How to scale points in scatterplot by area

Question

How can I scale the points in a Stata scatterplot by another variable? (I would like to get the area of each point, so I would like to scale by the area, or the square root of the variable, but this step is trivial.)

In R, I would do it as follows:

library(ggplot2)
ggplot(mtcars, aes(x = mpg, y = hp, size = sqrt(wt))) +
    geom_point() + 
    scale_size_continuous(name = "weight")

In Stata, I would like to do something like:

sysuse auto2, clear
gen weight2 = sqrt(weight)
scatter price mpg, msize(weight2)

But it doesn't work, and says:

(note:  named style weight2 not found in class symbolsize, default attributes used)

I can use weights, but apparently, these do not produce the desired result - see https://www.statalist.org/forums/forum/general-stata-discussion/general/1538359-scatterplots-with-weighted-marker-size-revisited?q=scatter%20area.

Any ideas of what I can do?

Nick Cox · Accepted Answer · 2025-02-25 13:47:15Z

3

Much depends what you want. Most people using this kind of plot seem to want mostly some almost qualitative specification of which data points correspond to very large ... very small values on some third variable.

Marker size is for that case given not by the msize() option, but by specifying weights.

. sysuse auto, clear
(1978 automobile data)

. twoway scatter mpg weight [fw=price^3] , ms(Oh)

answered Feb 25 at 13:47

Nick Cox

37.4k6 gold badges37 silver badges51 bronze badges

Sign up to request clarification or add additional context in comments.

4 Comments

bill999 Feb 25 at 15:18

Thanks! I would like the size of each point to exactly reflect the area as defined by the third variable.

Nick Cox Feb 25 at 15:31

I don't think any choice guarantees that. What would that mean if dynamic range was millions or bigger? What does size mean? Area or diameter?

bill999 Feb 25 at 15:39

Thanks. In my application, the dynamic range is much smaller (number of observations that a point is based on). If it were too big—millions or billions, say—I would probably take the log. And size would be area, not diameter.

Nick Cox Feb 25 at 15:52

I think you'd need to elicit more details from StataCorp technical support.

Bailey · Accepted Answer · 2025-02-25 04:21:24Z

-2

If it doesn't HAVE to be a scatter plot, the general consensus is to just use bubble. basically, the exact same code but use bubble instead of scatter plot

sysuse auto2, clear
gen weight2 = sqrt(weight)
twoway bubble price mpg, size(weight2)

answered Feb 25 at 4:21

Bailey

11 bronze badge

2 Comments

bill999 Feb 25 at 4:25

Thanks! I tried the code, and get this error: "bubble is not a twoway plot type"

Max_Stone Feb 25 at 9:11

The code is provided was not correct, it was a great solution but there was a simple syntax error, always try and test your answers first before submitting them

Collectives™ on Stack Overflow

How to scale points in scatterplot by area

2 Answers 2

4 Comments

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

4 Comments

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related