Absolute values and percentages

[MHMPires]

Hi,

I have created a Sankey diagram with your Stata package but was now wondering whether it is possible to include percentages alongside each of the absolute values?

Any help is greatly appreciated.

Thank you.

[moritzpoll]

I actually had the same use case and doctored around in the ado file of a previous version (1.6) which I had previously altered for something else. Now when I put the showtotal and percent options simultaneously, I get absolute values in parentheses and percentages as flow labels. The key is to create a second variable height, say height_raw, around where height is created that doesn't go through the percentage conversion but that gets carried through all other transformations that height goes through.:

gen flow1 = val1
gen flow2 = flow1
...
reshape long var val flow, i(id layer xcut) j(marker)
...
bysort layer2 var: egen flow_out_temp = sum(flow) if marker==1 // how much value is sent out
bysort layer2 var: egen flow_in_temp  = sum(flow) if marker==2 & markme!=1 // how many value comes in
bysort layer2 var: egen flow_out = max(flow_out_temp)
bysort layer2 var: egen flow_in  = max(flow_in_temp)
recode val_in val_out flow_out flow_in (.=0)
egen height_raw = rowmax(flow_out flow_in) // back-up for showing raw counts in percent graph

This then culminates in:

if "`nolabels'" == "" {
		if "`showtotal'" != "" {
			gen lab2 = name + " (" + string(height, "`format'") + ")" if tag_spike==1
			if "`percent'" == "" gen lab2 = name + " (" + string(height, "`format'") + ")" if tag_spike==1
			else gen lab2 = name + " (" + string(height_raw, "%9.0fc") + ")" if tag_spike==1
		}
		...
}

Result is below. The other difference to what the package does natively is that I flip the angle of labels for the categories I track (blue) differently to the ones that leave my study (grey) since some of them are pretty small while the purpose of the percent option for me is to keep the tracked categories at 100% and thus large enough for vertical labels. I'm sure @asjadnaqvi can come up with something much more elegant and I of course assume no responsibility for any of this working :)

[MHMPires]

wow many thanks for sharing! I'll try this out and let you know how it goes!

[asjadnaqvi]

@moritzpoll and @MHMPires, the issue is not this specific implementation, but generalizable rules that can hold across a wide array of Sankey visualizations. In that regard, it is also not clear what percentage represents. Depending on the context, users can request percentages to mean:

• Share of the global maximum height across all the layers
• Share of height in current layer
• Share of a node in the current layer (which is what the above customization shows).

So in theory one actually needs multiple percent options to fit all the cases.

I was already hesistant to implement percent as it currently exists in the package. It also also representing a very specific use case. So I would recommend to just clone the program and implement how you want it. I still need to think through the iterations and maybe decide on a generalizeable case but in theory I am not in favor of it.