Grouping students according to their first language is a far more accurate way of predicting ability than grouping them by ethnicity, new research reveals.

Umbrella ethnic groupings, such as “black African” or “white other” hide significant differences between high- and low-achieving linguistic groups, academics have found.

The findings come from an analysis of the exam results of almost 4,000 11- and 16-year-olds in inner London. Researchers found that, while students categorised as black African tended to perform well overall (71 per cent achieved five or more A*-C grades at GCSE), there was huge range of achievement within that group.

Eighty-three per cent of students who spoke the Ugandan language Luganda as their first language achieved five A*-C grades at GCSE in 2011. By contrast, only 45 per cent of Congolese Lingala speakers reached the same level.

One of the main reasons that black African students performed so highly as an ethnic group overall was that one in five black African teenagers spoke English as a first language, said Feyisa Demie and Andy Hau, who were conducted to do the research on behalf of the London Borough of Lambeth.

Their findings were presented at the British Educational Research Association conference, held this month at the University of Sussex.

In contrast to the relatively high-performing black African category, only 58 per cent of students classified as “white other” achieved five good GCSEs. But, again, this category comprised students from a diverse range of cultural, geographical and linguistic backgrounds. These included French speakers, three-quarters of whom achieved five A*-C grades. But it also included Portuguese speakers, among whom fewer than half reached the same level. Spanish speakers scored only slightly higher.

Similar patterns were apparent in key stage 2 tests taken by 11-year-olds. Italian and German speakers were significantly more likely to achieve level 4 in all subjects than Spanish and Portuguese speakers. Lingala speakers again performed poorly, but 95 per cent of Nigerian Igbo speakers attained level 4.

Even ostensibly homogenous categories could mask linguistic differences, the researchers said. Indian students in the local authority spoke Gujarati, Punjabi and Hindi; Pakistani students spoke Punjabi and Urdu.

The academics called for linguistic data to be made more broadly available. “There is a need to unpick how national ethnic categorisations may be used to improve our understanding of the performance of pupils who speak different languages,” they said.

Charmian Kenner, deputy director of the Centre for Language, Culture and Learning at Goldsmiths, University of London, agreed that information about students’ linguistic backgrounds should be widely shared.

“Ethnic-minority achievement needs to be looked at in a much more fine-tuned way,” she said. “It’s in looking at those groups that are doing well, and finding out the reasons, that we can then find out what we can do to help other children.”

But Julie Gibbings, from the National Literacy Trust, said that first language should not be used by teachers in isolation. “There are a huge range of factors to take into account when measuring ability,” she said.