Cognitive Daily

You can get a lot of information from a simple bar graph, but to what extent does the arrangement of the bars matter? You can find great commentary about good design, but what about a nice clean experiment? Martin H. Fischer led a team that asked participants to indicate if a given relationship was true or false, based on a variety of different bar graphs. For example, is A > B in this graph?

And what about this one?

If you are like most people, it was easier to confirm that A was larger than B in the first graph — where the bars were oriented vertically. In addition to manipulating whether the bars were vertical or horizontal, Fischer et al also looked at where the numbers fell along the number line. In the following graphs, the absolute difference between the black and white bars is a constant, but it might take you a minute to really see that.

In previous research, Fischer had found that participants are faster to make left responses to smaller numbers, and right responses to larger numbers. In other words, when confronted with two numbers, you are faster to identify the leftmost one as the smaller, and the rightmost one as the larger. This suggests that your underlying mental number line starts with small numbers on the left — which is certainly the way most rulers are printed. Given this left-right coding, it would follow that horizontally oriented bars would be easier to read. However, that’s not what Fischer et al actually observed: participants were significantly faster with vertically oriented bars depicting positive or negative values. The orientation difference didn’t hold for graphs depicting mixed pairs (one positive, one negative).

Participants were also significantly faster at making decisions regarding positive numbers, regardless of orientation. What is particularly interesting about these results is that our underlying mental number line, which seems to run from left to right, is not leading to an advantage for horizontally oriented bars. Instead, there is a suggestion that increased vertical height corresponds to “more” — something which might be more universal than a number line that matches Western reading habits.

One final, personal note about gridlines (those lines that extend across the graph for each scale mark): Yuck! Fischer et al included them in the stimuli for their experiment, and so I included them here, but I bet they get in the way as much, if not more, than bar orientation. Clearly, there is more work to be done.

Fischer, M. H., Dewulf, N. & Hill, R. L. (2005). Designing bar graphs: Orientation matters. Applied Cognitive Psychology, 19, 953-962.

Comments

#1 Mark Paris
February 27, 2006

We are encouraged to include grid lines so that people can take more accurate quantitative information from a graph. I am not certain that’s a good thing, but I have to face the fact that most of the people who “read” our reports or briefings usually only read a few summary bullets and look at the plots. People who actually need numerical information want the raw data.
#2 Owen Jones
February 27, 2006

You mention that we seem to visualise a number line from left to right. I have read that in countries like Iran where they write from right to left they visualise a number line from right to left and the difference in reaction time for comparing numbers is indeed in the other direction.
#3 z
February 27, 2006

i wonder about lefties; i know a young leftie who struggled an unusually long time with the right to left text reading frame, then all over again with the left-right on the piano vs. up and down on the music notation vs. high and low notes thing; and is now battling it again in the left-right negative-positive algebra number line.
#4 Jen
February 28, 2006

I’d be interested to see how these results would differ in different parts of the world (as you mentioned, with the whole Western left-to-right thing). Also, it seems that the way we read graphs is a learned interpretation. I’m tutoring a kid right now in math and teaching him how to do bar graphs. In math and science textbooks, it’s much more common to see and work with vertical graphs than horizontal graphs. Maybe I’m understanding this data wrong, but it seems like these results could be entirely explained by the conventions society uses and people becoming accostumed to these conventions. People just aren’t taught to read horizontal bar graphs like they are vertical bar graphs.
#5 coturnix
February 28, 2006

OT, but I have a question. I am looking for a good example I can use in class to show how a common-sensical explanation of a natural phenomenon is not neccessarily the correct explanation.

I want something that is familiar to most people and easy to demonstrate in the classroom. I would prefer a biological example, if possible (i.e., not demonstrating that Earth revolves around the Sun instead of common-sensical notion that it is the other way round).

Showing an image or a movie that messes up with people’s perception and leads them to a wrong answer should be coupled with a very easy and simple demonstration how one can, via experiment and/or simple calculations, figure out the correct answer.
#6 Mike
February 28, 2006

I would like to see the full range of sample graphs, but I think that the vertical style lends itself to easy scanability, regardless of whether the viewer reads left-to-right or right-to-left, based on the amount of contextual information that you can interpret in a compressed field of vision. The horizontal bars require truer ‘reading’ of the scale and context to quickly digest the information.

Perhaps another test would be to examine results based on graphs that do not have values for either axis-just pure spatial comparisons.
#7 PhilipJ
March 1, 2006

Those who enjoy this may also want to check out my quick recap of Edward Tufte’s classic, “Visually Displaying Quantitative Information” here.
#8 PhilipJ
March 1, 2006

… and for the record, Tufte abhors gridlines in more or less all situations. If they are absolutely required, he suggests using a light grey instead of thick black lines, as they take away from the actual data present and amount to nothing more than what he calls chartjunk.
#9 Dave Munger
March 1, 2006

My thought on gridlines is this: if you need gridlines to tell the difference between two data points, it’s very likely that the difference is not significant.