Next time you do a word count, check out one of the stats offered up by your word processor – the Flesch reading ease index. It was devised in the 1940s by an Austrian, Rudolf Flesch, and is a simple number crunching exercise that uses the number of words per sentence, and the number of syllables per word, to put a figure on how easy something is to read. This paragraph, by way of example, has a reading ease of 63.5 – the higher the number, the easier the read.
Indices are a product of mass information. Armies and bureaucracies love them. Now computers do too. And just as I rely increasingly on Google’s algorithms to search out the stories that fill up my RSS newsreader, I wonder too if I could search smarter. For me that means getting more information and less opinion.
So, what would be the features of an index that rated the news information in reports? The things I look for?
- Facts. I want places, organizations, names, numbers and dates. But I don’t want team-sheets and lists, and I’m not going to count things twice. Expressing these as a percentage won’t be ideal (e.g. the United Nations Security Council is four words long but will count as one ‘fact’).
- Attribution. I want names sitting within five words of the quotes in a story. Could be calculated by each attributed quote, or as a percentage of the reported speech word count. I prefer the latter.
- Talk, or reported speech. I want to hear from people in a story. What’s the ratio of words in quotation marks, to those out? Say 10% minimum, and a third maximum? I don’t want a report to be one long quote.
These would give you some pretty brute-force figures. They would need to be weighted, depending on assessments of the relative value of each component. They wouldn’t measure a lot of the things I also want in reporting, but like food labelling they would provide some indication of what’s inside a story. And just like food, it’s up to you what you consume. Fact, attribution, talk – yes, let’s call it the FAT content.
To road-test these figures I looked at two reports. One was a dispatch from John Simpson in Baghdad, dated 17 March 2007 – a typical blend of observation and opinion.
The other, by contrast, was a press release written in the form of a news story, filed for CentCom by Sgt. 1st Class Kap Kim on 30 March 2007.
Both pieces feature the Haifa Street area of Baghdad.
Simpson’s piece generates these stats:
- Fact ratio: 1:36. Not a bad ratio.
- 27.3% Attribution. Not so good. Although Simpson works in TV and repeats in writing what people told him on camera, there’s no easy way of checking. In fact none of the quotes listed come from the video report that sits by the story. Of course, we trust him, but a brute-force method wouldn’t give him the benefit of the doubt. However, Simpson quotes himself – and that keeps his attribution percentage from being zero.
- 10.1% Talk. Some of that percentage is, as noted above, Simpson quoting himself (probably not a tip for reporters starting out). Still, a macro wouldn’t mind.
The CentCom press release looks like this:
- 1:35 Fact ratio. There’s a lot of detail in here, but it’s about the same unit.
- 89.1% Attribution. The press release generously quotes numerous military types, and gives names and units. But it slips in an anonymous Iraqi who says: “With the security plan working now, we can come out again.” Yes, it’s almost as convincing as “Ambassador, with this Ferrero Rocher, you’re really spoiling us.”
- 29.4% Talk. We hear a lot from the boots on the ground.
The BBC’s world-weary World Affairs ed is short on facts, attribution and quotes, and long on colour and judgement. His Flesch reading level is 69.1. CentCom’s reporter sergeant crams his copy with attributed quotes and military detail. He scores 54.8 on the Flesch reading level.
So do the numbers help? As Bobby Kennedy said of GDP, it measures everything “except that which makes life worthwhile.” Still, if you can figure out a way to automate the numbers, or come up with a more convincing way to analyse copy, I’d like to see the results. In the meantime my RSS reader keeps filling up.