I have set up a new website at www.robertgrantstats.co.uk which will be a central point linking to this blog and the sites on animated graphs and interactive graphs. It also provides learning resources like online lectures. At present it is a copy of the old university site but will undergo a makeover to a new layout and CSS template in the next week or so. As the lectures go up I’ll post about them here.
Monthly Archives: August 2013
Here’s an exciting new visualization from Nathan Yau at FlowingData. You can read the detail here, but essentially he set up a program to query Google Maps across a 20-mile grid and find the nearest “grocery store”. The distances are the red lines and in some zoomed-in versions there are dots for the stores.
It’s a bit like the fairly common approach to mapping wind direction and strength, which in a sense goes all the way back to hand-drawn synoptic charts. I have never got to grips with automated querying from Google Maps (on the basis that the results are not entirely trustworthy) but I guess it’s not that hard to loop through the co-ordinates on the grid, searching for “grocery stores near 34.813839,-87.029149” (or whatever), select the first result, find and save its co-ordinates. It gets a bit more laborious because Google’s result A is not the nearest, it’s a combination of location, popularity, relevance… but I guess you could query them all.
Then the visualization can be a leaflet.js combination of markers and lines, or some other interface from R back to Google.
What would be really interesting as a next step would be changing the geographical distances to transport time when zooming into urban settings. You then have two versions: one if you own a car and one if you don’t. I wonder if Google Maps would provide that easily? I know it’s there on the screen but might not be so easy for the computer to get it several thousand times…
Somehow I have neglected to post this before now, but I am giving a talk at the London SUG at Cass Business School on 12-13 September 2013. The topic is becoming something of a staple: multiple imputation for residual confounding (coarsened covariates). Come one, coma all*.
* – this was a typo but I think I’ll leave it!
Ahead of the LondonR talk on 10 September on sonification, I am giving an even shorter 3-minute ‘trailer’ for my life and work with data and sound. It’s at the Barbican centre in London as part of Hack The Barbican, 21 August, 6-9 pm. I’m hoping to learn a lot from an interdisciplinary art + science gathering.
I’ll explain what motivates me, show someone else’s sonification work (not decided what yet), and then an example of my own which is a type of sonic graph never heard before, I believe…
For bureaucratic reasons you don’t wanna know, my university URL is changing. That means the tinyurl.com link won’t work any more (you can’t change them; yes, this did worry me a tinybit at the time…)
A new copy will be back before long. I’m tempted to move it entirely to a more permanent site that is entirely under my own control, but y’know, I’d have to pay for that… I’ll think about it over the weekend.
A follow-up to my post on Andrew Gelman’s encounter with social psychology and the color of women’s clothes: a short editorial that is really worth a read is “Promoting healthy skepticism in the news: helping journalists get it right” from the Journal of the National Cancer Institute in December 2009.
This tells the story of a Phase I trial of a new cancer drug (in other words, they just wanted to see that it didn’t harm humans, or make the condition worse, so they gave it to 19 people whose cancers had not responded to previous treatments), olaparib. Twelve of those 19 people didn’t get any worse for four months, Now, that’s a valuable first step, but it doesn’t really tell you why they didn’t deteriorate, whether the effect would be sustained longer term, whether the drug works for a particular subgroup of people. All of that was years away in the Phase II and III trials, and then if all goes well, the post-marketing surveillance. Yet the study hit the news in a big way, with NBC opening the TV coverage with “some are calling this the most important cancer breakthrough of the decade”. Unhelpful!
They also highlight the case of the alcohol and cancer paper that came out of the Million Women study in March 2009, also in JCNI. I use this paper for teaching quite a lot. It’s a nice example of a good quality cohort study that went a little wrong at the last minute. Check out Table 2, where the non-drinkers have higher risks than the 1-2 drinks a week crowd. Now look at Figure 3. Whoa, where did those non-drinkers go? Maybe the graph just didn’t look so good without a nice straight line! (I am indebted to Doug Altman for first criticising this graph.)
And on the basis of this straight line, the printing presses rolled and rolled. The invited editorial got a bit over-excited about it, coining the quote ” there is no level of alcohol consumption that can be considered safe”. The PR department picked up on that, then the TV and the radio stations and the newspapers (viz BBC, I’m not even going to speculate on the alarming quote from the first author in this). Even august medical campaigners on alcohol jumped on the bandwagon too. Messy.
Back to the JCNI December 2009 editorial. Well done to them for tackling this important topic in the year when their own journal had fallen victim to one of these health news feeding frenzies. They suggest that a range of handy guides to accurate and fair reporting will help journalists improve their act, but I find this hard to believe. In business, even if you think what you are doing is wrong or not a great long-term brand-building choice, if the competitor is going to steal a march on you with an alarming, attention-grabbing headline, you’ve got to print it too. In fact, you’ve got to print it first (no time for that handy guide) and it’s got to be even more alarming!
The only way to tackle this is at the PR source. Journalists for the most part (until a story gets hot) simply copy out of the press release. Get that right and everything else will follow. And best of all, we are actually in a position where we can influence this. Without patronising them, set up a free stats workshop for your university / organisation’s PR and comms people. They can stick it on their CVs. Make yourself available for stats queries when they are drafting stuff. Build out from there to a sustainable posse of stats people who are happy to answer marketing / PR / comms queries. And accept that you are not going to appear on the TV news (followed by years of opprobrium from the likes of me).
On the back of my Significance “a life in stats” article about Nathan Yau, Radical Statistics (a group, of whom I am pleased to be a member, interested in using statistics to promote progressive social and political change, and resisting the abuse of statistics by those in positions of power) asked me to write something expanding on these ideas for their magazine. So there is an article in the new issue number 109, which will soon be on their website.