Sunday, 14 September 2014

International reading comparisons: is England really doing so poorly?

I was surprised to see a piece in the Guardian stating that "England is one of the most unequal countries for children's reading levels, second in the EU only to Romania". This claim was made in an article about a new campaign, Read On, Get On, that was launched this week.

The campaign sounds great. A consortium of organizations and individuals have got together to address the problem of poor reading: the tail in the distribution of reading ability that seems to stubbornly remain, despite efforts to reduce it. Poor readers are particularly likely to come from deprived backgrounds, and their disadvantage will be perpetuated, as they are at high risk of leaving school with few qualifications and dismal employment prospects. I was pleased to see that the campaign has recognized weak language skills in young children as an important predictor of later reading difficulties. The research evidence has been there for years (Kamhi & Catts, 2011), but it has taken ages to percolate into practice, and few teachers have any training in language development.

But! You knew there was a 'but' coming. It concerns the way the campaign has used evidence. They've mostly based what they say on the massive Progress in International Reading Literacy Study (PIRLS), and the impression is they have exaggerated the negative in order to create a sense of urgency.

I took a look at the Read On Get On report. The language is emotive and all about blame: "The UK has a sorry history of educational inequality. For many children, this country provides enormous and rich opportunities. At the top end of our education system we rival the best in the world. But it has long been recognised that we let down too many children who are allowed to fall behind. Many of them are condemned to restricted horizons and limited opportunities." I was particularly interested in the international comparisons, with claims such as "The UK is one of the most unfair countries in the developed world."

So how were such conclusions reached? Read On, Get On commissioned the National Foundation for Educational Research (NFER) to compare levels of reading attainment in the UK with that of other developed countries, with a focus on children approaching the last year of primary schooling.

Given the negative tone of "letting down children", it was interesting to read that "In terms of its overall average performance, NFER’s research found England to be one of the best performing countries." I put that in bold because, somehow, it didn't make it into the Guardian, so is easy to miss. It is in any case dismissed by the NFER report in a sentence: "As a wealthy country with a good education system, that is to be expected."

The evidence of the parlous state of UK education came from consideration of the range of scores from best (95th percentile) to worst (5th percentile) for children in England. Now this is where I think it gets a bit dishonest. Suppose there were a massive improvement in scores for a subset of children, such that the mean and highest scores went up, but with the lowest scoring still doing poorly; presumably, the shrill voices would get even shriller, because the range would extend even further. This seems a tad unfair: yes, it makes sense to stress that the average attainment doesn't capture important things, and that a high average is not a cause for congratulation if it is associated with a long straggly tail of poor achievers. But if we want to focus on poor achievers, let's look at the proportion of children scoring at a low level, and not at some notional 'gap' between best and worst, which is then translated into 'years' to make it sound even more dramatic.

The question is how does England compare with other countries if we just look at the absolute level of the low score corresponding to the 5th percentile. Answer: not brilliant – 16th out of the 24 countries featured in the subset considered by the NFER survey. But, rather surprisingly, we find that the NFER survey excluded New Zealand and Australia, both of whom did worse than England.

So do we notice anything about that? Well, in all three countries, children are learning English, a language widely recognized as creating difficulty for young readers because of the lack of consistent mapping between letters (orthography) and sounds (phonology). In fact, when looking for sources for this blogpost, I happened upon a report from an earlier tranche of PIRLS data, which examined this very topic, by assigning an 'orthographic complexity' score to different languages. The authors found a correlation of .6 between the range of scores (5th to 95th percentile again, this time for 2003 data) and a measure of complexity of the orthography. I applied their orthography rating scale to the 2011 PIRLS data and found that, once again the range of reading scores was significantly related to orthography (r = .72), with the highest ranges for those countries where English was spoken – see Figure below. (NB it would be very interesting to extend this to include additional countries: I was limited to the languages with an orthographic rating from the earlier report).
PIRLS 2011 data: range of reading attainment vs. orthographic complexity
International comparisons have their uses, and in this case they seem to suggest that a complex orthography widens the gap between the best and worst readers. However, they still need to be treated with caution. I haven't had time to delve into PIRLS in any detail, but just looking at how samples of children were selected, it is clear that criteria varied. In particular, there were differences from country to country in terms of whether they excluded children who were non-native speakers of the test language, and whether they included those with special educational needs. Romania, which had the most extreme range of scores between best and worst, excluded nobody. Finand, which tends to do well in these surveys, excluded "students with dyslexia or other severe linguistic disorders, intellectually disabled students, functionally disabled students, and students with limited proficiency in the assessment language." England excluded "students with significant special educational needs". Needless to say, all of these criteria are open to interpretation.

I'm not saying that the tail of the distribution is unimportant. Yes, of course, we need to do our best to ensure that all children are competent readers, as we know that poor literacy is a major handicap to a person's prospects for employment, education and prosperity. But let's stop beating ourselves over the head about this. Research indicates that the reasons for children's literacy problems are complex and will be influenced by the writing system they have to learn (Ziegler & Goswami, 2005) and constitutional factors (Asbury & Plomin, 2013), as well as by the home and school environment: we still have only a poor grasp of how these different factors interact. Until we gain a better understanding, we should of course put in our best efforts to help those children who are struggling. The enthusiasm and good intentions of those behind Read On, Get On are to be welcomed, but their spin on the PIRLS data is unhelpful in implying that only social factors are important.

Asbury K, and Plomin R. 2013. G is for genes: The impact of genetics on education and achievement. Chichester: Wiley Blackwell.

Kamhi AG, and Catts HW. 2011. Language and Reading Disabilities (3rd Edition): Allyn & Bacon.

Ziegler JC, & Goswami U (2005). Reading acquisition, developmental dyslexia, and skilled reading across languages: a psycholinguistic grain size theory. Psychological bulletin, 131 (1), 3-29 PMID: 15631549


  1. 20% of children still failing. Say No more.

  2. Thanks Dorothy! I wanted to draw that graph for a long time, as here in France we also have heated debates about the interpretation of international comparisons (mostly PISA).

    Three remarks about your graph:
    - first, it would be useful to draw orthographic complexity against mean reading performance, since after all the real prediction is that orthographic complexity influences reading acquisition on average. I would therefore expect a stronger correlation than on your graph.
    - Then, there is the complex issue of the relationship between mean and range. Many factors probably intervene. It might be that increasing the mean mechanically entails increasing the range. But maybe some countries manage to escape that law, and then it would be interesting to investigate why. At any rate, it would be informative to draw that graph for a start.
    - the orthographic complexity scale that you used is obviously arbitrary. In fact it seems to be a simple ranking of languages (with some missing). Using a real estimate of orthographic complexity would push French and English higher up, and might change the relationship to some extent (perhaps more linear?). You can find such estimates in papers by Seymour, Borgwaldt, Landerl...