) appear to have its closest counterparts in the British cluster 2. The Greek cluster 1 has been named as ‘Small and agriculture dependent, good accessibility to the mainland but not Athens’. In other words, these islands are again small and agriculture dependent. Whilst this is fairly similar to the British cluster 2 islands, there is an important difference. The Greek cluster 1 islands are less remote (Athens is 236.5 km away on average for Greek cluster 1 islands, whereas British cluster 2 islands are a massive 915.8 km from London). Moreover, although it is true that the Greek cluster 1 islands have relatively poor direct transport links to Athens, almost a third of them do have direct sea links to the mainland. By contrast, absolutely none of the British cluster 2 islands have direct sea (or air) links to either the mainland or London.

The better access to the mainland would appear to have had some interesting effects on cluster 1 Greek islands. They exhibit some of the classic characteristics of agriculture-dependent economies: high rates of self-employment (27.1%) and high employment rates (88.8%). However, the female activity level is also high (29.9%), and in stark contrast to Greek cluster 2 islands there is low unemployment (at 8.7%, the lowest of all five Greek island clusters). Perhaps more importantly, unlike the British cluster 2 islands (and the Greek cluster 2 islands), which are also agriculture dependent, their property occupancy rates are extremely low (44.5% the second lowest of all the Greek clusters). This would suggest a high degree of second-home ownership exists in Greek cluster 1 islands. The mainland appears to be casting its shadow over these islands. The majority of Greek cluster 1 islands are located in the Cyclades to the southeast of mainland Greece.

Turning to the Greek cluster 3, Table 3 shows this to be another large group, made up of 23 islands. We have labelled this as ‘Closest to Athens, accessible to Athens (sea and air) and mainland (sea) and diversified’. This cluster superficially has a fairly close British counterpart (cluster 5 in Table 2). Like the British cluster 5, the Greek cluster 3 is the least remote of all from the capital city (Athens is only 126.6 km away on average), with good direct transport links to Athens and the mainland, and economies that have diversified away from agriculture. There, however, the similarities end. The British cluster 5 was made up of only three large English and Welsh islands, which, whilst relatively close to London, nevertheless exhibited very unfavourable labour market characteristics. By contrast, there are no fewer than 23 islands in the Greek cluster 3, showing that many more Greek islands lie close to the capital city (and with good direct connections to it) than is the case in Britain. The labour market characteristics of the Greek cluster 3 are distinctive. Unemployment rates are relatively low (10.4%), while the female activity level is high (31.1%), as are the employment rate (89.5%) and working population (65.1%). On the other hand, male activity levels are low (68.9%) and so too is self-employment (18.2%). The Greek cluster 3 islands also have the smallest agricultural sectors of all the Greek clusters (10.4%) and the lowest rate of property occupancy of all (42.1%).

Examination of Figure 2 shows that cluster 3 is sub-divided into two fairly distinct sub-clusters (labelled sub-clusters 3a and 3b on Figure 2), and that sub-cluster 3a comprises islands lying very close indeed to Athens (Aegina, Poros, Agistri, Spetses, Hydra, Salamina and Evia). By contrast, the remainder of cluster 3 (sub-cluster 3b) are islands further away from Athens (mostly in the Cyclades). This offers the opportunity to examine whether extremely close proximity to the capital city (something not found for any of the British offshore islands) may be having distinctive effects on the labour market. Table 3 therefore presents mean variable values not only for cluster 3 as a whole (i.e. all 23 islands) but also separately for the seven ‘inshore’ islands of sub-cluster 3a and the remaining 16 islands that make up sub-cluster 3b.

