I am obliged by my profession to read between thirty and fifty thousand words on most working days. This has continued for long enough that I no longer regard reading as an activity separate from the rest of life. It is what I do while earning my living, and it remains one of my principal recreations when the day’s work is finished.

My preference, when reading a favourite work of literature, is still for a printed book. There is a settled quality in print that no electronic device quite reproduces. A book has weight and position. One remembers that a passage appeared towards the foot of a left-hand page, perhaps a third of the way through the volume. Its physical existence gives a shape to the act of reading. I do not require an experiment to tell me that this has psychological value.

Even so, my house is of finite dimensions. The number of books I should like to possess is not. After several decades of accumulation, this conflict between desire and architecture has been decided in favour of architecture. I still buy books, though I can no longer pretend that every book worth reading must also occupy several inches of shelving until my executors come to dispose of it.

I have been a heavy user of computer technology since 1983. Since about 1995, I have probably read more from screens than from paper. I am now seventy-three, and therefore qualify as an older reader for the purposes of academic classification. I mention these facts because they bear on a question that has attracted a surprising quantity of research: whether reading from a screen is inferior to reading from paper.

Virginia Clinton has considered this question in an article published in the Journal of Research in Reading in 2019. Its title is “Reading from Paper Compared to Screens: A Systematic Review and Meta-analysis.” I approached it with low expectations. Most published research nowadays is worthless, and much of it involves scholarship so dubious that the word “research” functions less as a description than as a claim to immunity from ordinary disbelief. Where this activity is supported by the taxpayer, it is generally a fraud on the public.

Miss Clinton’s article is a rare exception. It asks a comprehensible question and sets out the evidence bearing on it. More unusually, it distinguishes between a small statistical effect and a universal law of human behaviour. The paper has weaknesses, some of them considerable, but its author identifies many of these herself. This is what scholarship used to mean before research became an industry.

The article is not a report of one new experiment. It is a systematic review and meta-analysis of experiments conducted between 2008 and 2018. Clinton identified twenty-nine reports containing thirty-three studies. These provided 2,799 participants for the analysis of reading performance. Smaller bodies of evidence were available for reading time and for what the researchers call “calibration,” meaning the accuracy with which readers estimate their own comprehension.

The studies had to satisfy sensible conditions. They needed to compare the same text on paper and on a screen. Their participants had to possess established reading skills, and the experimental arrangements had to permit a reasonably clear comparison between the media. Studies of people learning to read were excluded, as were studies involving second-language reading or identified visual and learning disabilities.

The main result was a small advantage for paper in tests of comprehension. Expressed in the customary statistical form, the average difference was a Hedges’ (g) of about minus 0.25 when screen reading was compared with paper. This is not a finding that readers understand paper while becoming helpless before a screen. It is a modest average difference collected from studies with varied participants and methods.

There was no reliable difference in reading time. Readers spent broadly similar amounts of time on the material, though their test performance was slightly better after reading it on paper. They also appeared to judge the adequacy of their comprehension somewhat more accurately when using paper. Clinton therefore concludes that paper may be a little more efficient, since it produced a small improvement in results without requiring additional time.

The most interesting qualification concerns the kind of material read. For expository writing, the estimated advantage of paper was larger, though still modest. For narrative writing, the estimated difference was close to nothing and statistically unreliable. This may mean that screens are less suitable for concentrated study than for reading novels. It may also mean that the experiments have not yet isolated the relevant causes.

Clinton is properly cautious. The review contained many more experiments involving expository texts than narrative texts. The narrative estimate came from only seven studies. It would therefore be foolish to turn a tentative distinction into a commandment that textbooks belong on paper while novels may safely be entrusted to glass.

There is a similar problem with the category of “screen.” A large computer monitor is not the same object as a telephone. An electronic reader displaying well-set type is not the same as a badly scanned PDF opened on an elderly office machine. Pagination and scrolling may impose different demands. Search facilities, text enlargement and the ease of annotation may also alter the experience. To combine every electronic device into one category is convenient for statistical purposes, though less satisfactory for deciding how one ought to read.

The evidence about reading time is especially uncertain. The results varied greatly between the studies. Some found that participants read more slowly from paper, while others reported the reverse. At least one experiment imposed different physical methods for answering questions in its paper and screen conditions. One may doubt whether the resulting difference measured reading rather than the inconvenience of the apparatus.

The article is also less informative about older readers than its analysis of age may suggest. Clinton divided participants into children and adults, but most of the adults were university students. Only one of the included studies directly compared younger adults with a group of older adults, whose average age was 66.8. It found broadly similar patterns in the two groups. This is useful evidence against the belief that advancing age necessarily makes screen reading difficult, but it tells us little about a seventy-three-year-old who has used computers intensively for more than forty years.

Indeed, I am unlike the ordinary participant in nearly all these experiments. A university student reading a short passage under supervision for the purpose of answering several questions is not doing what I do during a working day. Nor is he necessarily doing what I do in the evening, when I read for pleasure. An experiment can measure what has been placed within it. It cannot quietly expand its findings into circumstances never examined.

