Choice plot

If you have an agent (or multiple agents) that can choose between two mutually exclusive states (1 and 2), then its preference for either state can be plotted with achoice plot.

I just made up the name…

If the agent has equal probability -at any given time- to be in state 1 or 2, you can plot the deviation from the chance level (i.e. 50%) of being in either states using a combination of color and location (e.g. above Vs below the x axis).

The axis

The Y axis represents the % time spent in state 1 (above X axis) and state 2 (below X axis). In this case I have plotted a group performance (see standard error bars).

The X axis is used to plot the minute by minute performance/choice.

Example plots

How different behavioral patterns look like in the choice plot:

random_choice_plot_multipanels
[A] Random choice between states 1 and 2, the agents spend an equal amount of time in both states. [B] Most of the time is spent in state 2 (below x axis). [C] The initial preference for state 2 is lost around the 6th minute of the session. [D] The opposite of panel C.

What I like about this plot

What I really like about this plot is its flexibility. This is how a good plot should be: it nicely bends to your will! I want a plot that can be easily customized to emphasize the story I want to tell. For example,

  • You can choose the location of the states (above or below the x axis) to emphasize a rising or descending trend (see panels C and D, respectively).
  • Deviations from the chance level are immediately visible thanks to the color on gray background (see panels A and B)

What I DON’T like about this plot

There are few things that I don’t like about this plot (mostly related to its implementation – i.e. the code in R):

  • On the Y axis the % time for state 2 is reported as negative (this is a quick & dirty hack-trick to plot the state in the reverse direction). However, this can be easily fixed.
  • The code is rather verbose and WET (… if WET is the oppposite of DRY – Do Not Repeat Yourself?) most of the code could use a refactoring to get rid of very similar instructions – basically everything in the for loop (lines 11~21).
  • When plotting group performances the graph may be a bit awkawrd because the single bars do not add up to 100% – *see note at the end of the post.
  • The managing of signs (thresholds, direction and +/-) may be a bit convoluted, but it makes sense if read by a human being …

 

The code (made in R)

This plot was generated (with random data*) in R – see code below.

 

Leonardo Restivo

Behavioral Neuroscientist, M.Sc., Ph.D. - Passionate about Behavior, Data Visualization & Psychology. Read my CV+résumé. Follow me on twitter @scipleneuro