1 Introduction

Migration patterns have been a major force for the past three decades. The habitual immigrant receiving countries such as the United States, Canada and Australia has not only seen an increase in numbers but also a change in the composition of said immigrants. There has been a significant change in that immigrants are now predominantly originating from countries in Africa, Asia and Latin America rather than from the historically prevailing Europe, see Massey et al. (1993).

In explaining the reasons for international migration and changes in the overall pattern, a number of theoretical models has been used but even if the ultimate goal of these models remain the same, they utilize fundamentally different concepts, assumptions and frames of reference.

This paper will focus on modeling international migration for sixteenFootnote 1 OECD countries between 1991 and 2000. Many of the explanatory variables that appear frequently in the trade literature are also used in this paper—stock of immigrants, population size, income differences, common language, colonial ties, etc. The contribution of this paper is twofold. First, we will focus on generating two unique quality of life (QOL) indices that are objective but non-economic in nature. Following Rossouw and Naudé (2008), we construct a comprehensive demographic and geographical QOL index, for all source and destination countries used in our data sample. Consequently, the second contribution of this paper focuses on the empirical link between these constructed QOL indices and migration patterns for the OECD group of countries.

This paper proceeds as follows: Section 2 provides a literature review in terms of both the QOL concept and its link to migration, Sect. 3 details the development of two indices of quality of life to be applied, Sect. 4 develops an immigration model based on gravitational factors, Sect. 5 reports and discusses the results of the panel data, and Sect. 6 concludes and offers future research direction.

2 Literature Review

Migration is caused by a push from behind and/or a pull from an appealing prospect in front. The combination of push and pull factors and research into which specific determinants play a significant role in migration patterns has received a lot of attention in the empirical literature. For example, recent research on this front includes work by Naudé (2008), Zaiceva and Zimmermann (2008), Warin and Svaton (2008), Mayda (2007), Tolbert et al. (2006) and Hatton and Williamson (2005). In general, many of the determinants of migration flows can be categorized under four headings: Political, Economic, Demographic and Environmental factors. Economic factors have undoubtedly received the most attention, probably partly because of data availability, which increases the ease with which it can be statistically investigated. There is now a sizeable set of literature that has uncovered a very strong correlation between the rate of emigration and better economic conditions in the host country, compared to the source country, see Chaps. 1–3 of Bodvarsson and Van den Berg (2009) for a review.

In terms of the literature related to QOL, there appear to be two very distinct approaches towards the computation of this concept. They are respectively subjective and objective in nature. Objective QOL measures are influenced by economic variables which can be seen as physical facts and are more readily accepted by policymakers and more easily interpreted, see Sumner (2003). Two of the best known composite objective QOL measures are the Physical Quality of Life Index (Morris 1979) and the Human Development Index (UNDP 1990). The primary problems with these two measures are that they do not cover enough QOL domains (for example, the Human Development Index only consists of three variables) and GDP per capita often plays too important a role in these indices. This had led to the construction of more recent, broader defined QOL indices such as the Index of Economic Well-being (Osberg and Sharpe 2000), the Economist Intelligence Unit’s (2005) QOL index, and the State of the USA (2011).

Subjective QOL refers to the well-being as declared by a particular individual. It is based on the declaration by a person and can be seen as a measure that incorporates all life events, aspirations, achievements, failures and emotions (Rojas 2003). In considering subjective QOL studies, one must note that this area of QOL has been greatly influenced by Sen’s (1984, 1993) capabilities approach. Sen’s idea that a person’s capabilities influences his/her functioning’s has led to a wealth of studies such as those done by Griffin (1986, 1991 pp. 45–69), Cummins (1996), Alkire (2002) and Narayan et al. (2000), all of which add to shaping the current trends in measuring subjective QOL. The basic reasoning for using subjective QOL measures stem from the idea that an individual should be consulted when determining his/her perceived QOL.Footnote 2 Veenhoven (2002, pp. 40–41) argued that it is necessary to include subjective measures when determining social policy as this will enable policy makers to distinguish between ‘wants’ and ‘needs’ and help to assess the success of policy through goal attainment and public support rendered. Despite these advantages of subjective measurements of quality of life there still remains some uncertainty with regard to its conceptual and interpretative capacity, as life domains are evaluated by the individual’s position in the social system as well as by their personal characteristics or cultural specificity (Bălţătescu 2007).