Sub-cluster 3a is the most unexpected and interesting of all the Greek island groups. As Table 3 shows, sub-cluster 3a comprises islands that have excellent sea links to both the mainland and Athens itself (88% with a direct link), but almost no sea or air links anywhere else. It could be argued that these islands are not major destinations for international tourists. On the other hand, these are islands which are popular locations for second-home owners from Athens and the mainland (the average occupancy rate is only 44.6%). Moreover, they are accessible enough to attract day trippers, weekend and overnight visitors, and Salamina acts as a commuter dormitory area for Athens. The result appears to be a most unusual set of labour market indicators. Female activity levels are high (29.7%), as are employment rates (87.6%) and the working population (67.7%). However, unemployment is very high (12.4%), self-employment rates are low (14.9%) and the male activity level is low (70.3%). In the case of the sub-cluster 3a islands, one can perhaps argue that they are too close for comfort to Athens. The influx of visitors at a weekend or at even more infrequent intervals can leave communities with little in the way of an income stream during the week and across many months of the year.

Sub-cluster 3b islands are mostly smaller Cyclades islands further away from Athens (157 km on average), although with quite good air and sea links. This group includes some of the most successful mass ‘sun and sand’ tourism islands (e.g. Mykonos, Paros) and this may explain the somewhat higher employment rates, activity rate and female economic activity, not to mention a higher rate of self-employment than sub-cluster 3a. The sub-cluster 3b islands also seem to have been able to develop a little more manufacturing than sub-cluster 3a.

The remaining Greek island cluster (cluster 4) is a small cluster comprising only eight islands and we have labelled it ‘Large, rather remote but highly accessible (Athens and EU) and diversified’. It comprises larger islands (both in population and in land area) which are relatively remote from both Athens and the rest of the EU. Nevertheless, these islands have superb direct sea and air links both with Athens and internationally. They have larger than average manufacturing (5.2%) and service sector (58.9%) employment, the latter almost certainly being tourism and recreational activity, although the data set contains no direct evidence of this. Scrutiny of the membership of cluster 4 shows that these are the successful large tourist islands of both the Aegean (e.g. Rhodes, Chios, Lesvos, Kos, Samos) and Ionian seas (Kefalonia, Zakynthos and Kerkyra). It could be argued that the success of the tourism sector is reflected in the distinctive set of labour market characteristics. The islands have higher than average female economic activity (35.5%) with relatively low male activity levels (64.6%). On the downside, unemployment rates are rather high (13.1%) and employment rates correspondingly low (86.9%).

Conclusion

This paper has drawn upon cluster analysis to develop separate typologies for Greek and British offshore islands using variables which combine a variety of labour market, sectoral specialisation and geographical characteristics. Taking in turn each of the main research questions set out in the introduction to the paper:

In the introduction we argued that the paper would be an exploratory analysis seeking to develop comparative research using differing national data sets as a means of overcoming some of the weaknesses of harmonised Eurostat data. Our results suggest that it does indeed seem possible to undertake meaningful comparative research of this type and we would urge other researchers to seek to do this, particularly once the results of the 2011 censuses become available. There are, however, limits to how far this type of work can be taken, for we would hesitate ourselves to go beyond bilateral comparative work. Differences in the manner in which data are collected between EU member states will also always place limits on the number of variables that can be compared. There does, however, seem to be enough scope to conduct useful research that is not possible at the pan-EU level.

The EU fortunately exhibits an ‘accident of geography’ that should facilitate further work of this kind. The overwhelming majority of EU islands are concentrated in just five countries. Within each of these countries there are sufficient numbers to produce appropriate degrees of freedom for statistical analysis to be undertaken.

Finally, given the extraordinary severity of the economic crisis that has engulfed Europe as a whole, and Greece in particular, it is interesting to speculate a little on what the results of this paper may imply for the islands of the two countries. Great caution must, of course, be exercised in reaching conclusions. The data on which this paper is based date to 2001, whereas the financial crisis emerged only after 2007. In addition, the crisis clearly still has some considerable distance to run. While the initial financial crisis has given way to economic recession and an initial recovery, the EU economy has been hit by a further sovereign debt crisis whose full implications have yet to be felt.

