Thursday, November 07, 2013

When Big Data goes bad: 6 epic fails

Data, in the wrong hands, whether malicious, manipulative or naïve can be downright dangerous. Indeed, when big data goes bad it can be lethal. Unfortunately the learning game is no stranger to both the abuse of data. Here’s six examples showing seven species of ‘bad data’.
1. Data subtraction: Ken Robinson
Don’t let the selective graphical representation of data, destroy the integrity of the data. A good example of blatant data editing is the memorable ‘ritalin’ image used by Sir Ken Robinson in his TED talk at 3.47. This image is taken from its RSA animation.
Compare Robinson’s graph with the true source.
His has no legend and he’s recalibrated states to look as if there’s zero prescriptions. To understand this data you have to look at its source to understand that the white areas represent states that did NOT participate in the study or did not have reported prescription data. It’s a distortion, an exaggeration to help make a point that the data doesn’t really support
In fact, much of what passes for fact in Sir Ken Robinson’s TED talks are not supported by any research or data whatsoever.
2. Data addition: Bogus learning theory
Ever seen this graph, or one like it? It used to be a staple in education, training and teacher training courses. Only one problem - it’s bogus.
A quick glance is enough to be suspicious. Any study that produces a series of results bang on units of ten would seem highly suspicious to someone with the most basic knowledge of statistics.
But it’s worse than nonsense, the lead author of the cited study, Dr. Chi of the University of Pittsburgh, a leading expert on ‘expertise’, when contacted by Will Thalheimer, who uncovered the deception, said, "I don't recognize this graph at all. So the citation is definitely wrong; since it's not my graph." What’s worse is that this image and variations of the data have been circulating in thousands of PowerPoints, articles and books since the 60s.
Further investigations of these graphs by Kinnamon ((2002) in Personal communication, October 25) found dozens of references to these numbers in reports and promotional material. Michael Molenda ( (2003) Personal communications, February and March) did a similar job. Their investigations found that the percentages have even been modified to suit the presenter’s needs. 
The one here is from Bersin (recently bought by Deloitte). Categories have even been added to make a point (e.g. that teaching is the most effective method of learning).
The root of the problem is an image by Edgar Dale’s depiction of types of learning from the abstract to the concrete. He has no numbers on his ‘cone of experience’ and regarded it as a visual metaphor implying no hierarchy at all.
Serious looking histograms can look scientific, especially when supported by bogus academic research. They create the illusion of good data. This is one of the most famous examples of not ’Big’ but ‘Bad’ data in the history of learning.
3. Claims beyond the data – University League Tables
University league tables are used by politicians, Universities, parents and students. But they contain a dark, dirty, data secret. They claim to rank universities but, astonishingly, tell you absolutely nothing about ‘teaching’. They often claim to have ‘measures’ on teaching, but they actually draw their data from proxies, such as employment and research activity and use nothing but indirect measures to measure teaching.
The Times rankings are a case in point. They claim that their ranking scores include teaching. In fact, only 30% is based on teaching but they use NO direct metrics. The proxies include student/staff ratios (which is skewed by how much research is done) and, even more absurdly, the ratio of PhDs to BAs. It is therefore a self-fulfilling table, where the elite Universities are bound to rise to the top. There is no direct measurement of face to face tome, lecture attendance or student satisfaction.
4. Skewed data - PISA
Like the real Leaning Tower of PISA, the OECD PISA results are built on flimsy foundations and are seriously skewed. Nevertheless, they have become a major international attraction for educators, and regularly spark off annual educational ‘international arms’ races.
Both left and right now use the ‘sputnik’ myth, translated into the ‘Chinese competitiveness’ myth, to chase their own agendas – more state funding or more privatisation. This is a shame, as the last thing we need is yet another dysfunctional , deficit debate in education. Nations have different approaches to education, different demographic and social mixes and different agendas.
The problems in the data are extreme as PISA compares apples and oranges. PISA is seriously flawed because of the huge differences in demographics, socio-economic ranges and linguistic diversity within the tested nations. There are many skews in the data, including the selction of one flagship city (Shanghai) to compare against entire nations. Immigration skews include numbers of immigrants, effect of selective immigration, migration towards English speaking nations, and first-generation language issues. There’s also the issue of taking longer to read irregular languages and selectivity in the curriculum.
