Have statistics lost their power in public policy

Have statistics lost their power in public policy discussions? Jeff Evans j. evans@mdx. ac. uk ALM-24, Rotterdam, 2 -5 July 2017

PLAN • Contemporary crisis of statistics • The statistical approach: historical development • Aspects of the “overt crisis” • The “covert crisis”: “logic of statistics” “logic of data analytics” • Data analytics: Here comes “Big Data” • Issues with Big Data and Data Analytics • Overt & covert crises: Social & political consequences

The “overt crisis” of statistics • Lack of trust in statistics –> declining authority: e. g. survey results: US (68% Trump-ists distrust gov’t econ stats) … UK (55% distrust …“number of immigrants living here”) • Related to “post-truth” politics, “fake news” claims • Lack of generally accepted baselines for discussing competing claims about society resorting to “speaking one’s own truth” … drawing on “intuition” and emotion as alternative bases of knowledge, and of policy “Have statistics lost their power in policy discussions? – and why we should fear what comes next … ” (W. Davies, Guardian, 19 Jan 2017)

The statistical approach: historical development “Statistics” –> data + techniques + disciplines [gov’t, Univs. …] Communities of Practices [representative sampling, stable measurement, causal relns] A. Aim to understand an entire population [17 th C onwards]: Wm Petty & John Graunt est. counted deaths (England), NL? • “Statistics”… not necessarily numbers (various German states) • produced by trained cadres [experts], centralised office (France) B. Use of normal distribution to understand, quantify variation • approximation to gambling outcomes (de Moivre, early 18 th C) • errors of measurement (Gauss, late 18 th? ) • distribution of physical (mental? ) ‘characters’ (Quetelet, Galton)

The statistical approach: historical development C. Use to simplify diverse and complex populations, using specific indicators, clearly defined and systematically produced, e. g. population, Int’l Class’n Diseases (19 th C), ISCED D. Surveys / opinion polls of representative samples of a population/ subgroups, random sampling (20 th C) E. Attempts to maintain comparability across time + sometimes across nations, to support developing explanations e. g. PIAAC F. Experimental design (Randomised Controlled Trials) (20 th C) …. quasi-experimental designs

Aspects of overt crisis of statistics I. Dilemma: need to govern population as a whole vs. (increasingly) to respond to feelings of particular citizens in particular place and time mismatch between what politicians say about the general state of the labour market / unemployment & local / individual experience of it II. Strain on existing classifications and definitions ß more fluid and fragmented identities, attitudes, beliefs (“emotions”) ß reshaping of global economy and society, e. g. gender, social class • work, unemployment … e. g. “zero-hours contracts” • GDP … vs … levels of “well-being” Difficulty of satisfactorily portraying (general) state of the nation

Aspects of overt crisis of statistics III. Difficulty of measuring intensity – e. g. of employment (above) OR …. commitment, e. g. measuring the chance of actually exercising one’s “voting preference” on election day IV. Strains on comparability across time, and especially across nations, e. g. PISA “mathematics literacy”… PIAAC Numeracy: “the ability to access, use, interpret, and communicate mathematical information and ideas, in order to engage in and manage the mathematical demands of a range of situations in adult life”

The “covert crisis”: from a “logic of statistics” to a “logic of data analytics” ONS “experts”– bound by research ethics, monitored by UK Statistics Authority, committed to access to data for researchers) vs. … “Experts” of google, Facebook, Cambridge Analytica (Carole Cadwalladr, Observer, 26 feb 17) – appropriate data (via EULAs) … And sell data + analyse it [using “ data mining” / AI] … And link it with other datasets, e. g. for “sentiment analysis” producing “tailored messaging” – for customers, e. g. marketers, politicians, “opinion formers”, e. g. Vote Leave, Trump campaign … And hoard data without our knowledge, even if we produce it!

Data Analytics: Here comes “Big Data” Very large amounts of data, but also • accumulated by default, as a by-product, usually without “design” (e. g. sampling) • requiring electronic technology for capture, analysis, presentation (? ) E. G. Speed cameras behaviour monitoring Loyalty card purchasesbeh. monitor + demog (etc) correl’ns + experimentation? Electronic texts (soc. media) “AI” (data mining) + data linkage + “sentiment analysis”

Other examples of “Big Data” ? • “Citizen science” • “Citizen maths” observation (stars, bird movements) calculations / simulations / • Cf. Mass Observation (1937 -- ) ? not electronic, panels of volunteers

Issues with Big Data / Data Analytics Data is “Big”, BUT … [sampling, measurement, correlations? ] • “Haphazard” harvesting of data … A huge sample can still be biased (e. g. C. Marsh, 1979) … + if no known sampling design, generalisation to any recognisable population very difficult … • No settled categories self-selected identities (hard to “link”) • Correlation is (still) not causation … Aim to produce immediate fix … not widely applicable theories • No real “informed consent” … Most “Accept” the EULA – clear? “freely chosen”? proximate (to use of data)? time-limited? • No knowledge of what the data says about you

Overt & covert crises: Social & political consequences Without statistics & social research - to make relatively unambiguous, potentially consensus-forming claims about society, and to provide a corrective to faulty claims … • Few mechanisms to prevent people from instinctive reactions and emotional prejudices • Data analytics is “suited to detecting trends, sensing the mood, spotting things bubbling up” (Davies), and can reinforce such trends • BUT numbers are “generated behind our backs and beyond our knowledge”, and appropriated and owned by private concerns • THUS Open Data initiatives is not likely to be mirrored in the sharing of “the benefits of data analytics”.

Overt & covert crises: Social & political consequences • SO … battle is not between elite-led “politics of facts” versus a populist “politics of feeling” • RATHER … between those committed to public knowledge and argument versus those who profit from an appropriation and privatisation of information and an ongoing disintegration of public knowledge and argument

What can be done? A few modest suggestions 1. Read the EULA before you “click” and … 1 a. Cultivate a “healthy skepticism” of IT’s Fearsome Five: Apple, #2 ? ? … #3 ? … #4? … #5? … 1 b. … and ditch? E. g. Google search Duck. Go 2. Get your news, not from facebook, but from … independent (professional) journalists 2 a. Support e. g. Guardian membership, Financial Times Headlines 3. Support fact-checkers: e. g. Full Fact in UK Radical Statistics

What can be done? 4. Research topics 4 a. Which feelings are crucial here? • Fear (anxiety) / love • Trust / distrust • Anger (Pankaj Mishra, The Age of Anger, 2017)