Truncated y axis in interaction plots I am trying to visualise an interaction of two factors on a single Likert-scale item in R with data similar to the following: 
A <- c("P", "P", "V", "V")
B <- c("C", "R", "C", "R")
Y = c(6.71, 6.42, 6.75, 5.96)

When using the standard R command interaction.plot(A, B, Y), R automatically truncates the y axis to maximise the space taken over by the plotted means

Even though I have a significant interaction, I find it somewhat disingenuous to show only the small y range in which the interaction takes place. But if I set the y axis to show the whole range ´interaction.plot(A, B, Y, ylim=c(1,9))`, my interaction is nearly lost.

So I'm wondering which is the more common or illustrative style. Do I truncate the y axis in order to highlight just how my interaction works, or do I show the whole scale to give an indication of the magnitude in differences this interaction effect produces?
 A: It depends on a number of factors what you want to do here.  Are there additional graphs that might have different ranges of the y-axis?  You would need to select a common range then.  Is the range of values you have arbitrary?  Could they really have gone from -20 to 200?  Or, is there a very fixed possible set of values?  If the latter it's often customary to keep all but if the former you might want to select the graph that maximizes the effect.  You can also select other values such as the standard deviation around the grand mean as your y-axis range.  There are no hard and fast rules here.
Also, think of the conceptual questions and actual effect sizes.  Is your interaction small in the scheme of things?  Is it actually quite large because the mean values tend to vary very little?  
In the end, you should be considering what plot most honestly conveys the meaning in your data.  That's the question you want to answer, not what maximizes your interaction.  Consider in that honesty, or truth seeking, that you also have to convey your story to your audience.  If they cannot understand it, even though it honestly presents the numbers that's not the most honest way to convey the meaning.  You need to convey the meaning without distortion.  In some cases that really does mean maximizing the interaction, in others it does not.
Your range issue might be solved if you added error bars, like 95% CI.  It will necessarily broaden the range and convey your interaction.
As a final alternate solution you might consider displaying your data in a graph like you have and adjacent displaying a graph of your effects, A, B, and A:B.  That graph could have confidence intervals indicating both the magnitude of the effects and what range of effects you could have called significant.  Such a graphs also make any claims about the meaningfulness of non-significant effects much more compelling.
