Wednesday, October 18, 2017

Data Standards are Needed

There's an old saying that figures don't lie, but liars figure. Whether one agrees specifically with this saying, the fact is it draws out an important issue about numbers. They can bury many subjective elements under the appearance of objectivity. It's an old problem with statistics too. The same statistics can be used, for example, for different purposes in the political arena. An unemployment rate of 8% can be considered bad from the view point of the number of people out of a job, but good if it was 10% for all of the preceding year. Context is everything.

Cathy O'Neil made a similar point in her book "Weapons of Math Destruction". She pointed out that not only is data subject to mis-representation of the facts, but the algorithms used to analyze them can similarly embody biases in point of view that the analyst may not even be aware of. There are lots of biases available in our world.

She goes on to say that we should address this issue with development of proper standards in presenting data and the related algorithms. This may need to be approached on an industry basis. We have long had standards in presenting traditional financial information (such as balance sheets and income statements) but not so much when we talk about financial information presented in non-traditional format. That's why we need something like eXtensible Business Reporting Language.

This issue with data standards must be addressed when we talk about using data, perhaps particularly big data, for decision-making purposes. The use of data without standards can be very misleading.