

What is the function of the syntax highlighted in yellow? I am running this Multinomial Probabilities matrix
Matrix.
* intercept1 intercept2 pared public gpa.
* these coefficients are taken from the output.
compute b_gen = {1.689354 ; 0.057928 ; 1.162832 ; 0.629541}.
compute b_voc = {4.235530 ; 0.113603 ; 0.982670 ; 1.274063}.
* overall design matrix including means of public and gpa.
compute x = {{1 ; 1; 1}, make(3, 1, 52.775), {1, 0; 0, 1; 0, 0}}.
compute lp_gen = exp(x * b_gen).
compute lp_voc = exp(x * b_voc).
compute lp_aca = {1; 1; 1}.
compute p_gen = lp_gen/(lp_aca + lp_gen + lp_voc).
compute p_voc = lp_voc/(lp_aca + lp_gen + lp_voc).
compute p_aca = lp_aca/(lp_aca + lp_gen + lp_voc).
compute p = {p_gen, p_aca, p_voc}.
print p /title 'Predicted Probabilities for Outcomes 1 2 3 for ses 1 2 3 at mean of write'.
End Matrix.


How about running
Matrix.
compute x = {{1 ; 1; 1}, make(3, 1, 52.775), {1, 0; 0, 1; 0, 0}}.
print x.
End Matrix.
to find out.
From: Peter Spangler
Sent: Tuesday, February 12, 2013 8:51 PM
Subject: Syntax
What is the function of the syntax highlighted in yellow? I am running this Multinomial Probabilities matrix
=====================
To manage your subscription to SPSSXL, send a message to
[hidden email] (not to SPSSXL), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSXL
For a list of commands to manage subscriptions, send the command
INFO REFCARD

Administrator

compute x ={{1 ; 1; 1}, make(3, 1, 52.775), {1, 0; 0, 1; 0, 0}}.
Will result in an error!
Likely what is required is:
compute x ={{1 ; 1; 1}, make(3; 1; 52.775), {1, 0; 0, 1; 0, 0}}.
1 3 1 0
1 1 0 1
1 52.775 0 0
Peter Spangler wrote
What is the function of the syntax highlighted in yellow? I am running this
Multinomial Probabilities matrix
*Matrix.
* intercept1 intercept2 pared public gpa.
* these coefficients are taken from the output.
compute b_gen = {1.689354 ; 0.057928 ; 1.162832 ; 0.629541}.
compute b_voc = {4.235530 ; 0.113603 ; 0.982670 ; 1.274063}.
* overall design matrix including means of public and gpa.
compute x =
{{1 ; 1; 1}, make(3, 1, 52.775), {1, 0; 0, 1; 0, 0}}.
compute lp_gen =
exp(x * b_gen).
compute lp_voc = exp(x * b_voc).
compute lp_aca = {1; 1; 1}.
compute p_gen = lp_gen/(lp_aca + lp_gen + lp_voc).
compute p_voc = lp_voc/(lp_aca + lp_gen + lp_voc).
compute p_aca = lp_aca/(lp_aca + lp_gen + lp_voc).
compute p = {p_gen, p_aca, p_voc}.
print p /title 'Predicted Probabilities for Outcomes 1 2 3 for ses 1 2
3 at mean of write'.
End Matrix.*
Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email me.

"Nolite dare sanctum canibus neque mittatis margaritas vestras ante porcos ne forte conculcent eas pedibus suis."
Cum es damnatorum possederunt porcos iens ut salire off sanguinum cliff in abyssum?"

Administrator

My Bad:
Actually no error. No coffee more like it.
The result will actually be:
1 52.775 1 0
1 52.775 0 1
1 52.775 0 0
MAKE (nrows,ncols,value)
David Marso wrote
compute x ={{1 ; 1; 1}, make(3, 1, 52.775), {1, 0; 0, 1; 0, 0}}.
Will result in an error!
Likely what is required is:
compute x ={{1 ; 1; 1}, make(3; 1; 52.775), {1, 0; 0, 1; 0, 0}}.
1 3 1 0
1 1 0 1
1 52.775 0 0
Peter Spangler wrote
What is the function of the syntax highlighted in yellow? I am running this
Multinomial Probabilities matrix
*Matrix.
* intercept1 intercept2 pared public gpa.
* these coefficients are taken from the output.
compute b_gen = {1.689354 ; 0.057928 ; 1.162832 ; 0.629541}.
compute b_voc = {4.235530 ; 0.113603 ; 0.982670 ; 1.274063}.
* overall design matrix including means of public and gpa.
compute x =
{{1 ; 1; 1}, make(3, 1, 52.775), {1, 0; 0, 1; 0, 0}}.
compute lp_gen =
exp(x * b_gen).
compute lp_voc = exp(x * b_voc).
compute lp_aca = {1; 1; 1}.
compute p_gen = lp_gen/(lp_aca + lp_gen + lp_voc).
compute p_voc = lp_voc/(lp_aca + lp_gen + lp_voc).
compute p_aca = lp_aca/(lp_aca + lp_gen + lp_voc).
compute p = {p_gen, p_aca, p_voc}.
print p /title 'Predicted Probabilities for Outcomes 1 2 3 for ses 1 2
3 at mean of write'.
End Matrix.*
Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email me.

