Exploring goodness of fit of MDS solutions

[jsibley]

After running MDS, is there a way to explore which points are
contributing to the calculated stress?

For example, if I click on a point in the fit and transportation plot,
I get a number? If it is far from the ideal line, I assume it is
contributing to a higher stress value. Is that correct? Is there a way
to see which pair of items correspond to this point?

Also, is there a way to compare the input distances and calculated
distances to see where the biggest discrepancies are?

Many thanks for a great product!

Jonathan Sibley
Montclair, NJ
USA

[Pedro Valero-Mora]

Dear Jonathan,

I do not have my books with me and the real expert on MDS was Forrest
but I will try an educated guess to your answers. If somebody knows
better, please help me.

On 16/08/2011 18:18, jsibley wrote:
> After running MDS, is there a way to explore which points are
> contributing to the calculated stress?
>
> For example, if I click on a point in the fit and transportation plot,
> I get a number? If it is far from the ideal line, I assume it is
> contributing to a higher stress value. Is that correct? Is there a way
> to see which pair of items correspond to this point?

For data without error, the plot of fit and transformation plot will
show all the points falling on straight lines (try the Miles&Kms file
for an example). Points that fall away of the straight line are
contributing more to the Stress.

You are right, the labels of the points are not correct. You may have to
guess what are the pair of items that correspond to the point by
printing the matrix and counting. In the future I might correct the plot

>
> Also, is there a way to compare the input distances and calculated
> distances to see where the biggest discrepancies are?

The Fit and Transformation Plot is the way to see that.

>
> Many thanks for a great product!
>
> Jonathan Sibley
> Montclair, NJ
> USA
>

--
Dr. Pedro M. Valero Mora (val...@uv.es)
Facultad de Psicolog�a
Universitat de Valencia
Mob. 669842979 Ext.93564
http://www.uv.es/valerop

SINTEC-INTRAS
Driving Simulation and Human Factors of New Technologies in the Vehicle
Institute of Traffic and Road Safety
Tlf. (34) 96 3393880 (Interno: 93910)
http://www.uv.es/sintec

METHODOLOGY OF BEHAVIOURAL SCIENCES
Data Visualization and Computational Statistics
Tlf. (34) 96 3983564 (Interno: 83564)
http://www.uv.es/visualstats/Book/
http://www.uv.es/prodat

CHECK THE VISUAL STATISTICS BOOK!
http://www.uv.es/visualstats/Book/

[Jonathan Sibley]

Thank you for your answers, Pedro.

In addition to what now seems to be a feature request to have a more
useful labeling of the points on the Fit and transformation plot, I
was hoping for away to generate a table showing the original and
fitted distances as an alternative to the chart. Is this also a new
fearure request?

Thank you.

Sent from my iPad

On Sep 2, 2011, at 12:47 PM, Pedro Valero-Mora <val...@uv.es> wrote:

> Dear Jonathan,
>
> I do not have my books with me and the real expert on MDS was Forrest but I will try an educated guess to your answers. If somebody knows better, please help me.
> On 16/08/2011 18:18, jsibley wrote:
>> After running MDS, is there a way to explore which points are
>> contributing to the calculated stress?
>>
>> For example, if I click on a point in the fit and transportation plot,
>> I get a number? If it is far from the ideal line, I assume it is
>> contributing to a higher stress value. Is that correct? Is there a way
>> to see which pair of items correspond to this point?
>
> For data without error, the plot of fit and transformation plot will show all the points falling on straight lines (try the Miles&Kms file for an example). Points that fall away of the straight line are contributing more to the Stress.
>
> You are right, the labels of the points are not correct. You may have to guess what are the pair of items that correspond to the point by printing the matrix and counting. In the future I might correct the plot
>
>
>>
>> Also, is there a way to compare the input distances and calculated
>> distances to see where the biggest discrepancies are?
>
>
> The Fit and Transformation Plot is the way to see that.
>
>>
>> Many thanks for a great product!
>>
>> Jonathan Sibley
>> Montclair, NJ
>> USA
>>
>
>
> --
> Dr. Pedro M. Valero Mora (val...@uv.es)

> Facultad de Psicología

> Universitat de Valencia
> Mob. 669842979 Ext.93564
> http://www.uv.es/valerop
>
> SINTEC-INTRAS
> Driving Simulation and Human Factors of New Technologies in the Vehicle
> Institute of Traffic and Road Safety
> Tlf. (34) 96 3393880 (Interno: 93910)
> http://www.uv.es/sintec
>
> METHODOLOGY OF BEHAVIOURAL SCIENCES
> Data Visualization and Computational Statistics
> Tlf. (34) 96 3983564 (Interno: 83564)
> http://www.uv.es/visualstats/Book/
> http://www.uv.es/prodat
>
> CHECK THE VISUAL STATISTICS BOOK!
> http://www.uv.es/visualstats/Book/
>

> --
> You received this message because you are subscribed to the Google Groups "Visual Statistics" group.
> To post to this group, send email to visual-s...@googlegroups.com.
> To unsubscribe from this group, send email to visual-statist...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/visual-statistics?hl=en.
>

[Pedro Valero-Mora]

Try CREATE DATA in the model menu

>> Facultad de Psicolog�a

>> Universitat de Valencia
>> Mob. 669842979 Ext.93564
>> http://www.uv.es/valerop
>>
>> SINTEC-INTRAS
>> Driving Simulation and Human Factors of New Technologies in the Vehicle
>> Institute of Traffic and Road Safety
>> Tlf. (34) 96 3393880 (Interno: 93910)
>> http://www.uv.es/sintec
>>
>> METHODOLOGY OF BEHAVIOURAL SCIENCES
>> Data Visualization and Computational Statistics
>> Tlf. (34) 96 3983564 (Interno: 83564)
>> http://www.uv.es/visualstats/Book/
>> http://www.uv.es/prodat
>>
>> CHECK THE VISUAL STATISTICS BOOK!
>> http://www.uv.es/visualstats/Book/
>>
>> --
>> You received this message because you are subscribed to the Google Groups "Visual Statistics" group.
>> To post to this group, send email to visual-s...@googlegroups.com.
>> To unsubscribe from this group, send email to visual-statist...@googlegroups.com.
>> For more options, visit this group at http://groups.google.com/group/visual-statistics?hl=en.
>>

--
Dr. Pedro M. Valero Mora (val...@uv.es)

Facultad de Psicolog�a

[sibleyj]

That is a reasonable suggestion, but the options I see are:

Stimulus coordinates
Fitted Distances
Transformed Data
Analyzed Input Data

Wouldn't I need a way to compare a matrix of the original distances to
the fitted distances, or a matrix showing those differences?

Are there ways to compare two matrices in Vista?

Thanks.

Jonathan Sibley

[Pedro Valero]

The plot mentioned before is the way to carry out the comparsion. You may identify points that fall far away of the diagonal and then look for the pairs corresponding to the point in the tables. It is a rather cumbersome process but it is the only way I can think of right now.