Inclusion/Exclusion Criteria

brain drain, migration intention/decision/behaviour, skilled migration, and mobility propensity. For databases providing only abstract and bibliographical citations, every feasible attempt has been made to obtain the full text by searching in another database, or contacting the relevant authors. As suggested by [9], after the databases have been searched through, a last search via Google Scholar and Google is used to round up unpublished or ‘fugitive’ articles [14]. For every study selected to be included in the meta-analysis, its references section is also checked for relevant studies. At the end of the search, a total of 22 studies that fulfil the inclusion/exclusion criteria are selected. To construct the meta-analysis database, I first extract the coefficients of the education variable and their corresponding standard errors from the selected studies. For studies with multiple model specifications, the coefficient and standard error from the base model is extracted. These coefficients are the impact of education on migration behaviour. Along with these two statistics, I also compile and code other characteristics from each of the selected studies. The ‘Meta-regression’ section explains further.

FINDINGS & DISCUSSIONS

Meta-analysis

Pane A in Table 2 shows the effect size of each study, i.e. the coefficient estimate of the education variable. As these coefficient estimates are obtained from discrete choice models, their magnitudes are not directly interpretable. The meta-analysis obtains a summary effect of +0.300, i.e. the average coefficient estimate of the impact of education on migration behaviour. Pane A can also be read along with the forest plot shown in Figure 1. The squares in a forest plot represent the effect sizes while the horizontal lines represent the confidence intervals of the effects. A larger sample size gives a smaller confidence interval, indicating a more precise effect size.

The area of the squares reflects the weight that a particular study [15]. Studies with better precision are given more weight, which is a function of sample size. The location of the squares indicates the direction and the magnitude of the effect [16]. It is obvious from Figure 1 that most of the effect sizes border on the positive pane, indicating that the more educated tend towards migrating.

Pane B shows how the impact of education shifts over time, since the studies are sorted chronologically from 1993 to 2011. Pane B can be read along with Figure 2. The effect size displayed in each row is the summary effect based on all the studies up to and including that row. There appears to be three clusters of summary effects shown in Figure 2, with the impact of education stabilizing starting mid-2000s. This indicates that in recent years, the impact of education has been consistently at a magnitude of approximately 0.300 (which is translated to its corresponding marginal change in the probability to migrate, depending on the discrete choice model used).

Figure 1.            Forest plot

Figure 2.            Cumulative forest plot
Pane C and Figure 3 reveal two influential studies in the meta-analysis – [17] and [12]. If we omit Fan’s study from the meta-analysis, the average coefficient estimate of the education variable on migration behaviour will drop from 0.300 to 0.147. Similarly, if we omit Chiquar & Hanson’s study, the average coefficient estimate will increase from 0.300 to 0.394. This finding is not surprising because from Pane A of Table 2, Fan obtained a coefficient estimate of 3.14 for its education variable, while Chiquar & Hanson obtained a negative estimate of 3.537. These two estimates stand out when compared to the others. In this sense, these two influential studies can be regarded as outliers.

          Figure 3.  Influential studies

Publication Bias

I also check for publication bias to ensure no significant bias in the 22 studies selected for this meta-analysis (in which 18 are journal articles, 3 working/discussion papers, and 1 book chapter). Publication bias can arise when (i) statistically insignificant results are unpublished or being put away in the file drawer, (ii) publication in reports or working papers are excluded in the meta-analysis, and (iii) only publication in a certain language are taken into consideration.

Figure 4 shows a funnel plot typically used to check for presence of publication bias. A funnel plot is a scatterplot of the coefficient estimates from the studies and their corresponding standard errors. A funnel plot follows the rationale that when the sample size of a study increases, so does the precision of its coefficient estimate (i.e. as measured by the standard errors). In the absence of publication bias, the coefficient estimates will scatter symmetrically, with those from studies with smaller sample size making up the base of the plot, and coefficient estimates from larger studies will funnel up the plot. Publication bias however, is only one of the possible causes of funnel-plot asymmetries [18]. 

Figure 4.  Funnel plot

Significant at the *10%, **5%, and ***1% level.
† Influential study.
Pane A, B, C.
Using the random-effects approach, the summary effect is computed to be 0.300 with a standard error of 0.143 and a p-value = 0.036.