In contrast, objective measures of quality of life are seen as being more receptive, changing much faster than subjective social data that sometimes suffers a time lag. These measures are also perceived as being cheaper and less complex to collect than subjective data. Sumner (2003) argues that the supremacy of objective measures of quality of life is, in addition to the reasons already mentioned above, due to the presumption that objective measures are more adaptable to quantification as they are tangible. In contrast, subjective measures of quality of life are somewhat less adaptable to quantification and rely on more unsubstantiated proxies.

Empirical work has been done trying to link QOL (be it objective or subjective, economic or non-economic in nature) to migration patterns, with the earliest paper discussing this possibility dating back to Liu (1975). In his study, he found that net migration rates between 1960 and 1970 in all 50 States as well as the District of Columbia in the United States (US) responded positively and significantly to overall QOL indices. However, many of the QOL measures used in his study were economic in nature, e.g. cost-adjusted personal income per capita.

More recently, Osborne (2003) attempted to link global migration flows to several non-economic factors considered in the happiness literature—such as infant mortality rate, the nation’s carbon dioxide emissions, crime rate, and the level of political freedom. He applied this to migration to the US and within the US. His research found that the most consistent motivator of the migration decision was economic reasons. There was little evidence of the importance of environmental conditions, and insignificant impacts of crime and political freedom.

Rebhun and Raveh (2006) also focused on migration flows to the US and within the US when examining its relationship with QOL. They focused on two time periods 1965–1970 and 1985–1990. In the first time period, it was striking that many of the QOL variables were found to be insignificant. Three variables were found to be significant in explaining interstate migration and even more interesting is that two of these variables were economic variables—income per capita and employment. In the latter time period, employment opportunities was again found to be significant and across both time periods, the only non-economic QOL variable that was significant in both models, was the crime rate.

Another relevant study that focuses on the link between subjective well-being and migration is Bartram (2010a). His research found that although a higher income could be associated with a higher level of happiness after immigration it was other non-economic factors such as marital status, and health that had a stronger association with happiness. Other significant studies, which fall outside the scope of this paper, includes Bălţătescu (2007) and Bartram (2010b) both of which investigates the level of life satisfaction of immigrants after the fact and does not explicitly look at the relationship of subjective indicators being the driving force for said immigration. Other studies go the next step and just assume that migration will improve QOL and happiness, e.g. Blanchflower (2008) used life satisfaction data to forecast migration flows.

A review of the above literature indicates a clear gap in terms of research empirically investigating objective non-economic QOL drivers of migration flows. Much of the previous empirical work either included purely economic factors as the QOL variables (such as per capita income), or included QOL indicators that were subjective in nature, or were focused on the US, or specific cultural groups. Accordingly the contribution of this paper is twofold. Firstly, our paper will focus on generating two unique QOL indices that are objective but non-economic in nature. This is unique as most QOL studies focus either on subjective non-economic or objective economic measures. In following Rossouw and Naudé (2008), this paper makes use of 22 variables to construct a demographic and geographical QOL index for each of the source and destination countries used in the data sample. The second contribution of this paper is to focus on migration flows for the OECD. To our knowledge, no research has investigated in detail the possibility of a statistical relationship between non-economic QOL and migration flows for this grouping of countries.

3 Constructing the Quality of Life Indicators

Quality of life is a concept that has experienced wide-spread theoretical and empirical research. It is well recognized that GDP per capita does not solely reflect quality of life and that growth in per capita income does not always equate to increases in human well-being and development, see Qizilbash (1996).

As mentioned in Sect. 2, much of the past literature on the non-economic QOL has been done with the use of subjective indicators in order to measure how people perceive their non-economic QOL, or level of life satisfaction. Instead, we focus on using objective indicators in constructing our non-economic QOL indices for the 82 countries used in our analysis.Footnote 3 To date, the most progress in determining non-economic QOL has been made by McGillivray (2005). He extracted, by means of principal component analysis, the maximum possible information from various standard national non-economic quality of life achievement measures. McGillivray (2005) then empirically identified the variation in this extraction not accounted for by variation in income per capita, which he named μ i . This variable was then defined as being the residual yielded by cross-country regression of the extraction on the natural log of Purchasing Power Parity (PPP) GDP per capita. The variable μ i can therefore be interpreted inter alia as a measure of non-economic quality of life achievement, in the sense that it captures quality of life achieved independently of income.

The same methodology is applied in this paper, in order to determine the non-economic QOL residuals for all 82 countries included in our sample. It is important to note that the trends in the calculated residuals are determined by the choice of variables included as well as by trends in those variables. With the aim of constructing a non-economic index, the variables selected for this analysis are described in Table 1.