What, if anything, do our results suggest may be the eventual impact of the crisis on the two sets of islands? A key consideration in the likely impact of the crisis on island economies is their industrial structure. Because of the high degree of sectoral specialisation of the islands (Crete being the major exception), they are particularly vulnerable to any industrial asymmetry in the impact of the different phases of the recession. Since manufacturing is of only very limited importance on the islands (again Crete being an exception), attention is focused on three sectors. These are agriculture, private sector services (especially mass ‘sun and sand’ tourism in Greece and more specialised ‘activity tourism’ in the British offshore islands) and public sector services, the last being disproportionately represented on almost all small islands, but particularly those with regional and local administrative centres on them. Hadjimichalis (2011) has shown that across the EU the crisis initially hit ‘three highly connected sectors: banks, real estate and private and public debt’ (p.256). In this respect, the British and Greek offshore islands were probably initially little affected by this first phase of the crisis because, unlike their global small island state counterparts, offshore finance centres do not exist within the EU, and most of our two samples of islands have little in the way of private financial services. It has therefore probably been in the second and third phases of the crisis that the islands have been, and are being, most affected. The second phase was the recession triggered by the initial financial crisis. In Britain, the initial impacts (2008–2010) of the recession seem to have fallen on manufacturing and the wholesale and retail sectors (Office for National Statistics, 2011). The relative paucity of manufacturing in the British offshore islands would seem to have cushioned some of this initial impact there, and this is perhaps borne out by the finding that it has been regions such as the Midlands, North West and Yorkshire and the Humber (i.e. regions with relatively large manufacturing sectors) that were initially most badly affected (Office for National Statistics, 2011). It is therefore in the current third phase of the crisis that perhaps the major impact on the British offshore islands is likely to fall. In the post-2009 period, while British manufacturing benefited from the depreciation in sterling, this has probably been of little direct benefit to the islands given their sectoral specialisation. It is therefore to tourism and the public sector that the British offshore islands face their greatest threats. Here the omens are not good. It is possible that the high-value ‘activity tourism’ on which the British offshore islands rely may prove to be a victim of the on-going period of austerity. It can be argued that this may be the case with a lot of the islands in British cluster 1 (‘Sparsely populated, accessible to mainland and dependent on services’) as well as some of the islands in cluster 3 (‘Larger, remote, but accessible, and diversified’) and especially the larger islands of Orkney, which have a thriving tourism industry. Moreover, the reliance of many of the islands on a disproportionately large public sector and on an array of publicly funded subsidies (notably marine and air transport subsidies) makes the islands potentially very vulnerable. Based on our analysis, it could be argued that the less accessible and least diversified islands in British cluster 2 (‘Small, remote, inaccessible and agriculture dependent’) may be particularly affected. It is impossible at this stage to be absolutely sure of this rather gloomy prognostication, and it should be also noted that discretionary action may be used to offset the scale of the public sector cuts in the islands. A good example of the latter is the very recent decision of the UK government to introduce a pilot selective fuel duty cut for Scottish island groups. However, in the absence of more discretionary support of this type the British offshore islands do look very vulnerable to public sector retrenchment in the on-going third phase of the crisis.

In Greece, a series of austerity measures involving massive cuts to public spending, services and welfare payments seem to have amplified existing social and spatial disparities (Hadjimichalis, 2011; Matsaganis and Leventi, 2011; Monastiriotis, 2011). In particular, Matsaganis and Leventi (2011) estimate that the austerity measures have already resulted in the income share of the richest 20% of the population (relative to that of the poorest 20%) rising from 6.11 in 2009 to 6.19 in 2010. Hadjimichalis (2011: 257) argues that ‘according to some accounts those who have been first and worst affected are non-unionized workers (particularly women), lower-level civil servants and the lower middle classes in urban areas’ and that ‘urban areas are followed by former industrialized regions (such as Eastern and Central Macedonia), whereas rural and tourist areas will be affected last’. It is also interesting to note that there have been recent reports of ‘Greeks going back to the land’ and examples of what could perhaps be described as ‘return migration’, with educated skilled workers moving back to islands such as Chios (which according to our analysis is in Greek cluster 4 ‘Large, rather remote but highly accessible and diversified’), ‘looking to the nation’s rich rural past as a guide to the future’ (Donadio, 2012). Also, another recent report from the island of Karpathos (which according to our analysis is in Greek cluster 2 ‘Small, remote, inaccessible and agriculture dependent’) suggests that ‘the only positive outcome of the crisis is the return of young people, including graduates, to the village. There are plenty of empty houses here, no shortage of land, and good rains last winter have expanded the opportunities for new crops, as well as giving greater returns from old’ and that this reinforces community and family ties, ‘reversing a century-long trend of depopulation’ (Jilkinson, 2011). This seems to be consistent with Hadjimichalis’s argument (2011: 257) that ‘an important parameter of the Greek crisis … is the informal sector, which may provide, as in the past, buffer solutions via extended family and friends, production for self-consumption etc, in which house and land ownership in rural areas will play a major role’. It could be argued that the islands with a more developed agricultural sector (such as the islands in Greek clusters 1 and 2) would be more likely destinations for such ‘return migration’.

