The speaker wants to make clear from the start that he is a mathematical statistician---i.e., a mathematician who studies and developes statistical theory---with certainly a great (layman's) interest in environmental matters, but with very little actual professional practical experience in this big and important field (though with plenty of professional practical experience in other fields of application of statistics).
The body of this lecture is therefore a combination of anecdote and gut-feeling - not very scientific I'm afraid. But that's how it has to be and the organisers knew this when they asked me to speak.
My talk will have something to do with the environment but not much to do with water, sorry.
The title of the lecture is taken of course from Disraeli's famous statement corresponging closely to the modern `you can prove whatever you like with statistics'.
A different quote, similarly antagonistic to statistics, would be Rutherford's `if you need statistics, you did the wrong experiment'.
Perhaps one should counter Disraeli's words by noting that they were [?] a response to Florence Nightingale's efforts to improve the lot of soldiers during the crimean war through introduction of elementary hygiene; she pioneered statistics to make her point; a point which was unpopular since it appeared to be going to cost the taxpayer money but save lives. [check history].
Countering lord Rutherford, one should say that in environmental matters there are not that many experiments you can do. [One could say that human emission of greenhouse gasses is a global experiment with a sample size of one and where the experimental design does not satisfy the most elementary requirements needed to avoid confounding of the experimental effect of human intervention with the intrinsic variability in time of the earth's climate. Let us hope we survive our experiment; we may perhaps take some comfort from the fact that surely Nature will survive in some form or other].
This means that statistics is more than elsewhere NEEDED and that it is more than elsewhere DIFFICULT.
Statistics however runs counter to the grain of many people, as the above quotes illustrate.
The man in the street distrusts statistics and despises [his image of] statisticians, those who diligently collect irrelevant facts and figures and use them to manipulate society.
While GOOD statistical thinking is not that much more than good common sense, the man in the street is incapable of grasping basic (modern, frequentist) statistical ideas of significance level and statistical power; in fact he is a Bayesian without knowing it (as indeed maybe everyone should be in their personal decision making). [explain Bayesian versus frequentist paradigms]. Statisticians are involved in an everlasting internal foundations debate which further decreases their authority.
This issue is coming alive again through the recent emergence of powerful new techniques for Bayesian statistics.
Formerly the frequentists at least had on their side that they could analyse relatively complex models, whereas Bayesians only had solutions in the most simplistic of situations and under severe restrictions on the choice of prior distribution.
Now with Monte Marlo Markov Chain the tables are completely reversed, apparently; I say apparently because no one knows whether an MCMC algorithm has come close enough to convergence that the results are what they are taken for, and no-one has feeling for the implications of almost infinite dimensional prior distributions which may appear flat (representing no prior information) but actually imply strong beliefs in certain aspects of the parameters.
Yet more complications comes from the rise of various even more `soft' versions of statistical analysis, e.g., pseudo-scientific ways of mathematically balancing opposing expert opinions, fuzzy logic, and so on and so forth; and from the decelopment of pseudo-statistical methodology in areas like computer science (data mining, neural networks...) where whole diciplines are reinventing the wheel, and probably making some of the same mistakes as our forefathers did, through lack of appreciation of what alreay exists in statistical science; admittedly they may also be making some original contributions through their fresh look at the age old problems of how to make sense of uncertain data and how to find the right balance between bias and variance.
It seems to me that there are very big dangers here as sociey becomes ever more complex, and politicians and civil servants ever more a separate clan, that decision making in such areas as the environment where there IS scientific evidence to be taken account of, and there ARE scientists with relevant knowledge, but where there are powerful lobbies on all sides driven by very basic fears and emotions and desires, becomes more influenced by the possibilities of manipulating public opinion which modern technology also offers us, and bolstered by all kinds of pseudo-science which can be used to add an aura of expert respectability to the policies of the powerful lobbies in our society, mainly driven by short term economic gain rather than the long term well-being of society.
Clearly statisticians must also share some of the blame and I will return later to an analysis of what went wrong.
First let me bolster some of the above opinions by some stories, purely anecdotal I admit, and my main story (number 4, on batteries) is again a (probably somewhat biased) sample of size 1.
Still I think we have something to learn from it, and also we can learn from the debate which is made possible when someone is prepared (temporarily at least) to take an extreme view.
Some stories
1. The height of the dikes - what does sociey mean when it wants the probability of a disaster to be less than 1 in a hundred per hundred years?
Frequentist versus Baysian solutions?
Should we merge different kinds of uncertainty or not?
2. UK greatest export is privatisation-expertise: i.e. expertise in public relations.
Netherlands has best environmental-statistics in the world.
Big amounts of resources at central levels (the Hague, committees, task groups, )... cf. `Natuurmonumenten' crusade against alien trees! i.e. some tendencies in Dutch society... Positive side---VERY professional and important work done at RIVM, CBS, Universities, ...
3. Asthma, smog warnings, smog danger levels.
The critical levels were changed in order not to excite public unrest. In other countries (Paris, London) they take steps to decrease use of car transport in big cities!
4. Most recent personal experience: recycling of batteries.
Thanks to Erik van Zwet, for the statistical analysis and joint modelling and understanding effor. Big issue at RIVM, `stuurgroepen' and lobbiers in the Hague. The standard question: `How large should the sample be' (in order to prove that the Dutch households are recycling the industry-agreed minimum amount of batteries, hence no deposit system on batteries needed). Presently annual sample of (`official') household waste from more than 1000 households, representatively selected according to a modern and standard method in use by all professional marketing-bureaus. Conclusions come along with a statistical analysis using student-t and trimming of extremes.
At a closer look we discover that the RIVM department actually collecting and analysing the waste is severely underfunded and finds it very difficult even to give us access to the data; the presently used statistical method is completely inappropriate; the data is actually aggregated across households within a particular socio-economic stratum and the same sample of households are used in subsequent years.
Hence it is COMPLETELY IMPOSSIBLE to separate the between household variation from the variation across strata, and COMPLETELY IMPOSSIBLE to separate the within household random temporal variation from true time trends.
Only under strong (obviously untrue, but leading to an illustrative analysis) can we disentangle the various random and systematic effects and offer advice as to how large a sample should have been, of data collected at the household level (very much more expensive than filling a whole refuse van with the waste from 50 bins in one street) in order to get an answer, under various more or less plausible assumptions on the underlying variability in battery-refuse-behaviour.
Reports available from authors.
Conclusions
Where are the statisticians?
Some remarks on different statistical cultures and different status of statistical science in various countries; in Europe, roughly Britain, Scandinavia, France versus Netherlands, Germany; where are Italy and Spain? In UK statistics carried out at relatively high professional level and relatively visible in society cf. Sir Ronald Fisher, Sir David Cox,...! (typical British phenomenon but still it makes a point). Statistics came late to continent, and developed in shadow of pure mathematics; unhealthy split between pure and applied; slowly weakening in NL but in my opinion has done big damage in past.
What to do about it?
My suggestion: change the name (cf. Windscale, Cellarfield, ...). Statistics -> Knowledge Science!
Am I being serious? Yes and no. Anyway, this will be argued in detail elsewhere.