Attempts to identify factors which affect readability can be traced back a long way, probably back to the dawn of writing. But for our purposes they can be traced back to Thorndike (1921), who examined word frequencies and started a strand of thought which lives on today in corpus studies. Most subsequent readability formulae have included word frequency and/or word length. The first real readability formula was that of Lively and Pressey (1923) which was based purely on vocabulary difficulty.
A number of other factors affecting readability have been considered. Vogel and Washburne (1928) counted number of different words in a sample the number of prepositions, the total number of words not on the Thorndike 10,000 most frequent word list, and the number of clauses in 75 sentences. Gray and Leary (1935) listed factors under a) content, b) style, c) format, and d) general factors of organisation, although their readability formulae took into account only variables listed under style.
Perhaps one of the best known indices of readability (not least because it is still available in modern word-processing programs) is that of Flesch (1943, 1946, 1947, 1962). The Flesch Reading Ease Formula takes into account only words/sentence and syllables/word.
Dale and Chall (1948) also used two variables (average sentence length and word familiarity) in their formula.
Many other formulae of varying complexity followed, notably those of Fry (1964, 1977), Bormuth (1966), Coleman & Liau (1975).
Many other factors to insert in regression formulae or different methods of establishing readability have also been proposed such as (to name only two) counting abstract words (Flesch, 1943, Cohen, 1975) and propositional density, and inferences, (Kintsch, 1974) There are difficulties, however, in defining exactly what is or is not abstract and difficulties with conducting propositional analyses of long texts.
The objective of compilers of such formulae was to find a few simple text variables which correlated with reading difficulty in order to be able to predict the difficulty readers would have of comprehending a particular text. It is probably unfair to suggest that anyone was under the illusion that the few factors used in the formulae were the sole contributors to text complexity. It also seems to be the case that readability formulae were used for purposes for which they were not intended: formulae using few variables were intended as quick predictors of readability and not as suggestions as to how texts should be written. Nevertheless, the usefulness and validity of such formulae were called into question (Irwin and Davis, 1980; Davison & Kantor, 1982).
Criticisms of Formula-based Approaches
It is not surprising that formulae based on so few factors should have been criticised given the intuitive feeling that reading is a complex process (see Klare 1984 pp682-683 for a list of criticisms and critics).
Although readability formulae can usually be shown to correlate to some extent with text difficulty (or comprehensibility), they have little to say about causality. Furthermore, the percentage of variance attributable to the factors used in formulae has been shown to be quite small. Freebody and Anderson (1983) showed that vocabulary accounted for less than 5% and Davison, Wilson & Herman (1983) similarly showed that sentence length accounted for a very small percentage.
Davison and Kantor (1982:189) state:
"Objections may be made to readability formulas in general. Reading difficulty may be affected by the purposes and background of the reader and the inherent difficulties of the subject matter; it is not just a function of measurable properties like length and vocabulary. "
It is quite possible that sentence length correlates with difficulty simply because length correlates with other factors which might contribute to comprehensibility (sentences with complex co-ordination or subordination are likely to be longer and it is surely the difficulty of establishing the relations of, for example, subordination which causes difficulty rather than sentence length per se).Smith (1988) considers length to be just one of a number of factors which may contribute to linguistic complexity. She distinguishes between systematic complexity, surface syntactic complexity, interpretive complexity, and phonological complexity and suggests that there are interactions between these components. Since "sentences that are high in interpretive complexity (with missing elements) tend to be low in amount, or length in number of words" (1988:250), it seems that length may not be a good indicator of difficulty simply because different types of complexity are confounded and length is not positively correlated with all of them.
Intercorrelation of indices
It is easy enough nowadays with a modern word processor to compare the readability indices of a few texts. The results are often surprising. While there may be a doubling of difficulty for two different texts according to one index, another may register hardly any change at all. This fact alone should serve as a caution against uncritical use of such devices.
Problems of intercorrelations and the problems of relating reading difficulty indices to American grade levels have been pointed out by Klare (1984:706).