Quick Summary of Second and Third Sessions of “Visualisation and Presentation in Statistics”

Kevin McConway ( http://statistics.open.ac.uk/People/k.j.mcconway @kjm2 ): showing off some gratuitous use of numbers to illustrate Guardian stories #ouvpstats
Where do surveys reported in the press come from? ONS, market research companies. PR companies…… #ouvpstats
Get paid to do a (PR?) survey onepoll.com and youngpoll.com #ouvpstats
Not PR commissioned polls, err, maybe, err, hmmm…. http://72point.com/ #ouvpstats
Why are there numbers in the news? PR, Entertainment, eyecandy. Special status of “number facts” #ouvpn
Mary Poovey “A History of the Modern Fact” http://www.press.uchicago.edu/ucp/books/book/chicago/H/bo3614698.html #ouvpn
Need to distinguish between facts, analysis and narrative… #ouvpstats
What’s wrong with PR stats? ’tis the road to cynicism, or looking good rather than communicating well #ouvpstats
So what can we do about it? Statisticians need to engage with the public and work with journalists #ouvpstats
Statisticians’ view of journalists: innumerate, distort and oversimplify, don’t understand quantitative reasoniong, won’t listen #ouvpstats
Journalists’ view of statisticians: illiterate pedantic, boring, focus on ifs and buts, won’t listen #ouvpstats
Journalists work to tight timescales, have a view of “newsworrthiness”, are good storytellers #ouvpstats

Martin Bland ( https://hsciweb.york.ac.uk/research/public/Staff.aspx?ID=129 )
From papers during one issue from 1972 and 2010 Lancet and BMJ, mean population size has gone up 2-3 orders of magniture (tens to thousands+
Description of stats: very cursory, 2010: far more comprehensive statistical method reported. Shift from significance testing to estimation
Move towards evidence-based medicine starting around 1990s (bound to includes statistics)
“Why do we need some large, simple randomized trials?” Yusuf et al. 1984
Move to confidence intervals not p-values Gardner & Altman http://www.bmj.com/content/292/6522/746.abstract
Journals started to introduce systematic requires and statistical referees
Consort guidelines for stats in randomised medical trials http://www.consort-statement.org/
Statisticians should point out where wrong conclusions have been drawn as a results of stats mistakes…

Rosemary Bailey http://www.maths.qmul.ac.uk/~rab/
Problems with box and whisker plots (referred to as box and aerial/antenna plot?), which are now popular in medicine, biology, engineering (not least becuase folk don’t know what the whisker means). Antenna doesn’t take into account variability across conditions. [My naive understanding of these diagrams is that they are trying to say something different? But my knowledge is so hazy I can’t argue for what I do think they describe!]
Hasse diagrams – cords, dyes and constants(?) [I’m a bit lost at this point…]

Michel van de Velden http://www.erim.eur.nl/ERIM/People/Person_Details?p_aff_id=799
Perceptual maps – mutltivariate methods for plotting high-dimensional data
Exploit natural spatial recognition/visual abilities
Examples: Tufte 1983 cleveland and McGill 1987, Wainer 2005
Caption should convey enough info to allow reader in possession of data (and appropriate tools) to recreate the perceptual map
Shape paramter (aspect ratio) – ratio of x scale to y scale. If it can be 1, it should be… (changes aspect ratio of photo of Kate Middleton to make the point about distortion if not 1 when it could/should be…)
If perception of map relies in part on angle of point/line, need to know where the origin is.
Excel charts – hard to explicity set an exact aspect ratio (same with many tools?)
Perceptual maps may require guidance as to how to read a map – e.g. icons http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1572196


Jill Leyland, Vice President, Royal Statistical Society
Lots of folk think UK official statistics are not free of politcal interference, nor do they necessarily trust(?) them, scores very poorly compared to rest of Europe.
National Stats have high integrity and free of political interference. Perception of political interference is one reason why low degree of trust. UKSA (UK Statistics Authority) scrutinises official statistics: “promoting and safeguarding the porduction and publication of statistics that serve the public good”
No politicial interference, but: many key stats produced in depts, UKSA role not fully understood (scrutineer as well as publisher); pre-release access – Ministers can see statistics 24 hrs before they are released (up to 5 days in Scotland and Wales), and suspicion that Ministers may use this time for mischief…
Role of media – UK media are interested in statistics, but “stats are wrong” stories get more covereage than “stats are right”, and journalists often don’t understand statistical issues (as well as tight deadline, no specialist knowledge). BUT official statisticians could do better; ONS website a joke… (though new one due to launch at end of August). Far too little interaction with stats users outside government.
What can be done? Continuing efforts to improve presentation; need to differerntiate between independent national statistics and those produced by departments. Better education for journalists [and statisticians eg ito communications?]; reduction/elimination of pre-release access.

Author: Tony Hirst

I'm a Senior Lecturer at The Open University, with an interest in #opendata policy and practice, as well as general web tinkering...

2 thoughts on “Quick Summary of Second and Third Sessions of “Visualisation and Presentation in Statistics””

Comments are closed.

%d bloggers like this: