In John D Cook’s blog post (http://www.johndcook.com/blog/2010/12/15/big-data-is-not-enough/) he quotes Bradly Efron in an article from Significance. It is somewhat counter-culture (or at least thought-provoking) to the mainstream ‘Big Data’ mantra – Given enough data, you can figure it out. Here is the quote, with John D. Cook’s emphasis added:
“In some ways I think that scientists have misled themselves into thinking that if you collect enormous amounts of data you are bound to get the right answer. You are not bound to get the right answer unless you are enormously smart. You can narrow down your questions; but enormous data sets often consist of enormous numbers of small sets of data, none of which by themselves are enough to solve the thing you are interested in, and they fit together in some complicated way.”
What struck a chord with me (a data guy) was the statement ‘and they fit together in some complicated way’. Every time we examine a data set, there are all kinds of hidden nuances that are embedded in the content, or (more often) in the metadata. Things like:
- ‘Is this everything, or just a sample?’ – If it is a sample, then how was the sample created? Does it represent a random sample, or a time-series sample?
- ‘Are there any cases where there are missing cases from this data set?’ – Oh, the website only logs successful transactions, if it wasn’t successful, it was discarded.
- ‘Are there any procedural biases?’ – When the customer didn’t give us their loyalty card, all of the clerks just swiped their own to give them the discount.
- ‘Is there some data that was not provided due to privacy issues?’ – Oh, that extract has their birthday blanked out.
- ‘How do you know that the data you received is what was sent to you?’ – We figured out the issue – when Jimmy saved the file, he opened it up and browsed through the data before loading. It turns out his cat walked on the keyboard and changed some of the data.
- ‘How do you know that you are interpreting the content properly?’ – Hmm.. this column has a bunch of ‘M and F’s.. That must mean Male and Female. (Or, have you just changed the gender of all the data because you mistakenly translated ‘M-Mother and F-Father’?)
All of this is even more complicated once you start integrating data sets, and this is what Bradly Efron was getting at. All of these nuances are exacerbated when you start trying to marry data sets from different places. How do you reconcile two different sets of product codes which have their own procedural biases, but essentially report on the same things?
Full article here: