Never Mind The Data, Feel The Noise (or, seek the signal, young Jedi)

Everyone in education loves data.

This is the only time it is correct to use the word "Data" in the singular...

Or at least claims to. One sometimes wonders what would happen to the UK education system if a computer virus disabled every Excel spreadsheet overnight — h’mmm, perhaps someone should get in touch with those nice hacker people at Anonymous . . .

However, I digress. I wanted to share a recent epiphany that I’d had about data, particularly educational data. Perhaps it’s not much of an epiphany, but I’ve started so I’ll finish.

It came when I was listening to an interview on the evergreen The Jodcast (a podcast produced by the Jodrell Bank Radio Observatory). Dr Alan Duffy was talking about some of the new technologies that need to be invented in order to run the new massive Square Kilometre Array radio telescope (due to begin observing in 2018):

And then we have to deal with some of the data rates . . . essentially we recreate all of the information that exists on the internet today, and we do that every year without fail, it just keeps pouring off the instrument. And what you’re looking for is the proverbial needle in the haystack . . . how do you pick out the signal that you’re interested in from that amount of data?
The Jodcast, October 2014, 18:00 – 21:00 min approximately [emphasis added]

The realisation that hit me was: it isn’t the data that should be centre stage — it’s the signal that’s contained within that data. And that signal can be as hard to find as the proverbial needle in a haystack, even without data volumes that are multiples of the 2014 Internet.

A simple example from the history of science: Edwin Hubble’s famous graph from 1929 that was one of the first pieces of evidence that we exist in an expanding universe. The data are the difficult and painstaking measurements made by Hubble and his colleague Vesto Slipher that are plotted as small circles on the graph.


The signal is the line of best fit that makes sense of the data by suggesting a possible relationship between the variables. Now, as you can see, not all the points lie on, or even close, to the line of best fit. This is because of noise — random fluctuations that affect any measurement process. Because Hubble and Slipher were pushing the envelope of available technology at the time, their measurements were unavoidably ‘noisy’, but they were still able to extract a signal, and that signal has been both confirmed and honed over the years.

In my experience, when the dread phrase “let’s look at the data” is uttered in education, the “search for a signal” barely extends beyond simplistic numerical comparisons: increase=doubleplus good, decrease=doubledoubleplus ungood.

The way we use currently use data in schools reminds me of SF author William Gibson’s coining of the term cyberspace (way back in the pre-internet 1980s) as the

consensual hallucination experienced daily by billions of legimate operators . . . a graphic representation of data abstracted from the banks of every computer in the human system
— William Gibson, Neuromancer (1984)

In my opinion, almost the whole statistical shebang associated with UK education, from the precipitous data-mountains of the likes of RAISEOnline (TM) to the humblest tracking spreadsheet for a department of one, is actually nothing more than a ‘consensual hallucination’.

The numbers, levels and grades mean something because we say they mean something. And sometimes, it is true, they can tell a story.

Let’s say a student has variable test scores in one subject over a few months: does this tell us something about the child’s actual learning, or about possible inconsistencies in the department’s assessment regime, or about the child’s teachers?

My point is that WE DON’T KNOW without cross referencing other sources of information and using — wait for it — professional judgement.

I believe that the search for a signal should be central to any examination of data, and that this is best done with a human brain through the lens of professional experience. And, given the inevitability of noise and uncertainty in any measurement process, with a generous number of grains of statistical salt.

12 thoughts on “Never Mind The Data, Feel The Noise (or, seek the signal, young Jedi)

  1. ijstock December 23, 2014 / 10:22 pm

    As usual, you’re so very right. But don’t I know all too well from this last term that that is *not* how data is often used…

    Good to read you again.

    • e=mc2andallthat December 27, 2014 / 4:59 pm

      Glad to be back. BTW, are you still working on the book?

  2. educationresearcher December 24, 2014 / 10:28 am

    This is a the best description of statisitics have read in a edu blog.

    So important edu gains ownership of stats as a tool for interperting messy measurment and not treat it as arithmatic.
    Especially with increasing use of averages and inferential statistical techniques

    • e=mc2andallthat December 27, 2014 / 4:54 pm

      Thank you very much for the comment. Stats are so important in education and yet there is often quite shocking ignorance of the basics (and sometimes, I suspect, wilful ignorance.)

      • e=mc2andallthat December 30, 2014 / 8:50 pm

        Me too — but still further to go..

  3. RequiresImprovement December 30, 2014 / 11:19 am

    It’s A Physics Thing, isn’t it? Once, at The School I’m On Balance Glad Not To Be At Anymore, Team Physics were sent a set of directives from the Data SLT. I remember us looking at each other, and saying something along the lines of “you can’t do that to data”, in the same way you’d talk about torturing a kitten…
    When I’m hyping up error analysis with year 12s, I usually tell them that the most important things they will learn in physics A level are:
    1 All measurements are at least a bit uncertain
    2 If you try to subtact two similar but uncertain measurements, it doesn’t work well, because the uncertanity often ends up bigger than the answer.
    And that is exactly what most progress spreadsheets try to do: compare an uncertain assessment with another uncertain assessment, or with a target that could be a grade higher or lower.
    The trouble is that it’s hard to say much of this without sounding picky, excuse-making or not-can-do.

    • e=mc2andallthat December 30, 2014 / 11:58 am

      “Data SLT”? Oh lordy…
      I remember being very pleased with myself when Physics got some nice looking statistical feedback from an official source. I discussed it with a qualified statistician who pointed out the 95% confidence bars above and below the data point for Physics. “What this means,” he said, “is that there is a 95% chance that your department is very good, or only so-so, or somewhere inbetween.” Gee, thanks a bunch.

  4. michaelslav January 4, 2015 / 3:06 pm

    I really like the way you have articulated this! You may have already come across The Signal And The Noise by Nate Silver ( I was led to it by a blog from Daisy Christodoulou last year –

    What you term ‘professional judgement’ is akin to the need for a theory that Daisy demonstrates we need when analysing (big) data.

    • e=mc2andallthat January 4, 2015 / 5:38 pm

      Many thanks for the kind comment, Michael. I’d read Daisy’s blog a while back but I haven’t read Silver’s book yet. I think Silver (quoted by Daisy) is spot on when he says “statistical inferences are much stronger when backed up by theory or at least some deeper thinking about their root causes”.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s