Data journalism training – some reflections

OpenHeatMap - Percentage increase in fraud crimes in London since 2006_7

I recently spent 2 days teaching the basics of data journalism to trainee journalists on a broadsheet newspaper. It’s a pretty intensive course that follows a path I’ve explored here previously – from finding data and interrogating it to visualizing it and mashing – and I wanted to record the results.

My approach was both practical and conceptual. Conceptually, the trainees need to be able to understand and communicate with people from other disciplines, such as designers putting together an infographic, or programmers, statisticians and researchers.

They need to know what semantic data is, what APIs are, the difference between a database and open data, and what is possible with all of the above.

They need to know what design techniques make a visualisation clear, and the statistical quirks that need to be considered – or looked for.

But they also need to be able to do it.

The importance of editorial drive

The first thing I ask them to do (after a broad introduction) is come up with a journalistic hypothesis they want to test (a process taken from Mark E Hunter’s excellent ebook Story Based Inquiry). My experience is that you learn more about data journalism by tackling a specific problem or question – not just the trainees but, in trying to tackle other people’s problems, me as well.

So one trainee wants to look at the differences between supporters of David and Ed Miliband in that week’s Labour leadership contest. Another wants to look at authorization of armed operations by a police force (the result of an FOI request following up on the Raoul Moat story). A third wants to look at whether ethnic minorities are being laid off more quickly, while others investigate identity fraud, ASBOs and suicides.

Taking those as a starting point, then, I introduce them to some basic computer assisted reporting skills and sources of data. They quickly assemble some relevant datasets – and the context they need to make sense of them.

For the first time I have to use Open Office’s spreadsheet software, which turns out to be not too bad. The data pilot tool is a worthy free alternative to Excel’s pivot tables, allowing journalists to quickly aggregate & interrogate a large dataset.

Formulae like concatenate and ISNA turn out to be particularly useful in cleaning up data or making it compatible with similar datasets.

The ‘Text to columns’ function comes in handy in breaking up full names into title, forename and surname (or addresses into constituent parts), while find and replace helped in removing redundant information.

It’s not long before the journalists raise statistical issues – which is reassuring. The trainee looking into ethnic minority unemployment, for example, finds some large increases – but the numbers in those ethnicities are so small as to undermine the significance.

Scraping the surface of statistics

Still, I put them through an afternoon of statistical training. Notably, not one of them has studied a maths or science-related degree. History, English and Law dominate – and their educational history is pretty uniform. At a time when newsrooms need diversity to adapt to change, this is a little worrying.

But they can tell a mean from a mode, and deal well with percentages, which means we can move on quickly to standard deviations, distribution, statistical significance and regression analysis.

Even so, I feel like we’ve barely scraped the surface – and that there should be ways to make this more relevant in actively finding stories. (Indeed, a fortnight later I come across a great example of using Benford’s law to highlight problems with police reporting of drug-related murder)

One thing I do is ask one trainee to toss a coin 30 times and the others to place bets on the largest number of heads to fall in a row. Most plump for around 4 – but the longest run is 8 heads in a row.

The point I’m making is regarding small sample sizes and clusters. (With eerie coincidence, one of them has a map of Bridgend on her screen, which made the news after a cluster of suicides).

That’s about as engaging as this section got – so if you’ve any ideas for bringing statistical subjects to life and making them relevant to journalists, particularly as a practical tool for spotting stories, I’m all ears.

Visualisation – bringing data to life, quickly

Day 2 is rather more satisfying, as – after an overview of various chart types and their strengths and limitations – the trainees turn their hands to visualization tools – Many Eyes, Wordle, Tableau Public, Open Heat Map, and Mapalist.

Suddenly the data from the previous day comes to life. Fraud crime in London boroughs is shown on a handy heat map. A pie chart, and then bar chart, shows the breakdown of Labour leadership voters; and line graphs bring out new possible leads in suicide data (female suicide rates barely change in 5 years, while male rates fluctuate more).

It turns out that Mapalist – normally used for plotting points on Google Maps from a Google spreadsheet – now also does heat maps based on the density of occurrences. ManyEyes has also added mapping visualizations to its toolkit.

Looking through my Delicious bookmarks I rediscover a postcodes API with a hackable URL to generate CSV or XML files with the lat/long, ward and other data from any postcode (also useful on this front is Matthew Somerville’s project MaPit).

Still a print culture

Notably, the trainees bring up the dominance of print culture. “I can see how this works well online,” says one, “but our newsroom will want to see a print story.”

One of the effects of convergence on news production is that a tool traditionally left to designers after the journalist has finished their role in the production line is now used by the journalist as part of their newsgathering role – visualizing data to see the story within it, and possibly publishing that online to involve users in that process too.

A print news story – in this instance – may result from the visualization process, rather than the other way around.

More broadly, it’s another symptom of how news production is moving from a linear process involving division of labour to a flatter, more overlapping organization of processes and roles – which involves people outside of the organization as well as those within.

Mashups

The final session covers mashups. This is an opportunity to explore the broader possibilities of the technology, how APIs and semantic data fit in, and some basic tools and tutorials.

Clearly, a well-produced mashup requires more than half a day and a broader skillset than exists in journalists alone. But by using tools like Mapalist the trainees have actually already created a mashup. Again, like visualization, there is a sliding scale between quick and rough approaches to find stories and communicate them – and larger efforts that require a bigger investment of time and skill.

As the trainees are already engrossed in their own projects, I don’t distract them too much from that course.

You can see what some of the trainees produced at the links below:

Matt Holehouse:

Many Eyes _ Rate of deaths in industrial accidents in the EU (per 100k)

Rate of deaths in industrial accidents in the EU (per 100k)

Raf Sanchez:

Rosie Ensor

  • Places with the highest rates for ASBOs

Sarah Rainey

Advertisements

10 thoughts on “Data journalism training – some reflections

  1. Pingback: links for 2010-12-02 « Sarah Hartley

  2. Martin Griffiths

    Really interesting, thanks. Statistically speaking I think you were a bit lucky with your coin toss experiment. On average, you’d expect 4 or 5 as the longest sequence of heads in 30 tosses.

    Reply
      1. Martin Griffiths

        Framing it as longest sequence of Heads OR Tails would boost it to 5 or 6 expected.

        Another good classroom stats experiment is on regression to the mean. Get everyone to roll a die, with their roll representing the number of accidents at a certain location. Then give everyone who rolled a 5 or 6 a ‘speed camera’ marker, to represent installing cameras only at the worst locations. Work out the average number of accidents in the ‘speed camera’ locations and in the other locations. Then get everyone to roll again, work out the average in the two groups, and calculate the change.

        You should see that the ‘speed camera’ group has a bigger reduction in accidents than the other group, even though it’s pure chance.

  3. Pingback: links for 2010-12-06 « Science Training for Journalists

  4. Pingback: links for 2010-12-07 « Sarah Booker

  5. Keneilwe Eileen Segopolo

    Hello i have been looking at your profile and some of your work. I am part of a team of editors who have just started a portal for services.Its tough but inspiring and we need training. Do reply and keep the good work.

    Reply
  6. Keneilwe Eileen Segopolo

    The portal i reffered to uses Episerver CMS 5. There is a debate on whether its sustainable to use close source mangement system.For some of our colleagues it seems expensive but at least we dont have to worry about security features.Whats your take on this?

    Reply
    1. Paul Bradshaw Post author

      I think both open and closed source have their advantages and disadvantages, and wouldn’t presume to write off either.

      Reply

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s