Stats Chat

Explore, share and soon create infographics and data visualisations at the brand new website, Visual.ly

The United Nations Economic Commission for Europe has been releasing a series of guides about communicating with statistics and part 3 is now out:

Communicating with the media (PDF) – a practical tool for producers of statistics find the best way to get their message across and to communicate effectively with the media.

The Making Data Meaningful guides are intended as a practical tool to help managers, statisticians and media relations officers in statistical organisations use text, tables, charts, maps and other devices to bring statistics to life for non-statisticians.

Earlier editions are below:

• Part 1: A guide to writing stories about numbers (PDF)
• Part 2: A guide to presenting statistics (PDF)

The thorough guides are excellent practical primers and the latest one features quite a bit of social media applications.

Thanks to our colleague Peter Davis for pointing out this resource set.

Charlie Park has a great post on Edward Tufte’s “slopegraphs” where he explains what they are, how they’re made, why they haven’t been used too much to date and why they may become more popular very soon.

Great interview on Radio NZ this morning, with Eli Pariser author of ‘The Filter Bubble: What the Internet is Hiding from You’.

Content as well as advertising is now targeted to users. So internet users receive information that reflect their own opinions and interests and as Pariser comments “we don’t see how our experience of the web is different from anyone else’s”. By making our searches personally relevant and individually likeable (“if you like this fact you’ll like that fact”) we don’t get exposed to different perspectives. And we have not had the opportunity to ‘opt out’ of this search personalisation.

Online personalisation started for Google Dec 4 2009. Evidently 57 data points are used including such variables as your past searches, technology and font you use, where you are located, your speed of clicking through. Pariser says, “the watch word in Silicon Valley these days is relevant”.

Mark Zuckerburg of Facebook has been quoted “a squirrel dying in your front yard may be more relevant to you right now than people dying in Africa.” Pariser comments, “I worry about what a web and what a news ecosystem that is built on that idea of relevance might look like”. Presentation by Pariser on this is also here.

The discussion on the benefit of the editorial role in main stream media with the premise of the civic responsibility and ethics, compared to the algorithmic feed of personalised news on the web is interesting food for thought given the current News of the World story.

Once upon a time, in a distant town in a far away land, there was a regular ballroom dancing event every weekend, an Argentine Tango Spectacular.  Everyone in the town was invited, and since there wasn’t much else to do, nearly everyone turned up.    Now, in tango, as you know, you dance in couples, one man and one woman.  As the night wore on, some people danced over and over again with the same  partner.  Some did every dance with a different partner.  Some mostly just sat out and watched, and socialised. Some boycotted the whole event.

The King decided to commission a survey in all the towns in the country, asking people how many times they had danced tango in the past year and how many different people they had danced with.  In our town, the results came back showing that women had an average of 20.3 dancing partners and men had only 16.8. In most other towns, the survey had found that men danced with more different partners than women.

What can we conclude from this survey?  What does it tell us about who was calling the shots in dancing decisions? Does it mean that women liked more variety in their dancing partners?

Of course, we can’t conclude anything.  Every time a woman dances with a new man, that man is dancing with a new woman.  If the number of men and women in the town is close to equal, the average number of dance partners for men and women must be the same as a matter of simple arithmetic. All we learn about is the inadequacies of the survey.  Of course, this is just a story, and any resemblances to real statistical errors are almost entirely accidental.

The anti-smoking drug varenicline (brand name Champix in NZ, Chantrix in US) is in the news because of risks of heart disease.  Recent summaries of research suggest that varenicline increases the risk of serious cardiovascular disease by about 70%. “Serious”, here, is a grab-bag of diagnoses, but they are all things that at best will put you in hospital and at worst are fatal.  Smokers are already at higher risk of cardiovascular disease, so this sounds bad.  On the other hand, varenicline is supposed to help stop smoking, and that will decrease the risk of cardiovascular disease.

How do the risks balance out? Well, that depends on a lot of details, which is why MEDSAFE doesn’t want to commit itself until they’ve had a chance to look at all the evidence.  Still, smoking increases the risk of serious cardiovascular disease by a factor of three to six, and nearly all this increased risk goes away rapidly if you stop smoking.  Varenicline increases the chance of successfully stopping smoking by about 10 percentage points, so we can run some examples with plausible numbers.

Suppose you have 1000 smokers with pretty high risk, who have a 4% chance of a heart attack or death in any given year.  In the six months trying to quit you would expect 100×0.04/2=20 heart attacks without varencline, or 35 with the drug, an excess of 15.  Now, for 900 of these people the drug didn’t have any impact on quitting — they either would have succeeded anyway or would have failed anyway.  That leaves 100 people who were helped to quit, which will save about 3 heart attacks a year for the rest of their lives.   The break-even point is about five years later, and that’s ignoring the other health benefits of quitting.   The basic calculation works the same way for lower-risk smokers — the benefit is smaller, but so is the risk.  Varenicline looks pretty good now.

