The data stem from a 2x2-factorial design where both factors have two levels. The x-axis shows the words of a sentence and the y-axis the mean reaction time for each single word. Here is how I visualize the data so far with base R:

enter image description here

I find the usual method of visualizing such data rather unappealing but I couldn't come up with a nice idea myself. Are there more appropriate ways of summarizing such data in one single plot?

  • $\begingroup$ What are the bars above and below the points? What are the stars? They need to be described or removed. Why is the axis broken when there are no points below the break, and the scale doesn't seem to actually break there? $\endgroup$ – shadowtalker Feb 28 '15 at 13:25
  • $\begingroup$ And if there's any kind of natural (read: substantively relevant) ordering on the x-axis, you should use it. BTW I'm posting these as comments and not answers because they don't really relate to 2x2 experiments, but I think they do need to be addressed $\endgroup$ – shadowtalker Feb 28 '15 at 13:29
  • $\begingroup$ The bars are error bars according to Cosineau (2005). The stars show significance but don't really matter here as this is random data. The scale with the original data is broken between 0 and a larger number. In short: This is a pure dumb plot only in existence to illustrate the problem without a lot of words. It obviously lacks the precision one would wish for. $\endgroup$ – lord.garbage Feb 28 '15 at 13:34
  • $\begingroup$ my suggestion stands: label it or drop it. Also, don't break plots just to get 0 in there, unless 0 is an important reference point for this data (and usually it isn't) $\endgroup$ – shadowtalker Feb 28 '15 at 14:08
  • $\begingroup$ I agree with the labeling but error bars et. al. I would outsource to a figure caption. The breaking-the-axis-part is controversial so we probably won't settle this here. Thanks for the tips. $\endgroup$ – lord.garbage Feb 28 '15 at 14:12

The X-axis is not a continuous variable, therefore connecting the adjacent X-values is not appropriate. In other words: ditch the connecting lines.

As for visualisation, one other option would be to use a bar chart. But I am afraid it would only make things worse.

IMHO, the best visualization would actually be to put the values in a table. One word in a row with your four values as columns.

  • $\begingroup$ "The X-axis is not a continuous variable, therefore connecting the adjacent X-values is not appropriate." I completely disagree. $\endgroup$ – shadowtalker Feb 28 '15 at 13:22
  • 2
    $\begingroup$ But I do agree that a table is a great idea. $\endgroup$ – shadowtalker Feb 28 '15 at 13:23
  • $\begingroup$ There is nothing prohibiting you from connecting the dots with lines, but the clarity of such visualisation is very poor. Just look at that graph from the Wiki page that you linked - you have to look it from a distance to see some patterns... $\endgroup$ – alesc Feb 28 '15 at 13:26
  • $\begingroup$ Call it preference, then. But try drawing that graph without the connecting lines: good luck seeing any patterns at all. $\endgroup$ – shadowtalker Feb 28 '15 at 13:28
  • $\begingroup$ I don't see the table-"plot" to be a viable option. Regarding the bar-chart-option: A bar chart were you put four bars for each word next to each other? $\endgroup$ – lord.garbage Feb 28 '15 at 14:19

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.