## Which provides more information--probability density distribution or scatter plot with locally weighted regression line? Which is easiest to comprehend?

I don't mean in *all* contexts, but here, which is better? Probability density distributions or scatter plots with locally weighted regression? Why?

**Pair A**

**Pair B**

## Reader Comments (2)

In Pair A, the probability density distributions seem to be saying something about how the blues might be two separate populations, while the reds are just one. If that's relevant to the point you're trying to make, then the probability density distributions are better. If not, then they are worse as that irrelevant point might overwhelm the point you're trying to make.

In Pair B, if you include a regression line with the probability density distributions, they would be more informative as scatter plots with too many dots are hard to differentiate. Hence, in the scatter plot, I cannot see that CRT 0 has two peaks at med-hi and hi Religiosity. But, without a regression line, it's too hard to extract that info from the probability density distributions by eye - one can see a dip somewhere in the middle (although the min looks to be around -0.5 instead of +0.25), but it's hard to tell which end is higher.

@Jonathan-- I agree that both approaches are not fully satisfying. In B, the PDDs, as rendered (in ggplot), creates the false impression that each level of CRT was equally populated w/ rspts-- definitely not true, as one can see from the scatterplot.