From Figure 4, there seems to be a bias towards publication with positive coefficient estimates. Formal tests of publication bias however, suggest otherwise. At the 5% significance level, both Egger’s test (p-value = 0.586) and Begg’s test (p-value = 0.055) suggest no evidence of publication bias. Rosenthal’s (1979) classic fail-safe N is 2,642 (a p-value much less than 1%), suggesting that the possibilities are remote for us to have missed out on more than 2,600 studies in the related migration literature. We however, cannot preclude the possibility of language bias since only English-language publications are searched. Nevertheless, it is believed that non-English scholarly publications would not have been significant enough to cause a serious publication bias.

Meta-Regression

Meta-regression investigates the extent to which statistical heterogeneity between results of multiple studies can be related to one or more characteristics of the studies [19]. It is the best way to account for heterogeneity or between-study variance [20]. Through meta-regressions, observed heterogeneity can be accounted for, in which study characteristics explain some of the variations in the coefficient estimates between the studies. That is, meta-regression can help to answer questions like why some studies obtained positive/negative coefficient estimates, why some of their magnitudes are larger than those of other studies, and why some of the estimates in certain studies are significant while others insignificant. In doing the meta-regression here, we take into account study characteristics as listed in Table 3.

Results from the meta-regression indicate presence of heterogeneity in the coefficient estimates, with a between-study variance (or the variance of the true effect sizes) of t2=0.836, significant at the 1% level. The proportion of observed variance reflecting real differences in effect size is I2=99.7%, i.e. 99.7% of the observed variance is due to real differences in the studies rather than to random error. Such heterogeneity is also obvious from Figure 1. Had there been no heterogeneity issues, all the squares would have aligned to a straight vertical line.

Results from Table 3 suggest that the source of heterogeneity in the coefficient estimates of the education variable come from six of the study characteristics, i.e. the significant meta-regression coefficients. In running a meta-regression, we are basically estimating the following.

where,

bj            = the reported coefficient estimate of the education variable of the jth study from a total of L studies
             = the ‘true’ value of the parameter of interest
         = the study characteristics
          = the meta-regression coefficient that reflects the biasing effect of particular study characteristics
            = the meta-regression error term
j               = 1, 2, …, L

The publication year of a study is positively associated with the coefficient magnitude of the original studies, i.e. +0.309; more recent publications found larger positive impact of education on migration intention/decision. This result is also supported by the absence of publication bias discussed in the previous section. Publication bias tends to occur when publications are

focused on obtaining positive and statistically significant results. The fact that we found no evidence of publication bias and that the meta-coefficient shows a significant +0.309, may be pointing to the fact of an increasing impact of education after all. The positive meta-coefficient of the ‘skilled’ variable (i.e. 3.986) indicates that when a study examines skilled migration or the brain drain phenomenon, the effect size of the education variable increases, resulting in corresponding increases in the probability to migrate. When skilled migration is the issue, the education variable is understandably important and has a significant positive impact on the migration probability of the highly educated. The meta-coefficient of the education variable is +1.658, suggesting that when education level dummies are used in the original studies instead of using years of education, the impact of education on migration intention/decision tends to be positive. One possibility is perhaps the real effect of education is more readily captured by the level of education (i.e. a real difference between the level of a high school diploma and that of a doctoral degree, for instance), than by the years of having been in formal schooling.

At –0.0001, the negative impact of sample size on the coefficient magnitudes is relatively negligible although it has a stronger statistical significance. The negative impact of sample size here suggests that as more respondents are surveyed on their migration intention/decision, the impact of education (i.e. the coefficient magnitude of the education variable) decreases. The negative relationship casts doubts on the presence of genuine empirical effect of the impact of education on migration intention/decision [21]. This result however, could also be due to noise from a heterogeneously large cross-sectional sample. The practical significance of sample size nevertheless remains somewhat trivial. When the migration issue is examined in European region, the impact of the education variable decreases and translates into decreasing probabilities to migrate. This suggests plausibility of easier mobility within the European region, and therefore less importance might be placed on education as a mobility passport. Similarly, when students instead of working professionals are examined, the impact of education on migration behaviour also decreases. The students examined in the original studies are typically foreign students studying in a host country. Since they are already in the host countries, the level of education they are pursuing there is not as important as say other reasons such as assimilation process, insiders’ information, and networking, for example. On the contrary, the education level or years of education that working professionals possess might be more crucial in influencing their migration behaviour.

