Thursday, November 28, 2013

Why you shouldn't put numbers in text

Read any piece of business writing, and you're bound to encounter this sort of gibberish:
In the total throughput table, XYZ is ranked at the market leader, with a throughput of 66.3 million units, up from 60.9 million units in 2006, and a global share of 13.3%. ABC is in second position, with 60.3 million units (12.1%) followed by PQR with 54.7 million units (11%), DEF with 43 million units (8.7%) and JKL, which moved 27.3 million units and has a market share of 5.5%.
Business reports typically have dozens of tables. That’s where the numbers should go. We can’t completely avoid putting numbers in the text, but it should be avoided wherever possible. 

Here’s why.
If you can decipher the following quotation (academics often seem incapable of writing clearly, even when their studies are about comprehension), it explains why we should keep numbers separate from words:
Locating information such as facts, names, or numbers in text is a reading task requiring comprehension that is distinct from text recall in two respects:
(1) cognitive processes that control reading comprehension and locating information are expected to be different;
(2) the frequency of engagement in comprehension and locating are expected to be independent.

Journal of Educational Psychology, Vol 79(3), Sep 1987, 220-227. Guthrie, John T.; Kirsch, Irwin S. 
Put simply, a person reading text has his mind in ‘story mode’. To understand any facts or numbers that appear in the text, he needs to stop and switch to ‘data mode’. Either he won’t bother, in which case your numbers have no value, or he will switch mental modes and so lose the thread of your argument. This happens automatically when you break your reading to look at a graph or table, but it doesn’t happen at all when you read text - unless you stop reading.

Readers don’t absorb numbers in text, which is why we can write “rates improved by 25% from $3,000 to $4,000” without noticing the error. You usually have to read it twice to spot the mistake, switching your brain from narrative mode to mathematics mode. This means you have to read it twice to understand it. You should never have to read anything twice.

To demonstrate, try reading the following paragraph, which is a typical piece of business writing and is by no means the worst example I could have found. Read it once only and at normal reading speed, as if it were part of a 30,000-word report and you didn’t have a spare week to get through it all (which most of our readers don't):

“In fact, net sales fell from $3,470 million in 2009 to $3,227 million in 2010 – a reduction of 7%. A similar impact was felt on operating income, which fell by $39 million (26%) from $147 million in 2011 to $108 million in 2012.”

If you're familiar with financial reports, you probably feel you understood it. But all you really understood was: “Sales fell and so did income.” If you’re a skilled and attentive reader, you might also have gleaned that income fell by quite a lot. No reader can absorb 11 items of data in one 43-word paragraph. The author spent a lot of time finding those numbers, working out the percentages and carefully transcribing them. That time was completely wasted.

Moral: Tables are for supplying data. Text is for telling a story. It is only by telling a story that we convert data into knowledge


  1. Thank you for this. It shed new light for me on the type of technique that writers of reports should use in the summarization of test results--for example, school psychologists who are reporting the results of IQ assessments. Normally this information is presented in both table form and narrative form after the table. What you're saying is that the narrative summary of the results should stick to the verbal interpretation of the scores and leave the actual numbers out. For example, the table can say Johnny scored a 98 in Verbal Comprehension on the Wechsler, but the narrative following the table should just say that Johnny's score was within average range. Similarly, the narrative is not the place for percentiles, just the table?

    This is valuable information to know, especially when considering that the least prepared audience for interpreting these reports is parents, to whom the results are absolutely crucial regardless. If they have trouble switching from "story mode" to "data mode" to understand what a school psychologist is trying to convey about Johnny's ability, it behooves the psychologist to present the information in the easiest-to-digest format.

  2. Seen on a boxing website:

    WBA female light minimumweight (102-pound) champ, Japanese footworker Ayaka Miyao (16-5-1, 1 KO), 99.75, retained her belt for the third time as she fully utilized her speed and finesse to earn a unanimous decision (97-93 twice, 96-94) over mandatory challenger Gretchen Abaniel (13-5, 4 KOs), 101.5, Philippines, over ten fast rounds on Thursday in Tokyo, Japan.

  3. This is a well-intentioned suggestion, but overlooks the purpose of much of the writing to which you refer. It is not intended to be read as if it were 'faits divers' (I am sure there is an English equivalent, but at the moment it escapes me); moreover the suggestion of 'story' mode raises the question of how one would read Rimbaud or Henry James, Beckett or Sterne. The intended reader does want the figures, and is prepare to mull them over. A table needs interpretation (particularly when most tables are so badly presented). That is presumably the function of the text you quote, which saves the reader the trouble of reading throught he detailed accounts.. (Perhaps I could draw your attention, if you have not come across it, to Andrew (A.S.C.) Ehrenberg's "Data Reduction", which guided me when working as a statistician.)

    1. The idea that readers want the numbers and will make an extra effort is often used to justify such writing, but studies by psychologists have shown that this isn't the case. Regardless of their job, readers read as human beings. As you say, text should be used to analyse the data (to turn data into knowledge), but it shouldn't be the medium for presenting raw data. You don't correct bad tables by producing bad writing.

  4. I realize I’m late to the table but I just wanted to add my ‘thank you’. I’ve been working with scientists who have to write reports for non-technical audiences and this research explains what I’ve been saying but in a much clearer way! I really like the ‘story mode’ and ‘data mode’ contrast.