Table 1 Variables used in the quality of life indices
Full size table

After the selection of variables,Footnote 4 the data were divided into two distinct groups as argued by Johansson (2002) and Erickson (1993, pp. 67–83). Following Rossouw and Naudé (2008), the first group consists of variables pertaining to man and everything man made (called hereafter demography) and the second group is purely geographical and environmental of nature (called hereafter geography).

The next step after the categorization of the variables under the headings of demography and geography was to apply principal components analysis.Footnote 5 It was found that the first three components of demography (must have an eigenvalue above 1) explained 71.97% of the variation in that QOL index and the first four components of geography explained 67.15% of its variance.

To compile the two separate indices from these seven components, different weights had to be appointed to each one. Unfortunately, there is no proxy to use for the selection of weights and it is not statistically acceptable to apply equal weights to each of the components. Thus, the first component of each group (seeing as the first principal component accounts for the most variance and the components are ordered in size as they are extracted) was used in compiling separate demography and geography QOL indices.

In the same vein as McGillivray’s (2005) methodology, a regression analysis was next completed in order to determine the residual values of the demography and geography QOL indices. Similarly, the Human Development Index (HDI) was also used as an alternative QOL index, since this index is widely used and acknowledged, and will provide a good test of robustness of results. The three equations for country i at year t (t = 1991–2000) are specified as follows:

$$ { \ln }\,{\text{GEOQoL}}_{it} = \alpha_{3it} + \beta_{3it} \,{ \ln }\,{\text{percapita}}_{it} + \mu_{3it} $$
(1)
$$ { \ln }\,{\text{DEMQoL}}_{it} = \alpha_{2it} + \beta_{2it} \,{ \ln }\,{\text{percapita}}_{it} + \mu_{2it} $$
(2)
$$ { \ln }\,{\text{HDIQoL}}_{it} = \alpha_{1it} + \beta_{1it} \,{ \ln }\,{\text{percapita}}_{it} + \mu_{1it} $$
(3)

The above three equations aim to extract the part of the QOL index that cannot be explained by per capita income, and therefore reflects a more accurate and objective measure of non-economic QOL. It is these extracted residuals (μ it ) that provide values for the non-economic QOL for the 82 countries in our data sample and which are used in the gravity model in the following section, to determine if they influence immigration patterns in the OECD.

4 Model and Data

As mentioned in Sect. 2, economic incentives to migrate are a function of both undesirable conditions in the source country and attractive conditions in the destination country. Incentives to migrate are generally called pull-factors, and include higher wages, economic freedom, property right protection, employment opportunities, and social mobility. However, there are formal and informal costs of moving to consider, such as transportation costs, entry visas, time of travel, and psychological costs. Immigrants also face significant stay-away factors such as language barriers, discrimination, and uncertainty. For a detailed discussion of immigration determinants see Chaps. 1 and 2 of Bodvarsson and Van den Berg (2009).

Empirical work on international migration blend elements of the gravity model with elements of Borjas’ (1987, 1991) human capital investment model of immigration which suggests that migration flows are positively related to income differences and negatively correlated to migration costs, see for example Greenwood (1997), Hatton and Williamson (2005) and Lewer and Van den Berg (2008). This paper follows the previously cited work and suggests that a model based on geographic and gravitational factors, which has been traditionally applied to trading patterns, can sufficiently be used to explain immigration patterns. Following the foundational work of Greenwood (1997), Mayda (2007) and Lewer and Van den Berg (2008), the underlying immigration-gravity relationship is expressed:

$$ {\text{IMMIGRATION}}_{ij} = f\left[ {\left( {{\text{RELY}}_{ij} ,{\text{ POP}}_{i} \cdot {\text{POP}}_{j} } \right)/{\text{DIST}}_{ij} } \right], $$
(4)

where immigration to country i from country j is a positive function of RELY ij , the per capita income ratio of country i and j, and a negative function of distance between capital cities. Population size is the “mass” variable; ceteris paribus, the more people there are in a source country, the more people are likely to migrate, and the larger the population in the destination country, the larger is the labor market there. It is common for statistical regression models like Eq. 4 to be specified in natural logarithms. Representing natural logs in lower case letters, the gravity equation is:

$$ {\text{immigration}}_{ij} = a_{0} + \, a_{ 1} ({\text{pop}}_{i} \cdot {\text{pop}}_{j} ) + a_{ 2} ({\text{rely}}_{ij} ) + {\text{a}}_{ 3} ({\text{dist}}_{ij} ) + u_{ij} , $$
(5)

in which immigration ij represents immigration to destination country i from source country j, and rely ij is the ratio of destination to source country real per capita income. A priori, it is expected that the coefficients a 1 and a 2 will be positive and that a 3 will be negative.

