Sankey Plot for Panel Data Transition Probabilities

[Josephine Auer]

Hi all, hoping to get some assistance in constructing a Sankey Plot to visualise a simple 2x2 matrix of transition probabilities using Stata 17.0. I have read the help documentation attached to the sankey_plot package (author: Fernando Rios-Avila) but am having difficulties adjusting the code examples to my panel data set.

My data: I have a panel data set of just under 400,000 observations (about 20,000 distinct id units over 20 years). My variable of interest is a binary variable called "neet" (indicates whether an id was NEET in a given year or not) and in this example, I am restricting attention to a subgroup classified as "young20" and only considering transitions from 2019-2020 (last two years in my sample). I have constructed a 2 x 2 transition matrix to compute the probability of transitioning between "NEET" and "non-NEET" between 2019 and 2020 for this group using the code:

xttrans neet if young20==1 & year>2018

Goal: I would now like to visualise the transition probabilities defined for me in the above transition matrix using a Sankey Plot (open to any other suggestions on how to visualise transition probabilities, but from some reading online understand that a Sankey Plot is typically used). Unfortunately I am unable to post my data here and am not sure if this is typically required for Statalist posts (first time posting in a couple of years)! Any guidance would be much appreciated.

[Nick Cox]

Sankey plots look pretty -- those smooth curves look pleasing and there is always the possible reader buzz of how did you do that? -- but I've never seen one for one-step changes that seemed clearer than a two-way bar chart. In this case you have 2 x 2 probabilities and 2 suffice, as the probability of remaining in the same state is presumably just the complement of the probability of changing. I'd use a bar chart and label with the numbers.

I am positive about Sankey plots where there is a series of stages, say students dropping out, changing course or proceeding in the same course on a degree programme lasting several years, but otherwise they can be a lot of work for little or no obvious gain.

[FernandoRios]

HI Josephine,
I have recently added an update to sankey_plot but not yet pushed to ssc (i want to add a couple of other things before I do).
I think this will give you want.
See an example below:

**Data management
webuse nlswork
xtset id year
** this makes your data Wide
gen fmsp = f.msp
** labels the variables
label define msp 1 Married 0 Single
label values msp fmsp msp
** uses a cleaner scheme
ssc install color_style
set scheme white
** produces the sankey plot
sankey_plot msp fmsp if msp!=. & fmsp!=., wide xlabel(1 "Before" 2 "After") title("Transition of Marital Status")
** or if you want slightly different palette
color_style egypt
sankey_plot msp fmsp if msp!=. & fmsp!=., wide xlabel(1 "Before" 2 "After") title("Transition of Marital Status")

Hope this helps

[Dastan Aseinov]

HI Josephine,
I have recently added an update to sankey_plot but not yet pushed to ssc (i want to add a couple of other things before I do).
I think this will give you want.
See an example below:

**Data management
webuse nlswork
xtset id year
** this makes your data Wide
gen fmsp = f.msp
** labels the variables
label define msp 1 Married 0 Single
label values msp fmsp msp
** uses a cleaner scheme
ssc install color_style
set scheme white
** produces the sankey plot
sankey_plot msp fmsp if msp!=. & fmsp!=., wide xlabel(1 "Before" 2 "After") title("Transition of Marital Status")
** or if you want slightly different palette
color_style egypt
sankey_plot msp fmsp if msp!=. & fmsp!=., wide xlabel(1 "Before" 2 "After") title("Transition of Marital Status")

Hope this helps

Hi Fernando,

How can I define additional options for the sankey plot, if any?

Thanks in advance

Dastan

[Dastan Aseinov]

Hi Fernando,

How can I define additional options for the sankey plot, if any?

Thanks in advance

Dastan

I apologize to Fernando, this is no longer necessary.

my mistake was I didn't copy sankey_i and sankey_plot.shlp