CONCLUSION

This paper may well represent the first meta-analysis application on the impact of education on migration. This is a clear departure from the typical analyses used in the migration literature. Here, we have delineated the standard set-up of the analysis, by first doing a meta-analysis to obtain the summary effect size, followed by a publication bias check, and a meta-regression to identify the source of heterogeneity in effect size. A total of 22 micro-level studies that look at the determinants of migration have been analysed. These papers have included the education variable as one of the regressors, where it is operationalized either as the years of education or the highest level of education obtained. The outcome or dependent variable of these selected studies is either on actual migration behaviour, or on the intention to migrate. The meta-analysis conducted here comprises three parts: the meta-analysis, publication bias check, and meta-regression. Results from the meta-analysis conclude the summary impact of the education coefficient to be approximately +0.300. Results from formal tests on publication bias show no evidence of such bias. A meta-regression is used to account for the source of heterogeneity in the coefficient estimates. Six of the study characteristics are identified to be contributing to the heterogeneity.

An extension of this paper will incorporate coefficient estimates of the education variable from all different model specifications used within a study, instead of only including the coefficient from the base model. Another possible extension is to do a meta-analysis on a vector of coefficients of a number of variables, and not just limited to meta-analysing the impact of the education variable on migration decision.

References

Glass, G. V. (1976). Primary, secondary, and meta-analysis of research. Educational Researcher, 5(10): 3-8.
Glass, G. V. (1977). Integrating findings: The meta-analysis of research. Review of Research in Education, 5: 351-379.
Stanley, T. D. & Jarrell, S. B. (1989). Meta-regression analysis: A quantitative method of literature surveys. Journal of Economic Surveys, 3: 161-170.
Stanley, T. D. & Jarrell, S. B. (1990). A meta-analysis of the union-nonunion wage gap. Industrial and Labor Relations Review, 44: 54-67.
Longhi, S., Nijkamp, P. & Poot, J. (2010). Joint Impacts of Immigration on Wages and employment: Review and Meta-analysis. Journal of Geographical Systems, 12: 355-387.
Longhi, S., Nijkamp, P., & Poot, J. (2005). A meta-analytical assessment of the effect of immigration on wages. Journal of Economic Surveys, 19: 451-477.
Longhi, S., Nijkamp, P., & Poot, J. (2006). The impact of immigration on the employment of natives in regional labour markets: A meta-analysis. IZA Discussion Paper Series, No. 2044, Bonn: Institute for the Study of Labor (IZA).
Ozgen, C., Nijkamp, P. & Poot, J. (2010). The effect of migration on income growth and convergence: Meta-analytic evidence. Papers in Regional Science, 89(3): 537-561.
Genҫ, M., Gheasi, M., Nijkamp, P. & Poot, J. (2011). The impact of immigration on international trade: a meta-analysis. NORFACE Migration Discussion Paper Series, No. 2011-20, London: New Opportunities for Research Funding Cooperation Agency in Europe (NORFACE).
Vijverberg, W. P. M. (1993). Labour market performance as a determinant of migration. Economica, 60(238): 143-160.
Drinkwater, S. (2003). Go West? Assessing the willingness to move from Central and Eastern European countries. FLOWENLA Discussion Paper, No. 5, Hamburg Institute of International Economics.
Chiquiar, D. & Hanson, G. H. (2005). International migration, self-selection, and the distribution of wages: Evidence from Mexico and the United States. Journal of Political Economy, 113(2): 239-281.
Brucker, H. & Trubswetter, P. (2007). Do the best go West? An analysis of the self-selection of employed East-West migrants in Germany. Empirica, 34(4): 371-395.
Rosenthal, R. (1995). Writing meta-analytic reviews. Psychological Bulletin, 118(2): 183-192.
Harris, R. J., Bradburn, M. J., Deeks, J. J., Harbord, R. M., Altman, D. G. & Sterne, J. A. C. (2008). Metan: fixed- and random-effects meta-analysis. The Stata Journal, 8(1): 3-28.
Borenstein, M., Hedges, L. V., Higgins, J. P. T. & Rothstein, H. R. (2009). Introduction to meta-analysis. Sussex: John Wiley & Sons, Ltd.
Fan, C. C. (2002). The elite, the natives, and the outsiders: Migration and labor market segmentation in urban China. Annals of the Association of American Geographers, 92(1): 103-124.
Sterne, J. A. C., & Harbord, R. M. (2004). Funnel plots in meta-analysis. The Stata Journal, 4(2): 127-141.
Thompson, S. G., & J. P. T. Higgins. (2002). How should meta-regression analyses be undertaken and interpreted? Statistics in Medicine, 21: 1559-1573.
Poot, J. (2013 forthcoming). Meta-analysis of previous empirical research findings. In Stimson, R. J. (Ed), Handbook for spatially integrated social science research methods. Cheltenham UK: Edward Elgar.
Stanley, T. D. (2001). Wheat from chaff: Meta-analysis as quantitative literature review. Journal of Economic Perspectives, 15(3): 131-150.

