Sankey plot: separate scale for the right/left side

[Dastan Aseinov]

Hello Stata users,

I would be grateful for any assistance.

I employed the following syntax to generate a Sankey plot.

sankey x if shock!="." & copem!=".", from(shock) to(copem) by(y) labposition(100) labangle(35) labc(black) labsize(3) boxw(5) valsize(1) ctitles("{stSerif:{bf:HH affected by:}}" "{stSerif:{bf:Coped with shock by:}}") ctpos(top)  ctg(10) cts(3) smooth(7) sort1(name) format(%12.0f) recenter(mid) labscale(0.2) labprop percent valscale(0.5) valprop

Then I get this graph:


The percentage of each option on the right is shown on the same/common scale as on the left. Therefore, they appear so small that it is difficult to identify and compare the lobes on the right side relative to each other. I need your help so that the shares on the right side are shown in a separate scale.


Your guidance on resolving these issues is highly appreciated.

Best regards,
Dastan

[Nick Cox]

I haven't used sankey (from SSC?). What are the original data here? An 8 x 10 table, or something else? Why are values so much bigger on one side than on the other?

The author Asjad Naqvi asks people to raise issues on GitHub. Perhaps you've done that.

[Marc Kaulisch]

Maybe the alluvial -Package is more appropriate in your use case. https://github.com/asjadnaqvi/stata-alluvial

[Asjad Naqvi]

Thank you Nick Cox!

Dastan Aseinov, something is off with the figure. If the inflows and the outlows are fully balanced, then the bars should be the same height. Maybe you are using the "percent" option to show percentages? This is a beta option to test additional variations and I do not recommend using it. If you want to show percentages rather than actual values then calculate these directly and then use the standard sankey option.

Alternatively, you can use the "Alluvial" package as Marc Kaulisch suggested which is exactly designed for the type of visualization you are trying to generate.

Additional note: there is no left and right in Sankey. There can be one layer with a start and end destinations. So a request for rescaling one axis means adding the ability to rescale n number of layers. This is a dangerous step since it can lead to mis-representing data and this won't be implemented.