In many cases, researchers may want to control for other factors of immigration. The literature suggests that immigration is path dependent in that current immigration flows are related to past immigration patterns. For example, Kahan (1978), Murayama (1991), Rephann and Vencatasawmy (2000) find distinctive ethnic concentrations of immigrants, and Zavodny (1997) find that family connection is the most significant immigrant determination factor. Additional evidence suggests that immigration flows are larger, ceteris paribus, when a common language is spoken, see Greenwood and McDowell (1991). Adding these and other considerations to the model above creates the augmented immigration gravity equation:

$$ \begin{aligned} {\text{immigration}}_{ij} & = a_{0} + a_{ 1} ( {\text{pop}}_{\text{i}} \cdot {\text{pop}}_{\text{j}} )+ a_{ 2} ({\text{rely}}_{ij} ) + a_{ 3} ({\text{dist}}_{ij} ) + a_{ 4} ({\text{stock}}_{ij} ) \, \\ & \quad + a_{ 5} {\text{CONT}}_{ij} + a_{ 6} {\text{LANG}}_{ij} + a_{ 7} {\text{LINK}}_{ij} + u_{ij} , \\ \end{aligned} $$
(6)

in which stock ij is the stock of immigrants from an immigrant’s source country already living in the destination country, CONT ij , LANG ij , and LINK ij are dummy variables which take the value 1 for pairs of countries which have a contiguous border, common language, and common colonial linkage, respectively. The anticipated sign on all three dummy variables is positive, reflecting the idea that proximity, common language, and common historical ties create immigration networks.

To test the role that quality of life measures play in immigration patterns, this paper applies six unique indices of QOL to Eq. 6. First, this paper utilizes the two non-economic indices generated in Sect. 3 above: (1) GEOQoL and (2) DEMQoL as well as the income adjusted United Nation’s HDI (3) HDIQoL. The other three QOL proxies include: (4) the Frazier Institute’s economic freedom index (Free), (5) the World Database of Happiness index (Happy), and (6) the Socioeconomic Data and Applications Center’s environmental sustainability index (ESI).

Table 2 provides the definitions, sources and descriptive statistics of all data used in this paper. It is worth noting that the total sample of immigrant source countries captures nearly seventy percent of total immigration to the 16 OECD countries over the time-series.Footnote 6

Table 2 Data definitions and descriptive statistics
Full size table

The six QOL indices mentioned above were collected for both the immigrant source country and destination country. They are separately added to Eq. 6 in the form of relative ratios and relative differences, yielding the equations of interest:

$$ \begin{aligned} {\text{immigration}}_{ij} & = a_{0} + a_{ 1} ({\text{pop}}_{i} \cdot {\text{pop}}_{j} ) + a_{ 2} ({\text{rely}}_{ij} ) + a_{ 3} ({\text{dist}}_{ij} ) + a_{ 4} ({\text{stock}}_{ij} ) \\ & \quad + a_{ 5} {\text{CONT}}_{ij} + a_{ 6} {\text{LANG}}_{ij} + a_{ 7} {\text{LINK}}_{ij} + a_{ 8} ({\text{QOL}}_{i} /{\text{QOL}}_{j} ) + u_{ij} , \\ \end{aligned} $$
(7)
$$ \begin{aligned} {\text{immigration}}_{\text{ij}} & = a_{0} + a_{ 1} ({\text{pop}}_{i} \cdot {\text{pop}}_{j} ) + a_{ 2} ({\text{rely}}_{ij} ) + a_{ 3} ({\text{dist}}_{ij} ) + a_{ 4} ({\text{stock}}_{ij} ) \\ & \quad + a_{ 5} {\text{CONT}}_{ij} + a_{ 6} {\text{LANG}}_{ij} + a_{ 7} {\text{LINK}}_{ij} + a_{ 8} \left( {{\text{QOL}}_{i} - {\text{QOL}}_{j} } \right) + u_{ij} , \\ \end{aligned} $$
(8)

The augmented gravity models of immigration presented in Eqs. 7 and 8 contain both economic, non-economic and geographic variables. The contribution of this article is to test the marginal immigration effects of (a 8), the six alternative macro-non-economic quality of life factors. It is important to note that the non-economic quality of life variables are capturing unique information about immigration as the correlations between them and other non-economic variables in the model (e.g. common language) are quite small. Table 3 reports the correlation coefficients for all variables applied in this study.