•  
• Remaining studies included in the meta-analysis

Boheim, R. & Taylor, M. P. (2002). Tied down or room to move? Investigating the relationships between housing tenure, employment status and residential mobility in Britain. Scottish Journal of Political Economy, 49(4): 369-392.
Burda, M. C., Hardle, W., Muller, M. & Werwatz, A. (1998). Semiparametric analysis of German East-West migration intentions: Facts and theory. Journal of Applied Econometrics, 13(5): 525-541.
Chen, T. J. & Su, H. Y. (1995). On-the-job training as a cause of brain drain. Weltwirtschaftliches Archiv, 131(3): 526-541.
Constant, A. F. & D'Agosto, E. (2008). Where do the brainy Italians go? IZA Discussion Paper Series, No. 3325, Bonn: Institute for the Study of Labor (IZA).
De Jong, G. F., Richter, K. & Isarabhakdi, P. (1996). Gender, values, and intentions to move in rural Thailand. International Migration Review, 30(3): 748-770.
De Jong, G. F. (2000). Expectations, gender, and norms in migration decision-making. Population Studies, 54(3): 307-319.
Epstein, G. S. & Gang, I. N. (2006). The influence of others on migration plans. Review of Development Economics, 10(4): 652-665.
Fidrmuc, J. (2005). Labor mobility during transition: Evidence from the Czech Republic. William Davidson Institute Working Paper Series, No. 767, University of Michigan Business School.
Fidrmuc, J. & Huber, P. (2007). The willingness to migrate in the CEECs: Evidence from the Czech Republic. Empirica, 34(4): 351-369.
Huber, P. & Nowotny, K. (2011). Moving across borders: Who is willing to migrate or to commute? Regional Studies, DOI:10.1080/00343404.2011.624509.
Liebig, T. & Sousa-Poza, A. (2004). Migration, self-selection and income inequality: An international analysis. Kyklos, 57(1): 125-146.
Mora, J. & Taylor, J. E. (2006). Determinants of migration, destination, and sector choice: Disentangling individual, household, and community effects. In Ozden, C. & Schiff, M. (Eds), International migration, remittances, and the brain drain. Washington: World Bank & Palgrave Macmillan.
Papapanagos, H. & Sanfey, P. (2001). Intention to emigrate in transition countries: The case of Albania. Journal of Population Economics, 14(3): 491-504.
Sun, M. & Fan, C. C. (2011). China’s permanent and temporary migrants: Differentials and changes, 1990-2000. The Professional Geographer, 63(1): 92-112.
Van Dalen, H. P., Groenewold, G. & Schoorl, J. J. (2005). Out of Africa: What drives the pressure to emigrate? Journal of Population Economics, 18(4): 741-778.
Van Dalen, H. P. & Henkens, K. (2007). Longing for the good life: Understanding emigration from a high-income country. Population and Development Review, 33(1): 37-65.
Zhao, Y. (1999). Labor migration and earnings differences: The case of rural China. Economic Development and Cultural Change, 47(4): 767-782.

•  
• AUTHORS PROFILE
• Dr. Jan-Jan Soon is currently Senior Lecturer at Universiti Utara Malaysia. Her expertise areas include migration economics, labour economics, and applied microeconometrics..  
•  
•