Please cite this article as: Brown, D. (2017). Coverage-based frequency bands: A proposal. Vocabulary Learning and Instruction, 6 (2), 52–60. doi: 10.7820/vli.v06.2.Brown Vocabulary Learning and Instruction Volume 6, Issue 2, December 2017 http://vli-journal.org Coverage-based Frequency Bands: A Proposal Dale Brown Kanazawa University doi: http://dx.doi.org/10.7820/vli.v06.2.Brown Abstract Second language vocabulary research makes much use of word frequency lists and their division into bands. In recent years, bands of 1,000 items have become conventional. However, there does not seem to be any frm basis or rationale for this. Conventional banding may be questioned since the utility of words varies greatly depending on frequency, because there are enormous differences in frequency within higher bands, and because the reliability of the placement of words in bands becomes progressively poorer at lower frequency levels. This article suggests an alternative ap- proach: basing bands on coverage levels. Because of the frequency distri- bution of words, this means the highest frequency bands would contain very few words, while lower frequency bands would contain a great many words. The article shows how such bands can be constructed and presents a re-analysis of the results of a vocabulary test designed with conven- tional bands in terms of coverage-based bands. This re-analysis produces a very different profle of learners’ knowledge, and it is argued that the shape of this profle may be more useful in terms of guiding instruction in that it gives a clearer indication of which words should be targeted for a group of learners. It is further argued that the smaller number of words contained in coverage-based bands at higher frequency levels makes them a more feasible basis for instruction. The article thus concludes that coverage-based bands may be a fruitful avenue for researchers to explore. 1. Introduction Much work on L2 vocabulary makes use of word frequency lists and in par- ticular their division into bands. Bands of 1,000 items have become conventional through applications in vocabulary testing, such as Meara’s yes/no tests (Meara & Milton, 2003; Meara & Miralpeix, 2006) and the vocabulary size test (Nation & Beglar, 2007), and the lexical profling of texts (e.g., Nation, 2006; Webb & Ro- dgers, 2009) and of learners (Laufer & Nation, 1995; Milton, 2007). There does not, however, appear to be any frm basis or an explicit rationale for the use of 1,000-item bands. Kremmel (2016) has recently criticized this con- vention since the utility of words varies greatly with frequency. This is clear in the coverage levels provided by successive bands. For example, Webb and Rodgers (2009) looked at coverage in a corpus of flm scripts and found that while the frst 1,000 word families covered 86.5% of the words in the corpus, the second 1,000 covered 4.2%, the third 1.7%, down to the tenth 1,000 which covered just 0.1% of the scripts. A second problem is that at higher frequency levels there is massive variation in the frequency of the items within a band. In Davies’ COCA-based