Long use matters. A reader who has worked from screens for several decades acquires habits suited to the medium. He adjusts the size of the text and moves between documents without conscious effort. He knows when to search for an earlier occurrence of a name and when to return to the beginning of an argument. He may also have trained himself to ignore the distractions that trouble less experienced readers.

Medium preference may matter as well. Clinton discusses evidence that readers sometimes judge their performance more accurately when using the medium they prefer. The causal relation is not clear. We may prefer a medium because we use it well, or use it well because we prefer it. Either possibility should discourage the issuing of general instructions based on small average effects.

My own judgment is that reading electronic copies is at least as efficient as reading from hard copy. I do not say that the two experiences are identical. The printed book has psychological advantages, and I prefer it for literature to which I expect to return for pleasure. But electronic text offers searchability and enlargement. These are not minor conveniences, particularly for an older reader whose eyes are no longer those of a young articled clerk.

Searchability changes the meaning of efficiency. In a laboratory, efficiency is commonly treated as comprehension divided by time. In professional reading, it may involve locating every occurrence of a phrase, checking whether an author has contradicted himself, or recovering a passage remembered only imperfectly. A paper volume may help me remember where a passage stood. An electronic copy allows me to find it in seconds. Both qualities are useful, though they are useful for different purposes.

Scaling is equally important. The size of print in an old book is fixed, and not always by a printer considerate of readers in their seventies. Electronic text may be enlarged until it becomes comfortable. This can reduce fatigue and permit longer reading without altering the substance of the work. A study that treats the ability to enlarge text as an incidental feature has overlooked one of the reasons for using the medium.

There is a greater advantage. Electronic copies make available material that would otherwise be practically inaccessible. I have recently been reading through an electronic archive of The Quarterly Review and Blackwood’s Magazine, extending across the nineteenth and early twentieth centuries. To assemble the original volumes would require more shelf space than I possess. Even access to a good institutional library would not allow me to consult them at any hour or search them as a single body of writing.

Without electronic copies, this reading would have been impossible. That ends any simple argument about the superiority of paper. The true comparison is not always between the same conveniently available work printed on paper and displayed on a screen. It may be between an electronic copy and no copy at all. A small advantage in remembering details from paper cannot compensate for the total absence of the text.

This is one reason why academic experiments should not be allowed to dictate ordinary life. Their narrowness is often a virtue, since a controlled experiment must exclude complications. The trouble begins when researchers or journalists forget what has been excluded. A finding about short-term performance on comprehension questions becomes a pronouncement on reading as such. The statistical tail then begins to wag the intellectual dog.

Clinton largely avoids this offence. She notes that many of the underlying studies had small samples and that several failed to report the reliability of their comprehension tests. She accepts that her review does not cover people learning to read or those reading in a second language. She also acknowledges that the evidence does not settle why paper may confer its small advantage.

The suggested explanations remain conjectural. Readers may be more prone to wandering thoughts when using screens. They may treat paper as a signal that a task is serious, while treating a screen as a signal that it is casual. They may encode physical position differently when a text remains fixed on a page. These possibilities deserve investigation, but none has yet been established as the governing explanation.

There is also a danger of confusing a screen with the habits commonly associated with it. A computer connected to the Internet may offer messages and other interruptions. A printed book ordinarily does not. Yet distraction is not an inherent property of illuminated pixels. One can disable notifications and read a local document without keeping several diversions open beside it. An experiment that compares an undistracted book with a badly managed computer may discover more about human indiscipline than about reading media.

None of this disproves Clinton’s result. Paper appears to offer a small average advantage in certain tests, especially those involving expository material. Anyone preparing for an examination might sensibly take account of this. One might also use paper for a difficult document if experience shows that it encourages closer attention. What does not follow is that screen reading is generally inefficient or intellectually inferior.

The sensible reader will use each medium according to its strengths. I shall continue to prefer printed copies of favourite literature, because pleasure is not reducible to the score on a comprehension test. I shall also continue to use electronic copies for professional work and for the exploration of archives that no private house could contain. This is not inconsistency. It is the ordinary allocation of different tools to different purposes.

Miss Clinton has produced a serious and useful article. It does not prove that paper is always better, and she does not claim that it does. It establishes a small average advantage under limited experimental conditions, while leaving ample room for experience and circumstance. In an academic world where vast amounts of public money are spent proving what was assumed at the beginning, this degree of honesty is sufficiently uncommon to deserve notice.

My shelves will therefore remain full, though they will not be permitted to bring down the ceilings. I shall keep buying those printed books whose physical presence adds to the pleasure of possession and rereading. For the rest, I shall enlarge the type, search the archive and continue reading from my screen. At seventy-three, I have no wish to choose between civilisation in print and civilisation in electronic form when I can make sensible use of both.