There are two complications. The first is that there is a lot of uncertainty in both the benefits and risks.  If varenicline actually increases risk 150% and only gets 5% more people to quit, the break-even time is more like 20 years, and the benefits look more dubious. That’s especially true when you consider the other side-effects.  It’s been known for years that varenicline is no fun at all to take, and in particular there are accumulating reports of serious depression.

As usual, PHARMAC seems to have a sensible policy on the drug — the government will subsidise it, but only if you’ve unsuccessfully tried simpler and safer approaches to quitting, like patches or nicotine gum.

Brain Pickings has compiled a fantastic list of 7 books on data visualisation from the speakers at the eyeo festival.

Once you’ve explored those, jump over to eyeo‘s website to be linked to some incredible people in the data visualisation field, including:

Moritz Stefaner who worked on the OECD Better life project and Google’s Think Quarterly – both previously blogged about here on Stats Chat; Ben Fry of Fathom, Adam Bly of Visualizing, Amanda Cox who creates beautiful graphics for the New York Times, and Nicholas Felton. All their sites are well worth a visit.

eyeo was partly sponsored by Visual.ly which launches this week and promises to showcase more than 2,000 visualisations and enable people to create their own web-based infographics and visualisations.

And, for more fascinating visualisations, visit Data Insight – a recently held data visualisation competition.

Bobby Willcox, one of our PhD students at the University of Auckland, Department of Statistics was featured in the NZ Herald yesterday.

Bobby’s major area of interest is statistical research in netball. She completed her MSc in Statistics in 2004, and has been involved with Netball New Zealand and the Silver Ferns netball team ever since. She is currently the Performance Analyst for the Silver Ferns and is in the process of completing her PhD, started in 2007. Her research is focused around finding a way to objectively measure and evaluate player performance in netball and identifying different strategies adopted by players within each position on court.

Read more about Bobby’s statistical tools developed for the Silver Ferns.

Gareth Lloyd, a British software engineer, has put together a animation of every historical event in Wikipedia. Watch for the Taupo eruption just before 200CE — it’s a long time until New Zealand next shows up.

From 3 NEWS, last week: Mr Thompson went on radio this morning to debate gender inequality – it was largely uncontroversial until he tried to explain why women are paid, on average, 12 percent less than men.”Look at who takes the most sick leave,” said Mr Thompson. “Because you know, once a month they have sick problem. Not all women, but some do. They have children and they have to take leave off.”

In New Zealand there does not seem to be any comprehensive information on sick leave differences for men and women, but other countries collect this information, including Sweden, the UK, and Australia, and NZ information is available for public employees.

• The Public Services Association (the government employees’ union) obtained data under the Official Information Act. The figures based on 2010 data in the State Services Commission’s Human Resources Capability Study show that in 2010 men took on average 6.8 days sick and domestic leave annually while women took 8.4 days.
• In the UK, the 2004 Labour Force Survey showed that about 1.4% of men and 2.1% of women were taking sick leave on any given day. The gap between men and women showed up at all ages, but was largest for 25-34 year olds.  The survey also asked about children: the gap between men and women was smallest when there was a child under 5 in the household, and largest when the youngest child was 5-10.  Single parents, male or female, had the highest rates of sick-leave absence, with nearly 5% missing work for at least one day during the week they were sampled.  A less-detailed report from 2010 shows about the same difference between men and women
• The last Australia Bureau of Statistics report on sick-leave differences between men and women dates from 1995.  Over a two-week period, 9.3% of men and 10.3% of women took at least three hours’ sick-leave. As with the UK data, the difference was larger for people under 34. In the over-55 age group, men were slightly more likely than women to take sick leave.
• In Sweden, quarterly data are available right up to 2010. The number of sick days taken per quarter is about 1.5 for women and about 1.0 for men (more in winter, less in summer).
• A (highly technical) report from the US National Bureau of Economic Research looks at whether there are periodic cycles in women’s sick leave, and finds that there is no real evidence. Differences between men and women do appear to have a weak 28-day periodicity, but so do differences between older and younger men, showing that it isn’t PMS that’s the explanation.

Although the data are not as complete as we might like, there is a consistent picture across the world where data is available.  On average, women do take more sick days than men, but it’s not a big deal. Two or three days a year seems to be a reasonable estimate of the difference, and there’s no way that explains any meaningful fraction of a 12% difference in pay.   It’s also worth noting that the pay gap between men and women varies enormously around the world — it’s almost twice as large in the UK as in NZ — but the sick-leave differences are much the same.