A 1580 Map Shows Why Data Visualization Is Never Neutral | Hacker News Reader

A 1580 Map Shows Why Data Visualization Is Never Neutral

thereader.mitpress.mit.edu

81 pointsthe-mitr25 comments

Data visualisation can never be neutral for the same reason photography can never be neutral - you cannot do it without making endless decisions about framing even if you try and capture your subject ‘neutrally’. What you include, what you don’t, whether your axies focus on the charge range or full numeric ranges and so on. See Susan Sontag’s On Phototography for a deeply considered read on this.

Also relevant:

- https://en.wikipedia.org/wiki/All_models_are_wrong

- https://en.wikipedia.org/wiki/Map%E2%80%93territory_relation

There's a difference between showing why something is "never neutral" vs merely "not always neutral".

One example would be sufficient to establish that it is "not always neutral", but I think no example can establish that data visualisation is never neutral.

It's not even sufficient to show that maps are never neutral.

As far as neutrality is concerned, I'd like to see someone argue that a graph of sin(x) with x between 0 and 2pi and y between say -1.5 and +1.5 is not neutral, provided it is presented as a graphical approximation of the function and domain.

For other numerical graphs, such as stock prices, or population development, there's always the possibility that it isn't represented in sufficient detail, but that would depend on the user's knowledge and expectation, I think. How to lie with statistics is a good starting point for that. The article suggests the same idea, but makes it a social issue.

By limiting the graph to [0, 2pi] and presenting it as a graphical approximation of the function and domain, you've obscured the fact that the function is periodic. Visualizing the interval [-3pi, 3pi] would make it more obvious that the function is periodic.

Depending on the color choices, it may be more difficult for colorblind folks. Choosing to print or display it is a choice that affects the blind.

That doesn't change the neutrality of the graph itself, does it?

The idea of a 'graph in itself' is nonsense. Graphs are viewed by people and the functional relationship of a graph and its viewers is conditioned on choices the graph's author makes.

Assigning neutrality or bias without viewer of interaction is a category error. It is not an intrinsic property of the graph despite our wanting to simplify the world this way. It's like saying chocolate is intrinsically the best candy when what's being said is 'my relationship to chocolate is that it's my favorite.'

So your point is: there is never neutrality, and the OP "map shows why etc." is just a trivial remark. I think you're mistaking what neutral means.

Domain and function were given.

> you've obscured the fact that the function is periodic

No. Periodicity is not a property in a graph. The graph still shows the function in that domain neutrally, as far as I can tell. Neutral meaning: impartial, without strongly marked characteristics.

If you insist on periodicity: there's an infinite (uncountable) number of properties not shown. If that violates neutrality, every statement is biased, which makes the article's statement rather trivial.

The point of data visualization is to communicate through graphical representation of information.

If you have to specify everything in text and you don't intend to actually communicate anything through graphics and if you only focus on pure, sterile mathematical functions which don't have an ideological component that could be biased to begin with... then surely you can see that your statement is the trivial one.

> The graph still shows the function in that domain neutrally

The key word being "in that domain". You can use the same excuse every time you mislead people with statistics. The fact that your choices can mislead people proves that they're not neutral.

> The key word being "in that domain".

You have to accept the limitations. A map is not the territory. Everyone knows that and accepts the map for what it is (because it is useful). It is not misleading to present a graph in a limited domain. It conveys that it does not represent the full function.

But that's not what neutrality is about. Being neutral does not mean representing the full truth. It means: not distorting, being impartial, not stressing a particular aspect, etc.

That the graph exists at all is a non-neutral statement about that graph versus all other possible graphs that could have been drawn and were not.

You can only claim that if you think nothing can every be neutral, which makes the discussion rather moot.

Your wording is so ambiguous that it's impossible to determine which of two opposite claims you are making.

Easy. You’ve already bought into the cultural assumption that this equation will be graphed on a Cartesian plane. Using polar coordinates instead, with an angle of x and a radius of y (a perfectly valid interpretation, even if it’s not what you envisioned), gives us a circle. These outputs have their own cultural associations: the former, music (particularly synthesizers); the latter, radar or sonar indicators. Showing the radius enhances the effect. All else being equal, if a new unexplained technology is said to produce that output, a clever data visualizer can bias the viewer towards viewing it as militant or peaceful. This is strictly a modern cultural bias; a hundred years ago, viewers would not have made either of those connections.

> We too often assume that visualization is transparent, a method that lets us see through to the significance of the data underneath. But the reality is that the choices made when visualizing data are nearly as significant as the data itself.

This idea seems inconsistent with the idea that no map can be objective. The examples given have to do with an unfamiliar audience (the Spanish) misinterpreting the map. There is an explicit acknowledgement that a correct interpretation does exist! There is objective info in there if you know how to read it.

I think there is a hidden assumption that objective means "every single person who sees this will understand it." This is of course impossible but it's important to emphasize that the majority of people, in almost all cases, will mess something up. So I dislike the framing of data visualization is not neutral, the analyst is imposing their view on the world and exercising lots of power, etc and prefer "it's just stupidly hard and you gotta pick a few people you are actually talking too."

For a scientist this is a somewhat obvious point. The purpose of figures in the paper is to make certain specific points. You don't make a figure just to have it (if it is a good paper). In the same time, adjusting the plots to make specific points doesn't mean making them misleading (which by done by some, by playing with axis ranges, or coloring of points or smoothing).

I ran into this building a site that shows where the money in a litre of petrol goes in Pakistan. Every number comes straight from the regulator's published price notices, so I assumed the chart would be neutral. It wasn't, because the choices were all in the grouping.

I could have split the import price into "crude" and "premium", but the documents print cost and freight as one line. Any split would have been my estimate presented as fact, so it stays one line. A row the document doesn't carry shows as empty, not 0.00, because a zero tells the reader something the source never said. I also renamed a bucket from "The state" to "Government". The figures didn't change, but people read "the state" as something abstract and "government" as someone who takes a cut.

Same data and same sources, but each of those calls changes what a reader walks away believing.

If you had access to all the underlying data could you then achieve a neutral visualization?

No, because all lossy data compression is biased.

I don’t get why this was downvoted: it conveys a personal experience on the subject, and I for one enjoyed reading it; perhaps I’m missing something?

Because AI-generated text goes against HN guidelines.

How do you know it is AI generated? It didn't cross my mind when I read it.

That last sentence reads very AI, very common sentence structure for AI. Pangram does give it a high confidence 100% generated, and Pangram is tuned for a low false positive rate.

It's very possible GP used AI to translate to English. In my experience llms love to change sentences to match their own preferred ways of speaking when translating.

Reply on news.ycombinator.com