I'm always interested in ways of measuring things, particularly if they are unconventional. So I was drawn to this story on ABC News http://www.abc.net.au/news/stories/2008/09/20/2369772.htm where a University of UCLA academic, John Agnew used luminosity from cities and satellite photographs to infer ethnic cleansing between particular dates in Iraq.

I'm not sure how firm his assumptions really are, but it is an interesting technique and it made me wonder what other things unconventional statistical models could lurk out there.

From my perspective statistical methods were invented over the last 100 years to support politically motivated science.

Is it an exact science /a social product and thus raises the important question of - are statistical methods discovered" or invented ?

Clearly there is a tension between statistics as an exact science" defined as"objective, rigorous, culture free, and technique oriented and statistics as a social product which is produced as the outcome of human responses to a wide variety of mostly conflict-laden situations. (See for example the shonky stats published by US intelligence on civilian deaths in IRAQ)

For me the statistical methods developed by social scientists cannot be separated from the social, political and economic forces that motivate the research in the first instance.

Take for instance the inexact collection of data on Indigenous peoples by the ABS (inspired by the 1967 referendum) and how it failed to provide a solid foundation for the development of evidence based policy. And yet each successive government and their statistically trained bureaucrats - will return back to these same statistics without a second thought - and then wonder why policy failed to deliver.

What’s that saying about insanity – “doing the same thing over and over again and expecting a different result?

You're too postmodern for me Rainier. You're suggesting that people look at statistics to confirm their prejudices. Some do, but I do my polls to challenge my prejudices, and they always do. So I change my mind to fit into reality - the definition of sanity.

Anyway, thought some might be interested in the first instalment of analysis of my polling on federal leadership which is at http://whatthepeoplewant.nationalforum.com.au/archives/003393.html. Turnbull is doing better than Nelson I think, but Rudd is still the king.

GrahamY, for this method to work there would have to be a very strong correlation between luminosity and ethnic cleansing. I see trouble arising where other factors are contributing to reduced luminosity of cities. These may be security instability leading people to leave temporarily until security improves or destruction of infrastructure. If the same study were to be done for Indonesia in 2004, the erroneous conclusion would be that ethnic cleansing had occurred.

Equally, a situation where ethnic cleansing did occur and one group took over another groups houses, you might see no change in luminosity. This needs to be ground truthed before drawing too many conclusions.

Rainier, Statistics is the science of data management. It is a well established science with well established methods. It is only ever as good as the methods and analysis. Where those are shown to be robust the conclusions can be accepted, until better information comes along. This is not to be confused with 'statistics' the data collection that ends up in your newspaper. These are sometimes the product of data mining and should always be questioned until you know the methods and analysis are robust.

if you check my post you will see that I am saying these prejudices exist ubiquitously - no matter which way to design to test them. There is no such thing as unbiased polling.

If you poll to check against your own prejudices then you admit this outright. I fail to see how this is being postmodernist :)

Agro, The stats I am referring to are in published academic journals, government reports and policy - not in my local rag.

And yes they are well established methodological approaches – that are historically informed by well established doctrines and beliefs -which enables the creation of replicable information and explanations of the natural and social world.

The scientific myth of objectivity does not disappear simply because scientists believe this to be true. This illusion of objectivity is well known and acknowledged by statisticians themselves.

Interesting stuff. While everybody knows that it's very easy to lie with statistics (much of AGW denialism depends on distorting stats), the issue that Graham, Rainier and Agronomist are talking about is, of course, validity - i.e. do the statistics measure what they purport to measure?

In this case, I agree mostly with Agronomist - the apparent correlation between luminescenve and ethnic cleansing would have to depend on the degree to which other causes of luminescence can be discounted.

Rainier, in saying:

"Take for instance the inexact collection of data on Indigenous peoples by the ABS (inspired by the 1967 referendum) and how it failed to provide a solid foundation for the development of evidence based policy. And yet each successive government and their statistically trained bureaucrats - will return back to these same statistics without a second thought - and then wonder why policy failed to deliver."

highlights an interesting aspect of Australian population statistics.

If I understand Rainier correctly, he has noted inexactitude in the collection of statistical data on indigenous peoples AFTER the 1967 referendum. In matters unrelated to indigenous peoples policy formation, I did note a curious observation made in a submission to the Inquiry into the Conduct of the 2004 Federal Elections. That submission is not able to be reached directly with a link, but it can be reached from the List of Submissions page. See: http://www.aph.gov.au/house/committee/em/elect04/subs.htm . It is submission No. 123, and contains a document titled 'Australia - Aggregate Enrolment Levels 1947 - 1987'.

Pages numbered 19-22 in that document, under two headings, 'Retrospective Alteration of the 1961 Census Count', and 'Disregarding the Constitution?' contain an interesting perspective on population statistics collection around that time.

Interesting, too, the claim in a reference, at page 136 in Year Book Australia 1971, that "Aborigines have been enumerated in ALL censuses of the Commonwealth".

I'm not quite sure how to take what seems to have been the retrospective alteration of the statistics involved. I'd be interested in your view of it, Rainier. It might constitute published evidence for what you have alluded to as "[failure] to provide a solid foundation for the development of evidence based policy".

Returning to a more direct addressing of the topic, although not being a pollster, I have long felt especially in relation to political or electoral opinion polling that intention should be elicited by asking seemingly unrelated questions, the responses to which may be more accurately revealing. This may already be standard practice, for all I know.