Something Going Up Is Not Always Good
Even though our unique ability to deal with complexity propelled humans to the top of the evolutionary heap, we still crave simplistic (i.e., overly simple) explanations. I promote the value of simplicity in my work, but never simplicity that sacrifices truth. Simple things can and should be explained simply. Complex things can and should be explained as simply as possible, but never in a way that disregards or misrepresents their complexity.
When people hold simplistic assumptions about data, we should educate them, not accommodate their ignorance. One such assumption is that, in a time series, values going up are always good and values going down are always bad. I find it odd that people tend to interpret data in this manner, because no one interprets life in this manner. While we consider it good when our incomes go up or our health improves, we have no trouble recognizing that the cost of food going up or increases in suffering are bad. Why would we interpret data in this naive manner?
How do you deal with the commonplace exceptions to the “going up is good assumption,†such as the variance between actual and budgeted expenses? When considering expenses, being over budget is usually considered bad. Through the years of teaching data visualization courses, participants in my classes have often suggested that this assumption should be accommodated by reversing the quantitative scale, placing the negative values (i.e., under budget) above and the positive values (i.e., over budget) below. Is this an appropriate solution? Representing negative values as going up creates a new source of confusion, and does so unnecessarily.
Rather than accommodating ignorance by twisting data into awkward arrangements, why not correct the error instead? It is easy to explain that things going up aren’t always good in a way that everyone can understand. When specific cases of ignorance can be banished so quickly, easily, and permanently, why perpetuate it?
Data sensemaking and communication fundamentally seek to replace ignorance with understanding. Everything that we do in this venture should be done with this in mind. When we accommodate ignorance, we condone and encourage it. Doing so undermines the integrity of our work and the outcomes that we should be working hard to achieve.
Take care,
7 Comments on “Something Going Up Is Not Always Good”
I guess I was never really concerned with this tendency, since people familiar with a metric know when up is good or bad. For example, the phrase “costs are going up” really doesn’t sound good to anyone.
So where I draw the line on simplifying data is typically at the intended user – Many people may assume that “going up” is a good thing, but I don’t need to accommodate everyone. I only need to simplify my presentation to the extent that a person knowledgeable of the measurement would need it simplified, and no farther.
And if expenses or late shipments or machine failures or employee fatalities are going up, I would expect my audience to already know when it’s not a good thing.
I have encountered this “going up is good, going down is bad” in my teaching. Now I teach mostly graduate students and don’t see it so readily, but the years of teaching undergraduates produced repeated examples of this. It was always amusing, and distressing, to see students interpret downward sloping lines as bad, even when the data was on mortality rates, for example. I agree that manipulating the display is not the way to address the problem – in some ways, it compounds the problem since we still have the root cause (that down is bad) and now we have a poor visualization as well.
What is not so clear to me is how to address the problem. I suspect it is System 1 thinking (to use Kahneman’s term) – not due to rational thought, but almost a biological response (although I doubt it is biological – after all, in early human history, up probably meant climbing which is hard work, and down was a breeze). It was fulfilling to point out students’ misinterpretations of the slopes and I only hope that the examples did something to change the underlying problem. But I really have no way of knowing. Perhaps it was just an example of school work they got wrong. There is something fundamental about quantitative literacy that needs to be addressed to get at the cause of this thinking, but I am not at all sure what that is.
This is an issue I encounter fairly often, and I find it most problematic in situations where the same screen has multiple metrics with opposite directions, such as sales and expenses.
Perhaps the best example I’ve dealt with for this issue is currency exchange rates. These are traditionally displayed in a specific order, and the result can be that one currency pairing is positive when going up, while another is negative when going up, based on our typical cash flow. Unlike a more obvious metric, such as “expenses”, the reader may not even know which direction is meant to be positive (and this may change over time).
Some solutions we’ve worked with for this include trend arrow pairs (red up arrow / blue down arrow when up is bad, blue up arrow / red down arrow when up is good), verbal explanation (“Higher is better”), and coloring the line’s segments based on the trend direction (this one works well when we have a small number of lines, but is very difficult to track when we have a large number of lines or charts). I keep thinking there’s a better way, though.
Conceptual metaphors such as UP IS GOOD have been shown by cognitive linguists to exist nearly universally across the languages of the world, and to be intrinsically linked to human perception and cognition (for a start, check out Lakoff & Johnson 1980 “Metaphors we live by”). If so, it’s not a learned behavior to show a positive bias when interpreting diagrams that show something going up. Trying to make people unlearn this behavioral pattern may be both futile and also contrafactual.
Scmuddi,
I’m familiar with Lakoff’s work. I am not aware, however, of Lakoff or any other linguist claiming that “UP IS GOOD” is intrinsically linked to human perception or cognition. Particular assumptions are built into perception or cognition, such as “the light is coming from above” is built into perception and “up is quantitatively greater in value than down” appears to be built into cognition. (Note: The assumption that up is good is purely cognitive, not perceptual.) Even if some linguists do in fact make the claim that “up is good” is built into cognition, and that claim turns out to be valid, something that is intrinsically linked to human cognition can and should be countered when it is erroneous. Suggesting that we cannot or should not point out to people that “UP IS NOT ALWAYS GOOD” because the effort would be “futile and also contrafactual” is rather silly.
Just scream really angrily when delivering the numbers…that should get the point across unambiguously.
On a more serious note, I have noticed some cases where this phenomenon doesn’t seem to apply. For instance, when we look at certain error data that is usually zero (wrong-site surgery, for ex), it seems intuitively obvious that the one or two ‘bumps’ in the data are abnormal and not good. I wonder if others have noticed the same thing, and whether or not there might be a good explanation for it.
David,
There are many audiences for whom this erroneous assumption (i.e., up is always good) does not apply. People who understand their data don’t make this mistake. Unfortunately, we expect people who don’t understand their data to make effective use of it. The solution is education, not accommodation.