Re: st: 2 dimensional graph for joint distribution

[David Hoaglin <dchoaglin@gmail.com>]

Lulu,

That graph in Train's paper is definitely a type of display that one
should avoid.

Unless you have a fairly large number of combinations of the
coefficients of odo and bill, you will have little way to develop the
surface corresponding to the joint density function of those
coefficients.

For numbers of combinations like the six in your example, it might
work to use a "bubble plot" --- a scatter plot of the coefficient of
bill against the coefficient of odo in which each point is plotted as
a circle whose area is proportional to share.  I don't know whether
such a plot is readily available in Stata.

Regards,

David Hoaglin

On Wed, Oct 30, 2013 at 7:07 AM, Lulu Zeng <luluzengnz@gmail.com> wrote:
> Dear Statalist
>
> I am having trouble creating a two dimensional joint distribution bar
> graph in Stata.
>
> I'm trying to graph the joint distribution of 2 coefficients I
> estimated from a nonparametric choice model using fixed mass point
> method following Kenneth Train's approach in his 2008 paper (EM
> Algorithms for nonparametric estimation of mixing distributions). The
> graph I am trying to make is similar to the one on page 65 of Train's
> paper: http://elsa.berkeley.edu/~train/EMtrain.pdf
>
> My 2 coefficients (both discrete) are called odo and bill, and the
> variable share gives the distribution of the 2 coefficients (share
> adds up to 1), for example:
>
> odo       bill        share
> 0.25      0.58      0.15
> 0.48      0.93      0.17
> 0.33      0.20      0.09
> 0.76      0.64      0.23
> 0.87      0.28      0.25
> 0.53      0.83      0.11
>
> I want to graph the joint distribution of odo and bill with them on
> the x axis (2 dimensions), and their joint distribution shown as
> percentages on the y axis.
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/