Sven de Kreiner Danish statistician says PISA is not reliable at all. In the reading tests 28 questions were supposed to be equally difficult in every country. PISA has failed here as differential item functioning - items with different degrees of difficulty in different countries - are common. In fact he couldn't find any that worked without bias. Items are more difficult in some countries. He used his analysis to show that the UK moves up to 8 or down to 36. PISA assumes that DIF has been eliminated but not one single item is the same across the 56 countries
Politicians and activists distort PISA to meet their own ends. They cherry pick results and recommendations, ignoring the detail. Finland is famously quoted by the right as a high performing PISA country. Yet, it is a small, homogeneous country with no streaming, high levels of vocational education, no substantial class divisions and no private schools. Facts curiously ignored by PISA supporters.
5. Faked data
Eysenck worked with Cyril Burt at the University of London, the man responsible for the introduction of the standardised 11+ exam in the UK, enshrined in the 1944 Butler Education Act, an examination that, incredibly, still exists in parts of the UK. Burt was subsequently discredited for publishing largely in a journal that he himself edited, falsifying, not only the data upon which he based his work, but also co-workers on the research. To be precise, Burt's correlation coefficients on IQs in his twin studies were the same to three decimal places, across articles, despite the fact that  new data had been added twice twice to the sample of twins. Leslie Hearnshaw, Burt’s friend and official biographer, claimed that most of Burt's data after World War II were fraudulent or unreliable.
This is just one of many standardised tests that have become common in education but many believe that tests of this type serve little useful purpose and are unnecessary, even socially divisive. Many argue that standard tests have led to a culture of constant summative testing, which has become a destructive force in education, demotivating and acting as an end-point and filter, rather than a useful mark of success. Narrow academic assessment has become almost an obsession in some countries, fuelled by international pressure from PISA.
6. Dirty data deeds
One example of data gathering in education stands out as truly evil. In 1939, the CEO of IBM, Thomas Watson, flew across the Atlantic to meet Hitler. The meeting resulted in the Nazis leasing the mechanical equivalent of a Learning Management System (LMS). Data was stored as holes in punch cards to record details of people including their skills, race and sexual inclination and used daily throughout the 12 year Reich. . It was a vital piece of apparatus used in the Final Solution, to execute the very categories stored on the apparently innocent cards - Jews, Gypsies, the disabled and homosexuals, as documented in the book IBM and the Holocaust by Edwin Black. They were also use to organise slave labour and trains to the concentration camps.
This is not the first time the state has recoded educational details to keep tabs on potential dissent. It was common in the Stasi infused East Germany. I shared a room at University with someone who became a Stasi spy in the UK and have taken some interest in their methods. Perhaps the most meticulous storing of data ever taken by a state, right down to smells in jars, from clothing and towels placed under interviewees during interrogation. The idea was that dissenters could be found by dogs, when necessary.
I am a fan of Big Data in education, even though it’s really closer to ‘large or little’ sets of data. However, we must be wary of data when it is used to exaggerated claims through addition or subtraction or spearhead prescriptive programmes and extreme testing. I am appalled at the way politicians and educators take up PISA, PIAC and OECD data, with little or no detailed examination of their assumptions or relative values and use it to shape prescriptive policies that do more harm than good. Big Data in the hands of little brains is downright dangerous.


Peter Phillips said...

Good article, Donald.
The misuse of data is of course endemic. It is most obvious in politics, advertising and the Daily Mail, but more worryingly pervades pharmaceutical and food research - ref Ben Goldacre's excellent books.
I would also recommend "Fooled by Randomness" by Naseem Taleb of Black Swan fame.
Even e-learning is not immune, as anyone who has perused a recent flotation document will attest.

Anonymous said...

That last sentence says it all.

Dave Ferguson said...

A small addition to your examples is the typical geographic image used to represent demographic information.

Compare these two depictions of the electoral vote in the 2012 U.S. presidential election.

Alison said...

Perfect timing Don. I am writing an online PGCE and keen to blast some myths about EBT/Big data. I am very sad about the Ken graph though - I thought he was a good egg. Thanks for the content!