The controversial and ethically questionable Facebook Study published in the June 2014 issue of PNAS has by now been greatly criticized in the public media by privacy advocate and social scientists alike. Not without culpability is PNAS itself, which allowed this questionable work to be published.
I personally have strong opinions about the ethical implications of the researcher's actions, and feel that they are absolutely important and worth discussing; but since that has been greatly debated already I will not belabor the point.
Instead I shall nitpick, which is very much the calling of one of my intellectual stature. In the published paper, Kramer et al. display their data graphically using a bar chart. They show a change in measured emotional words percentage on a bar chart FOR WHICH THE ORIGIN IS NOT ZERO. The authors instead place the origin(s) at 5% and 1.50% respectively, which make the fluctuations see much larger. The offending graph is reproduced below.
I see this as being highly disingenuous. I also question the point in flipping the Negative words per second. An uncharitable explanation would be to use the relative size difference to further mislead the viewer, though I do not claim to know the minds of the authors. It could just be that they thought it looked nicer.
I have reproduced the graph(s) without those 2 questionable factors. This graph has a shared, Zeroed origin and constant orientation.
It is in my opinion a more useful representation of the data, which does not hide the relative variance of the control and experiment groups over multiple attributes.
The error bars are still there. Given how very difficult they are to see at this zoom level I can see the argument being made that non-zero axes were chosen to make the error bars visible. At any rate we have a more realistic picture of this data-set, from which we can draw our own conclusions about the efficacy of Facebook's manipulations.