For decades, we’ve had tools to measure the complexity of writing that still survive in some writing software as “Text Analysis.” For example, Flesch’s Reading Ease score and Flesch-Kincaid scores count syllables and words in sentences. However, the scoring is anything but transparent. To arrive at a Flesch score, you—or, more likely, your software—you rely on the formula:
206.835-(1.015 x Average Sentence Length) – (84.6 x Average Syllables per Word).
Alternatively, you can use the more user-friendly Flesch-Kincaid which correlates the Flesch score with the estimate percentage of Americans who can comprehend the content measured by Flesch Reading Ease. If an article scores between 0-30, only university graduates can fully understand the content. In contrast, a fifth grader can understand any paragraph in the 90-100 range, comprehensible to over 93 percent of Americans. The lower the score, the easier the reading. You can try this formula out for yourself. I ran Dr. Seuss’ Green Eggs and Ham through the Flesch-Kincaid formulas and discovered the book scores an improbable -1.3. This score suggests that the ideal reader of Green Eggs and Ham is a fetus.
Opaque formulas and bizarre outcomes aside, readability formulas reveal something valuable about how we measure the complexity of sentences: counting will only get you so far. To date, studies of students in primary and second education correlated growing sophistication in writing with lengths of sentences and clauses. This correlation makes sense, since longer sentences rely on phrases and clauses, a marker for a writer’s command over sentence structure. But counting alone is problematic. For instance, counting the syllables in a word to determine its difficulty can dramatically skew measures of complexity. Even a two-syllable word can vary dramatically in the demands it places on reader’s understanding. Consider praxis and baseball, both two-syllable words. While your average third grader can read baseball with ease, the word praxis can send even some PhDs scrambling for their dictionaries.
Enter the Lexile® Framework, commercially available software that uses sentence length combined with the frequency with which readers commonly encounter words. In addition to containing a corpus of over 100 million books, articles, and websites world-wide, Lexile is also highly influential in determining the reading levels of materials in primary and secondary education. Moreover, Lexile scores appear alongside articles in library databases, providing scores on everything from an article in a local newspaper or The New Yorker to books. But researchers have largely focused on Lexile’s ability to determine age-appropriate reading for students in elementary and secondary education. This omission prompted graduate student Samantha Miller and I to measure the validity of Lexile in assessing the overall sophistication of sentences and paragraphs in an article published in The International Journal of Business Administration.
We found that Lexile correlated highly with three robust measures of textual sophistication, which we measured using software that measured nineteen measures of the complexity of sentence structure. Lexile correlated highly with the three most robust measures of sentences’ complexity: the median length of sentences and clauses, as well as the use of complex nominals—or noun phrases. Lexile correlated most highly with the use of complex nominal per clause or sentence (p=<0.0001) and with median lengths of clauses (p=<0.0002) and median lengths of sentence (p=0.0013).
The takeaway? Lexile’s algorithms robustly determine the sophistication of words by matching words writers use against a still-growing body of 100 million texts, valuable when paired with other measures for assessing sentence-level complexity. The result? Those puzzling Lexile scores that crop up when you access an article from a library database reliably predict the difficulty of the content you’re about to read.