MATLAB Central Newsreader - scatter plot of raw data from mnrfit()

scatter plot of raw data from mnrfit()

Céldor — Fri, 20 Apr 2012 11:05:27 +0000

Hi All,

I have estimated regression model from mnrfit() and draw cumulative probability curves which works all right for me. I would like to see which points these curves are estimated from. Is it possible? It is not straight forward to me because as inputs I provide x continuous variable and y categorical variable with categories 1-5. How can I draw these points? What I want is equivalence of linear regression when I put data from which I estimate regression. In terms of ordinal probit model I get curves representing cumulative probabilities and I cannot use straight data I have estimated curves from.

For instance if I have
[B D S] = mnrfit(x,y,'model','ordinal','link','probit');
and
[P PL PH] = mnrval(B,x,'type','cumulative','model','ordinal','link','probit');

can I use simply P(:,cat) + S.resid(:,cat)? Where are the values the curves were estimated from ? 'cat' is category of my interest?

Please can someone help with that?

Re: scatter plot of raw data from mnrfit()

Céldor — Tue, 24 Apr 2012 08:12:07 +0000

"Céldor " <zebik@op.pl> wrote in message <jmrftn$3u1$1@newscl01ah.mathworks.com>...
> Hi All,
>
> I have estimated regression model from mnrfit() and draw cumulative probability curves which works all right for me. I would like to see which points these curves are estimated from. Is it possible? It is not straight forward to me because as inputs I provide x continuous variable and y categorical variable with categories 1-5. How can I draw these points? What I want is equivalence of linear regression when I put data from which I estimate regression. In terms of ordinal probit model I get curves representing cumulative probabilities and I cannot use straight data I have estimated curves from.
>
> For instance if I have
> [B D S] = mnrfit(x,y,'model','ordinal','link','probit');
> and
> [P PL PH] = mnrval(B,x,'type','cumulative','model','ordinal','link','probit');
>
> can I use simply P(:,cat) + S.resid(:,cat)? Where are the values the curves were estimated from ? 'cat' is category of my interest?
>
> Please can someone help with that?

Could someone please confirm whether it is possible or not? Thanks

Re: scatter plot of raw data from mnrfit()

Peter Perkins — Tue, 24 Apr 2012 12:52:58 +0000

There are examples right in the documentation that are as close to this
as I am aware of.

http://www.mathworks.com/help/toolbox/stats/mnrfit.html

The examples in the doc have a single predictor x that takes on integer
values. Even for a simple linear regression, it is not a simple matter
to plot versus multiple predictor variables. With MN response, you have
the added difficulty of (probably) needing to bin the data with respect
to the predictor variable(s) unless it is already discrete.

Hope this helps.

On 4/20/2012 7:05 AM, Céldor wrote:
> Hi All,
>
> I have estimated regression model from mnrfit() and draw cumulative
> probability curves which works all right for me. I would like to see
> which points these curves are estimated from. Is it possible? It is not
> straight forward to me because as inputs I provide x continuous variable
> and y categorical variable with categories 1-5. How can I draw these
> points? What I want is equivalence of linear regression when I put data
> from which I estimate regression. In terms of ordinal probit model I get
> curves representing cumulative probabilities and I cannot use straight
> data I have estimated curves from.
>
> For instance if I have [B D S] =
> mnrfit(x,y,'model','ordinal','link','probit');
> and
> [P PL PH] =
> mnrval(B,x,'type','cumulative','model','ordinal','link','probit');
>
> can I use simply P(:,cat) + S.resid(:,cat)? Where are the values the
> curves were estimated from ? 'cat' is category of my interest?
>
> Please can someone help with that?