Consider the distributions I have plotted below. They are of the same variables, one in normal histogram form and another in kernel density (Epachanov).

enter image description here

As far as I know, the auc of the kernel curves is 1, so I'm not sure why the y-axis would need to change sometimes for that. Yeah, the main thing is I'm not sure why sometimes the y axis is the same and sometimes the y axis is different.


Why did the y axis remain the same for the left two (0 to .8) but changed for the right two (histogram is 0 to 14, kernel is 1 to 7)? Feel free to explain about how the y axis factors into interpretation if you don't mind.


For the LHS plots, the areas under the curves are n near $1$, so you don't see a range change in y-axis. We can inspect it visually, by reading $y$ values for each bin, and multiplying it with the step size, $0.25$ for the blue histogram. For example, the it has an approximate area of $$(1/3\times3+1+2+3+3.5+5+7\times2+3.5+1+1.5)/10\times 0.25\approx 0.91$$ This is my simple approximation, you can come up with another, or directly calculate it from the data. But, it'll be around $1$. The situation is similar with the red histogram. KDE is not the same as normalized histogram but for simplicity you can think like we normalize the curve with 1/AUC. If AUC is close to $1$, then normalization won't affect the value on the y-axis a lot.

It's harder to do visual inspection for the red plot on RHS, however, clearly, the AUC won't be close to $1$ so that the resulting KDE curve's y-axis values will be in different range. Also, some jumps might not well represented in KDE smoothed PDF. And, for the blue curve in RHS, y-axis range didn't change significantly. In the histogram, most of the values fluctuate around $1$, and the AUC is therefore $1$ since the support spans $[0,1]$. The KDE also produces a curve with y-value near $1$.


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.