I’ve occasionally heard academics making a recommendation that can be summarized in four words: “Do not read, write!” The last time was at a conference dinner where a famous Stanford economist was mentioning that this advice had been given to him by an even more famous Harvard economist.[1]
At first sight, this is stupid advice. Why are we writing for? To disseminate ideas, change people’s views and ultimately, their practices. Ideas cannot obviously be disseminated if no one reads the sentences that express them. There is no point in writing if no one reads the end product of this activity. This is obvious enough, and therefore there is no chance that this advice can be understood as a universal prescription.
People who formulate this advice rather do so in terms of individual strategy. And this makes far more sense. Young scholars generally feel the need to read a lot. This is because they have to absorb a huge quantity of knowledge that has been produced over several decades by their predecessors. To contribute to a field of research on a specific topic, you need at least two things. First, to know the state of the art on the topic in which you are interested, and so be aware at least of the most recent literature. Second, to master the techniques and methods that can help you to produce some new and relevant knowledge in the field. You cannot acquire any of those without reading. The risk however is that you’re reading too much. Decreasing marginal returns also apply here. The additional benefits of reading a piece of research on a given topic tend to decrease as you read more, while the marginal (opportunity) costs remain the same, if they are not increasing. There is therefore somewhere an optimal amount of reading that must be done, precisely at the point where marginal costs and benefits are equal.
However, it is not easy to identify such a point. Partly, this is due to the fact that on any given topic, there are several questions and sub-questions that must be considered with non-fully overlapping literature on each of them. The computation is thus difficult to realize. This is equivalent to an optimization problem in search theory where you have to determine the optimal quantity of information you need before settling not on one good, but on a multidimensional vector of complementary goods. Not really a problem in theory, but a difficult task in practice.
There are two additional difficulties. First, the optimization exercise is by nature in terms of expectations. By definition, you cannot know the marginal value of a piece of information that you do not have yet. The best you can do is to form an estimate taking the form of a probability distribution of the marginal value of any new piece of information. Obviously, this involves the formation of beliefs about this probability distribution. Behavioral economics indicates that beliefs are easily biased. This is compounded by the second difficulty. It is far easier to read than to write. Reading is mostly a passive activity – though the most effective forms of academic readings are semi-active in the sense that you should take notes, try to summarize sections and chapters, and go back and forth between what you’re reading and the state of your thoughts or what you have already written. Whatever the way you read, it remains that writing requires far more effort and concentration. Moreover, while some readings may be difficult, it is always possible to skip (consciously or not) the difficulty by overlooking the parts that we don’t understand. This is not possible when you write. If you’re blocked, you’re blocked. If what you write is bad, it will be obvious to you and others almost immediately. If you didn’t understand what you’ve been reading, this will eventually manifest only later.
This second difficulty activates a second range of biases, especially those related to time preference. While over the long run, the benefits of (good) writing are superior to the benefits of mildly useful readings, the immediate costs are higher and also more salient. This can easily generate the kind of procrastination typical of hyperbolic or quasi-hyperbolic forms of discounting largely documented by behavioral sciences. The result is an indefinite postponement of the writing activity, with more or less relevant readings as a substitute.
With all this in mind, we can understand the “Don’t read, write” advice as a useful heuristic aiming at counterbalancing a natural tendency to differ difficult tasks. Any person who has done (or is doing) a PhD and eventually has supervised PhD students will acknowledge this point. An issue I have however with this recommendation is that it can easily be misunderstood as a statement about the usefulness (or lack thereof) of reading as part of the academic endeavor. This echoes my concern with the general tendency that academics are spending less and less time reading the work of others, especially when it takes the form of a long book. For sure, there are specificities tied to fields and disciplines. There are disciplines (intuitively, I would say mathematics and some natural sciences) where significant contributions can be made with a limited amount of well-targeted readings. This is far less obvious in humanities, with social sciences somehow occupying an intermediary position on the spectrum.[2]
Actually, in terms of social practice, I wonder if the problem is not the reverse of the one that emerges at the individual level. Are we not, collectively, writing too much? The number of academic journals has never been so high. This is related to the fact that there is a growing number of academics worldwide and that academics (essentially due to institutional pressures) have never published so much. On almost any given topic, there are so many new publications that it is impossible to keep track of all of them. Of course, in these circumstances, trying to read everything would be foolish. The question however is about the social value of these so numerous publications that almost no one reads – the question takes an even bigger dimension if we include non-strictly academic contributions, such as those that appear on Substack for example.
There are arguments in favor of the current model. For instance, if you see ideas as books that are kept in an infinitely large library and producers of ideas as readers searching for them, the more you have producers, the likelier to find the most relevant books. But this really applies only if the search in the library is necessarily random. Obviously, a non-random search is possible and more efficient. But to know where to search, you need to have an idea of what has been found and where in the library. Put less metaphorically, efficient writing at the collective level requires enough resources dedicated to reading!
[1] I must make the precision that I was not directly part of the conversion, just lucky enough to be sited next to the Stanford economist.
[2] The irony here is that articles in biology or physics routinely have more than a hundred references in their reference lists, resulting in journals having impact factors that are insanely high compared, e.g., to history of economics or philosophy journals.