Table 3 Correlation matrix
Full size table

5 Empirical Results and Discussion

Most studies estimate Eqs. 7 and 8 by using double logarithmic form. However, one problem with this technique is that country pairs whose immigration flows are zero are omitted. This paper includes all data by applying the methods recommended by Feenstra (2004) who prescribes using the scaled ordinary least squares (SOLS) method with fixed effects when working with censored data. An additional benefit from this method is that it corrects for standard error clustering, see Redding and Venables (2000) and Rose and van Wincoop (2001).

Tables 4 and 5 report the results from estimating the gravity model of immigration specified in Eqs. 7 and 8 above. Most covariates of the augmented gravity model have the correct sign and are with exception of contiguous border and the various non-economic QOL indices, highly significant. The adjusted R 2 measure ranges from 0.68 to 0.71 and indicates that the model performs well.

Table 4 Tests of relative quality of life on immigration patterns
Full size table
Table 5 Tests of quality of life differences on immigration patterns
Full size table

Interpretation of coefficients in Tables 4 and 5 are straightforward as they effectively measure by what percentage migration changes for a percentage change in the independent variable. For example, the coefficient dist ij estimates by what percentage migration changes for a given percentage increase in the distance between the source and destination country. In the second column of Table 4, the dist ij coefficient is −0.216 suggesting that for every 10% increase in distance; immigration between countries is reduced by 2.16%. Distance as a cost of migration tends to dampen immigration flows has been reported in many studies including Greenwood and McDowell (1991), Pedersen et al. (2004) Clark et al. (2007), Mayda (2007) and Lewer and Van den Berg (2008). Continuing with column two of Table 4, it is important to note that common language enhances immigration among source and destination countries. The statistically significant coefficient on LANG is 0.235, suggesting that, ceteris paribus, countries with common language observe roughly 26% (e0.235 − 1 = 0.264) more migration than when there is no shared language between two countries. Also consistent with previous empirical studies, our results indicate that the pull factors of real income differences and labor market size/opportunity have a positive impact on immigrant flows from source to destination countries.

It is worth noting the large and highly significant immigrant stock coefficient, stock ij , which confirms that immigration is indeed path dependent, see Zavodny (1997), Hatton and Williamson (2005) and Pedersen et al. (2004). The role of past migration on current migration flows has been firmly developed in the literature and is often referred to as a migrant network, dating back to Yap (1977) who suggests that destination country contacts have a significant impact on where immigrants settle. Taylor (1986) and Massey and Espana (1987) provide additional network rationale by suggesting that migrant networks lower information and psychological costs, lower job and housing search costs, and provide cultural familiarity to recent immigrants.

The primary contribution of this paper is to test the relationship between non-income based factors and immigration flows. While the non-economic quality of life instruments applied in this paper are nowhere near exhaustive, the findings of this paper indicate that none of the six non-economic quality of life alternatives significantly impacted the supply of immigrants to the 16 OECD countries.Footnote 7 Rather the findings suggest that while no single determinant seems to dominate the process, economic, historic (path-dependency and network) and geographic factors (such as distance) significantly determine the flow of immigration for the sample. These findings are in support of Osborne (2003), Pedersen et al. (2004), Rebhun and Raveh (2006), and Bodvarsson and Van den Berg (2009, Chap. 3) who find earnings differences and migration costs (among other things) are primary immigration drivers. Also relevant here is the summary from Michalos (1997) who states that “There is evidence from survey research that people tend to move in order to improve the quality of their lives…” (p. 155). His research goes on to say that economic and cultural/family networks play significant roles in an individual’s decision to migrate, but he also reports that mild climate, low pollution, low crime rates and other non-economic factors play a role (especially in regional migration), it seems clear that additional studies examining non-economic variables and immigration rates are needed before any formal conclusions can be made.

6 Conclusions

This paper contributes to the immigration literature in two ways; first, constructing objective non-economic QOL measures for OECD countries for the demographic and geographical indices respectively, and by testing the non-economic quality of life-immigration relationship.

Using an extended gravity model of international migration for sixteen OECD countries from years 1991 to 2000, the fixed effects panel indicates that non-economic QOL measures play a statistically insignificant role in determining immigration flows. The stock of immigrants from the source country already living in the destination country, population size, destination country income, common language, historical colonial ties all significantly increase the flow of immigration to the OECD sample countries. Geographical distance is found to erode the flow of immigration.

Clearly, much work still remains to be done in accessing the determinants of immigration with respect to non-income based measures. Research is needed with more complete data sets, different econometric tools and more inclusive models. There is also significant potential for testing the role of objective and non-economic QOL indices on other economic and global outcomes. Future studies may want to examine the linkages between QOL and economic growth rates and QOL and labor productivity among other areas.