[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: RE: St: Re: graph over

From   "Nick Cox" <>
To   <>
Subject   st: RE: St: Re: graph over
Date   Thu, 2 Apr 2009 12:41:32 +0100

The technique Allan mentions here can certainly be useful. 

But using the official command -separate- directly is much less work
than the route outlined here. This was added in Stata 6, and so is
certainly available to Allan as a Stata 9.2 user. 

The code would be something like 

. separate ESL, by(school) veryshortlabel 

. line `r(varlist)' time 

Conversely, the main motivation for the -separate- command was to
provide a direct tool for precisely this problem. 

However, for Caleb's original problem -xtline- is even more direct, as
already posted. 


Allan Reese

There's a generic answer to this that I prefer to call lateral thinking
rather than a kludge.

The data comprises three variables:
   ESL percentage        - the Y var
   time of observation   - the X var
   school (categorical)  - the group var

Scatterplot Y against X shows the overall pattern, and Stata has an
option to label each point with its group.
However, it's not immediately obvious how to plot the groups with
different symbols: msymbol(varname) is not available [HINT to StataCorp
- it would be nice.]  Nor how to join the points for each group.

The solution is to restructure the data, and one option would be reshape
long->wide, making ESL+school variables.  This might be more appropriate
if the time points of observations in each school were equivalent, and
cross-correlations would be calculated.

My usual solution is to keep the original cases but divide the
observations for each group into separate variables.
You can do this by a series of

   gen ESL1 = ESL if school==1   // etc
   xi , noomit|ESL
   recode _I* (0=.)
except that (in my Stata 9.2) generates the dummy main effect variables
as well, contrary to Manual [HINT to StataCorp - bug? Known?]

   scatter _IschoolX* time
plots each school with a different symbol (within system limits for no
of groups!).  Any other twoway plot with a varlist would be similar.

There are lots of graph problems that are best tackled by splitting one
variable into groups. 

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index