Mobility Intentions of Foreign Researchers: The Role of Non-economic Motivations

Abstract

Recent contributions suggest that non-economic factors could be important motivational drivers of scientific mobility. We investigate this hypothesis in a sample of foreign researchers in Italy and Portugal, examining their willingness to leave the host country. We distinguish between economic factors, non-economic relational factors and non-economic aspirational factors. Controlling for the relevant contextual variables, we find that foreign researchers, unsatisfied with aspirational factors (e.g. level of independence, autonomy, intellectual challenge and social status), are more likely to leave their host country and move to a third country than they are to return to their countries of origin. Relational and economic factors, such as salary and benefits, do not demonstrate any additional impact.

Keywords:

• International scientific mobility
• brain drain
• brain circulation
• motivation

Acknowledgements

The authors are grateful to the Portuguese Fundação para a Ciência e a Tecnologia (FCT) and to Tiago S. Pereira and colleagues for the access to Portuguese data.

Notes

¹ The term “aggregate factor” is used to indicate the factors resulting from the factor analysis to improve their recognition and thus readability throughout the text.

² For robustness purposes, the analysis was performed both keeping these variables as described (on a scale of 1–4) and as binary variables equal to 0 for values smaller or equal to 2 and 1 for values larger than 2. The results were similar.

³ In alternative to the GDP per capita we considered the Gross domestic expenditure in R&D and the Higher education sector expenditure in R&D by main field of science, both in absolute value and as percentage of GDP (data from the OECD Science and Technology Indicators database: http://stats.oecd.org). The use of these data determined a drop in the number of observations due to missing values for several countries. This, in some cases, determined a lower significance of the results. However, overall, the main results presented were robust to the inclusion of these additional control variables.

⁴ We introduced new categories within Natural Science and Engineering (the areas mostly represented in our sample). In particular we distinguished researchers in universities from researchers in public research centers and researchers involved from researchers not involved in collaborations with industry. We have not done the same for other research areas because no or very few researchers in these other areas were affiliated in Public Research Centers or were involved in collaborations with industry.

⁵ The survey did not allow distinguishing more precisely the intended period of mobility in case researchers indicated temporary mobility. Similarly, we have little information on the actual intentions of researchers indicating to not having taken a decision yet. Therefore we do not detail further the dependent variable in this direction. In order to address the concern that the ambiguity of these answers drive our results we estimated our models also assigning these two categories either to researchers intending to leave or researcher intending to stay, with similar results. Alternatively, excluding researchers indicating one of these categories from the sample also provides similar results. Finally, we run a logit for each one of the thresholds in the ordered probit regression. Results are consistent with the ones shown but are not always significant for each type of variable and for each dependent variable.

⁶ Similar figures to those discussed would be obtained through an analysis of the marginal effects of decisions regarding whether to leave the home country (both temporarily and permanently).

⁷ The inclusion of more than one item per category leads to weak results due to quasi-collinearity among the satisfaction variables. In Model 1 we present the results for the categories that show a higher impact. However, the differences encountered between factors of the same category are only indicative: we do not claim that one factor can be considered more relevant than another factor within the same category since this would require to consider them within the same model. In other words, in Model 1, the factors included are conceptually representative of the entire category to which they belong.

⁸ The decrease in the number of observations is due to very few missing values in the satisfaction variables for each dropped observation, impeding computation of the factor variables. In Model 2, the arithmetic means are computed using the variables available, ignoring missing values and allowing for consideration of the entire sample.

⁹ The standard deviation for the factor variables is fixed to 1.

¹⁰ As an alternative to this specification we estimated a probit model, on the subsample of researchers indicating the intention to leave the host country, where the dependent variable indicated the decision to return to the country of origin or move to a third country. Results were equivalent to those here discussed.

¹¹ We first run a quantile regression including country, age, gender, period of stay and research area as regressors of the number of papers per year. Secondly, we distinguished researchers that had a higher or lower productivity compared to the median predicted from the quantile regression model. In this way we identified two balanced groups of researchers with high and low productivity, relatively to other researchers with similar characteristics, at least based on the variables at our disposal. As such, by construction, the two groups have a significantly different productivity (about 2 and 0.18 papers per year, respectively) but almost identical main personal characteristics. We then run our analyses on the two different samples.