The passive voice and comprehensibility of biomedical texts: An experimental study with 2 cohorts of chiropractic students

Writings in the health sciences tend to be complex. This is natural, given what are often the relatively esoteric themes and the argumentative (in a scientific sense) nature of writings in the health sciences; for example, establishing cause-effect relationships from samples of populations. Efforts to render such writing more comprehensible are therefore laudable, especially as English becomes the lingua franca of health sciences education and research worldwide. As increasing numbers of health professionals undergo training in English, it becomes apparent that the language of their chosen discipline may present challenges.^1,2  This is likely as true for chiropractic students as for students in other areas of the health sciences.³  Furthermore, even health sciences students who are so-called native English speakers may be challenged by the vocabulary and writing conventions in their chosen discipline.⁴ 

Thus, the more understandable learning resources are, the more efficient the learning process will be.⁵  The overabundance of articles on how to write readable medical prose is testimony to this conventional wisdom (eg, see Alexandrov,⁶  Bredan and van Roy,⁷  and Steen⁸). However, commonly proposed strategies for comprehensible writing may not be grounded in good evidence. For example, it is not uncommon to see writers encouraged to favor the active voice over the passive voice—“the doctor treated the patient” versus “the patient was treated by the doctor”—perhaps based on the misconception that the passive voice results in longer and less-comprehensible constructions. Even the American Medical Association Manual of Style continues to encourage authors to “use active voice whenever possible,”⁹  while the British Medical Journal still advises authors to “use active voice but avoid ‘we did' or ‘we found.'”¹⁰ 

On the other hand, it has recently been argued that there are instances in medical writing in which the passive voice is preferable to the active voice, and that sweeping editorial guidelines may be overly simplistic and so do not serve their intended purpose.¹¹  To date, however, no experimental studies have challenged the assumption of enhanced comprehensibility for the active voice. In this age of evidence-based care, it is somewhat ironic that the teaching of health sciences literacy remains largely opinion based. Hence, the purpose of this study was to compare the comprehensibility of authentic samples of biomedical text expressed in the active and passive voices.

This study was approved by the research ethics board of Canadian Memorial Chiropractic College. As part of an annual assessment of incoming students, 161 2nd-year chiropractic students were recruited into the study by an in-class announcement on September 10, 2012. Once seated in the testing room and in sequence according to their seating, the students alternatively received 1 of 2 forms of the assessment. Hence, the allocation to cohorts was not randomized. Nonetheless, post hoc analysis of the cohorts showed that they were quite similar (Table 1).

The assessments used 2 strategies to test the effect of the passive voice on the comprehensibility of biomedical texts.

In the first instance, each of the 2 cohorts of students was presented with 1 sample of text in its original published form, with frequent use of the passive voice, and a sample of text in which the passive voice was replaced throughout the text with the active voice (text available from corresponding author). The text samples were extracted from the methods sections of 2 original research papers referenced within the undergraduate curriculum of the students.^12,13  The comprehensibility of the 2 examples of biomedical text was tested in the original published form and in the rewritten form. Thus, in more detail, 1 cohort received the original version of text 1 and the rewritten version of text 2. The 2nd cohort received the rewritten version of text 1 and the original version of text 2.

The readability statistics for the texts were obtained from the spelling and grammar checking function of Microsoft Word 2007 (Microsoft Corp, Redmond, WA), and as shown in Table 2, they were essentially identical for the original and rewritten versions of the texts. Additionally, the profile of the vocabulary was obtained by processing the text samples using the online software Web VP Classic v.4.0.¹⁴  Thus, the words within each version of each text sample were classified as belonging to (1) the general service list, the approximately 2000 most commonly occurring word families in the English language; (2) the academic word list, the approximately 570 word families, other than general service list words, commonly found in academic writings; and (3) all other words not belonging to the general service list or academic word list, so-called “off-list” words. Off-list words are more likely to be technical words specific to the theme of the text. As shown in Table 2, the vocabulary profiles were essentially unaltered by rewriting the sample texts in the active voice.

The students were asked to answer 10 true-or-false questions dealing with each of the 2 texts. Applying item response theory, the difficulties and sensitivities of each of the 10 questions were calculated for each of the 2 forms (active voice versus passive voice) of the text. The difficulty of each question was defined as the mathematical inverse of the prevalence of correct answers. For example, if half of the students answered a question correctly, that question would have a difficulty of 2.0 (the inverse of 1/2). The sensitivity of each question was defined as the prevalence of correct answers among the top-scoring quartile of the cohort divided by the prevalence of correct answers among the bottom-scoring quartile of the cohort.

Comparisons were also made between the raw and processed (see below) scores of the 2 cohorts for each of the 2 versions (original published form and rewritten form) of each of the 2 texts, essentially asking whether cohort scores were on average different depending on whether the text was presented in the active or passive voice. Raw scores were calculated simply as the percentage of correct answers. Processed scores were calculated by assigning each correct answer a weight equivalent to its calculated difficulty and then dividing the student's score by the total weight of all questions.