Nevertheless, it has also been suggested that there may be strong migration trends to the opposite direction. In particular, Monastiriotis (2011) argues that the demand deficiency and rising unemployment resulting from the austerity measures could lead to ‘brain-drain’ out-migration, whereby more skilled workers will have an incentive to move from less prosperous regions (including most of the islands) to the larger urban agglomerations, attracted by the larger pool of jobs there as well as the perceivably more and better-quality amenities. He also argues that owing to compositional differences the overall medium- and longer-term negative shock of the austerity measures will be smaller in the more central and high-income regions of Attica, Thessaloniki, Crete and the south Aegean region, whereas it will be significantly larger in the north and northwestern regions including the Ionian and Northern Aegean islands. For instance, he argues that the North Aegean islands are amongst the regions with unusually high shares of incomes generated in public utilities and therefore that they will be disproportionally affected by public sector cuts. Similarly, combined public sector pay and pensions comprise more than 50% of household incomes in the north and northwest regions (again including the North Aegean islands) whereas ‘they are less than 35% in South Aegean and Crete (close to 40% in Athens and Central Greece)’ (Monastiriotis, 2011: 328). Our Greek islands cluster 4 (‘Large, rather remote but highly accessible and diversified’) includes some of the islands that would be particularly affected by public sector cuts. Also, it is interesting to note that according to some accounts, there have been massive cuts in the asking prices of second home properties (and especially luxury properties) in some of the islands (Right Move Overseas, 2011; Tsakiri, 2011). It can be argued that this would particularly affect islands in Greek cluster 3 ‘Closest to Athens, accessible to Athens (sea and air) and the mainland (sea) and diversified’, which includes some of the most successful mass tourism islands as well as popular locations for second-home owners from the mainland. In addition, the austerity cuts have significant impacts on the quality of life of island regions, and one very important area where this may be particularly noticeable is that of health services provision, especially for islands with poor air and ferry boat transport links (Karamanoli, 2011). This would particularly affect the islands in Greek cluster 2 ‘Small, remote, inaccessible and agriculture dependent’. However, it has also been argued that the austerity measures may lead to a so called ‘internal devaluation’ with wages and prices going down (Kokkoris et al., 2010; Krugman, 2011; Monastiriotis, 2011) and that this process may potentially benefit some of the islands specialising in tourism (and in the case of islands such as Crete manufacturing exports as well). This would be particularly relevant to the islands in Greek cluster 3. Finally, there is strong evidence to suggest (see Matsaganis and Flevotomou, 2010; Monastiriotis, 2011) that if the efforts of the Greek government focuses on tackling suspected tax evasion in particular occupational categories then this will result in the considerable amelioration of some of the existing social and spatial disparities.

The analysis and discussion presented in this paper contributes to the on-going efforts to better understand the key issues and challenges affecting island economies in Europe. It can be argued that our results can be used as a basis to identify the types and numbers of islands that may be particularly affected by the financial crisis and austerity measures in Europe, which are particularly severe in Greece. It is also hoped that the results presented in this paper can contribute further towards an evidence base that could facilitate wider analyses of the regional and local impacts of the crisis (such as the work of Blažek and Netrdová, 2012; Hadjimichalis, 2011 and Monastiriotis, 2011) as well as the debates on what policies can be implemented to offset these impacts.