Linguistic and nonlinguistic factors determining proficiency of English as a foreign language: a cross-country analysis

Abstract

This article is one of the first studies to use econometric techniques to examine the factors influencing proficiency of English as a foreign language. Specifically, this article aims to investigate, at cross-country level, the linguistic and nonlinguistic factors influencing the variation of proficiency in the use of English as a second language. As a proxy for English proficiency, we use the average Test of English as a Foreign Language (TOEFL) scores achieved by residents of a total of over 60 countries. The regression results suggest that linguistic factors such as historical affinity and similarity in word order between English and a given language have an influence on the proficiency in English achieved by people whose native language is not English. Among the nonlinguistic factors, expected years of schooling and degree of globalization have a positive relation with proficiency in English. These results seem robust, even when the computer-based TOEFL score is used in place of the paper-based TOEFL score and even when the overall score is replaced by the respective score in listening comprehension, structure and written expression, and reading comprehension.

Acknowledgements

We are grateful to Joshua Aizenman, Paul Miller, Chan-Hyun Sohn, Chris Wolfe, the two referees of this journal and participants in the Korea and the World Economy V Conference, Seoul, Korea, July 2006, for their helpful comments. We also thank Euijeong Park and Jakyung Ko for their research assistance. All errors are ours.

Notes

¹ At the microlevel, there have been a number of studies that investigated the effect that linguistic distance has on English language fluency for immigrants in the United States. See, for example, Chiswick and Miller (1998, 2001, 2005).

² Chiswick and Miller (2005) developed a scalar measure of the distance between English and 44 different languages, based on the difficulty that Americans have in learning other languages. This measure can also be used as a determinant of English language proficiency for nonnative English speakers.

³ Satem and Centum are, respectively, the Avestan (an ancient Iranian language) and Latin words corresponding to ‘hundred’.

⁴ The correlation between Alesina et al.'s (2003) linguistic fractionalization index and ethnic fractionalization linguistic index is 0.70.

⁵ http://publications.worldbank.org/WDI

⁶ http://unstats.un.org/unsd/demographic/products/socind/education.htm

⁷ One may think that import share of GDP might also proxy the degree of everyday exposure to English and its importance. As a matter of fact, we also tried to include import share of GDP or total trade (sum of exports and imports) share of GDP in place of export share of GDP as an explanatory variable, but found that these alternatives were not as robust as export share.

⁸ We also attempted to include the number of outbound international travellers or the total number of international travellers but this is possible only with a great loss of degrees of freedom, as the data on outbound travellers are available only for a small number of countries.

⁹ Ho (2003).

¹⁰ This index has also been used to examine the link between globalization and economic growth. See Dreher (2006), for example.

¹¹ http://www.cepii.fr/distance/geo_cepii.xls

¹² As an alternative specification, we also take the log of the TOEFL score.

¹³ As a matter of fact, we also tried to lag these variables and found that the results were qualitatively the same.

¹⁴ Note that even though AFFINITY and WordOrder are time invariant, correlation coefficients are different for 1997/98 and 2004/05 because the samples are slightly different.

¹⁵ We used the robust errors correction procedure for heteroscedasticity in the STATA software package.