Twenty sentences were extracted from the Biomedical and Health Linguistics randomized controlled trials corpus¹¹  and presented in their original form (active or passive voice) to 1 cohort and then in the alternate (rewritten) form (text available from corresponding author) to a second cohort; for example:

Active voice version presented to 1 cohort: Question 21. “We also observed this minimal level of correlation in the subgroup of patients who subsequently had recurrent coronary events (r = 0.1, P = 0.004).” New England Journal of Medicine

Passive voice version presented to the other cohort: Question 21. “This minimal level of correlation was also observed in the subgroup of patients who subsequently had recurrent coronary events (r = 0.1, P = 0.004).” New England Journal of Medicine

Each version of the test contained the same number of sentences in each voice. Hence, within each cohort (ie, version of the test), students were presented with 10 original and 10 rewritten sentences, and among these sentences, 10 were in the active voice and 10 were in the passive voice. The sentences came from the “big five” medical journals: Journal of the American Medical Association, New England Journal of Medicine, Annals of Internal Medicine, British Medical Journal, and the Lancet. Students were asked to subjectively rate the comprehensibility of each sentence using the following scale: 1, completely clear; 2, quite clear; 3, moderately clear; 4, quite unclear; 5, completely unclear.

Seventy-eight complete copies of test 1 and 82 complete copies of test 2 were returned. The numerical ratings for the 2 versions of each sentence were averaged, and the averages for the active-voice and passive-voice versions of each sentence were compared using a paired, 2-tailed t test.

The readability statistics and vocabulary profiles for the 2 texts are presented in Table 2 and suggest that the readabilities of the 2 versions of each text (supplementary files) were not substantially different.

Table 3 shows, for text 1, the calculated difficulties and sensitivities of the 10 true-or-false questions when the text was presented in the passive versus the active voice. There was no statistically significant difference, per paired, 2-tailed t test, between the difficulties and sensitivities (p = .69 and .35, respectively) for the true-or-false questions when tested against text 1 in the passive versus active voice.

Table 4 shows, for text 2, the calculated difficulties and sensitivities of the 10 true-or-false questions when the text was presented in the passive versus the active voice. There was no statistically significant difference, per paired, 2-tailed t test, between the difficulties and sensitivities (p = .48 and .81, respectively) for the true-or-false questions when tested against text 2 in the passive versus active voice.

As shown in Table 5 and on the basis of subjectively rated comprehensibility, the versions of sentences presented in the active voice were, on average, somewhat less comprehensible (mean 2.21, SD 0.48) than the versions presented in the passive voice (mean 2.07, SD 0.47; p = .003 per 2 tailed, paired t test).

This study challenged the hypothesis that texts written in the active voice are more comprehensible than texts written in the passive voice. The study was conducted using students who were just commencing their 2nd year of study at a Canadian chiropractic college. Hence, all students already possessed at least an undergraduate university degree and had completed 1 year (approximately 1000 classroom hours) of additional study focusing on human biology but also involving a yearlong course in critical appraisal of research literature. The texts presented in this experiment were therefore representative of the reading materials which the students would be expected to access in their current year of study.

Objective measures of readability and vocabulary profiles were obtained for 2 samples of text which, in their original forms, made extensive use of the passive voice, as is conventional in biomedical texts.¹¹  When these texts were rewritten to eliminate the use of the passive voice, the changes in readability measures were miniscule or none. This contradicts the argument that the use of the passive voice produces more-complex or less-comprehensible texts. Furthermore, when students were asked questions pertaining to the original (passive voice) versus rewritten (active voice) versions of the 2 texts, there were no statistical differences in the mean difficulties or sensitivities of the questions. The individual comprehension questions satisfactorily discriminated between different levels of reading ability among the students. Specifically, 36 of 40 questions had sensitivities of >1.00, while 4 of 40 questions had sensitivities just equal to 1.00, and so overall the questions provide a valid measure of the students' reading comprehension. This indicates that the 2 versions of the same text were equally comprehensible to the students.

When students were asked to subjectively rank the comprehensibility of sentences written in the active versus passive voice, they indicated that overall sentences presented in the passive voice were somewhat (and statistically significantly) more comprehensible. When compared with the objective measures of comprehensibility, this seems to reflect a bias on the students' part in favor of the passive voice, and, given that students are familiar with medical texts that make frequent use of the passive voice, this general preference seems perhaps not surprising. On the other hand, it is not clear if students prefer to encounter particular verbs in 1 voice versus the other. Each of the main verbs in the sentences listed in Table 4 was presented 4 times—twice in the active voice and twice in the passive voice. The verb obtain was the only verb for which the passive voice was twice rated as much more comprehensible than the active voice. The verb obtain was previously shown to have a strong affinity for the passive voice in reports of randomized controlled trials.¹¹  On the other hand, other verbs which had previously shown a strong affinity for the passive voice (eg, associate, consider) were not consistently ranked as more comprehensible when presented in the passive voice in the present study. Hence, there is not a strong argument that students are conditioned by medical writing to prefer certain verbs in the passive voice.

A limitation of this study is that subjects were not randomized to the 2 test cohorts. Further, there may be important differences in the literacies of students at 1 institution versus another, or, indeed, 1 country versus another, limiting the generalizability of this work.

Collectively, these results indicate that for the test subjects and texts used in this study, use of 1 voice versus the other did not affect objectively rated comprehensibility, and that there was only a marginally greater subjectively rated comprehensibility for the passive voice. These results are congruent with a previous paper which proposed, on theoretical grounds, that the passive voice is appropriate in medical writings. Furthermore, these results argue against editorial guidelines which advocate for deliberate selection of the active voice in medical writings. Given that articles in leading journals and accessed by health sciences students are already difficult to read,^4,15  editors of publications in the health sciences may wish to take a more evidence-based approach to developing recommendations on language use.

The authors declare that they have no conflicts of interest. No external funds were received in support of this work.