"Nolite dare sanctum canibus neque mittatis margaritas vestras ante porcos ne forte conculcent eas pedibus suis."
Cum es damnatorum possederunt porcos iens ut salire off sanguinum cliff in abyssum?"


What is the role of the two columns of 0,1 combinations in the calculation of this probability matrix? . {1, 0; 0, 1; 0, 0}}
On Tue, Feb 12, 2013 at 6:51 PM, Peter Spangler <[hidden email]> wrote:
What is the function of the syntax highlighted in yellow? I am running this Multinomial Probabilities matrix
Matrix.
* intercept1 intercept2 pared public gpa.
* these coefficients are taken from the output.
compute b_gen = {1.689354 ; 0.057928 ; 1.162832 ; 0.629541}.
compute b_voc = {4.235530 ; 0.113603 ; 0.982670 ; 1.274063}.
* overall design matrix including means of public and gpa.
compute x = {{1 ; 1; 1}, make(3, 1, 52.775), {1, 0; 0, 1; 0, 0}}.
compute lp_gen = exp(x * b_gen).
compute lp_voc = exp(x * b_voc).
compute lp_aca = {1; 1; 1}.
compute p_gen = lp_gen/(lp_aca + lp_gen + lp_voc).
compute p_voc = lp_voc/(lp_aca + lp_gen + lp_voc).
compute p_aca = lp_aca/(lp_aca + lp_gen + lp_voc).
compute p = {p_gen, p_aca, p_voc}.
print p /title 'Predicted Probabilities for Outcomes 1 2 3 for ses 1 2 3 at mean of write'.
End Matrix.

Administrator

What is the point of this whole exercise?
Does the concept 'design matrix' resonate?
When in doubt write out the math!

Peter Spangler wrote
What is the role of the two columns of 0,1 combinations in the calculation
of this probability matrix? .
*{1, 0; 0, 1; 0, 0}}*
On Tue, Feb 12, 2013 at 6:51 PM, Peter Spangler < [hidden email]>wrote:
> What is the function of the syntax highlighted in yellow? I am running
> this Multinomial Probabilities matrix
>
> *Matrix.
> * intercept1 intercept2 pared public gpa.
> * these coefficients are taken from the output.
> compute b_gen = {1.689354 ; 0.057928 ; 1.162832 ; 0.629541}.
> compute b_voc = {4.235530 ; 0.113603 ; 0.982670 ; 1.274063}.
> * overall design matrix including means of public and gpa.compute x = {{1 ; 1; 1}, make(3, 1, 52.775), {1, 0; 0, 1; 0, 0}}.compute lp_gen = exp(x * b_gen).
> compute lp_voc = exp(x * b_voc).
> compute lp_aca = {1; 1; 1}.
> compute p_gen = lp_gen/(lp_aca + lp_gen + lp_voc).
> compute p_voc = lp_voc/(lp_aca + lp_gen + lp_voc).
> compute p_aca = lp_aca/(lp_aca + lp_gen + lp_voc).
> compute p = {p_gen, p_aca, p_voc}.
> print p /title 'Predicted Probabilities for Outcomes 1 2 3 for ses 1 2 3 at mean of write'.
> End Matrix.*
>
>
Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email me.

"Nolite dare sanctum canibus neque mittatis margaritas vestras ante porcos ne forte conculcent eas pedibus suis."
Cum es damnatorum possederunt porcos iens ut salire off sanguinum cliff in abyssum?"


Annette Dobson’s _An introduction to generalized linear models_
may be a good reference to understand those matrix operations.
Juan Zuluaga (jzuluaga – stcloudstate.edu)
Research Analyst at the Office of Precollege Programs,
St. Cloud State University.


The point of this exercise is to better understand the syntax. The design matrix resonates, calculating log probabilities from the intercept and coefficients. I'll look into Annette Dobson’s _An introduction to generalized linear models
On Fri, Feb 15, 2013 at 4:39 AM, David Marso <[hidden email]> wrote:
What is the point of this whole exercise?
Does the concept 'design matrix' resonate?
When in doubt write out the math!

Peter Spangler wrote
> What is the role of the two columns of 0,1 combinations in the calculation
> of this probability matrix? .
>
> *{1, 0; 0, 1; 0, 0}}*
>
>
>
> On Tue, Feb 12, 2013 at 6:51 PM, Peter Spangler <
> pspangler@
> >wrote:
>
>> What is the function of the syntax highlighted in yellow? I am running
>> this Multinomial Probabilities matrix
>>
>> *Matrix.
>> * intercept1 intercept2 pared public gpa.
>> * these coefficients are taken from the output.
>> compute b_gen = {1.689354 ; 0.057928 ; 1.162832 ; 0.629541}.
>> compute b_voc = {4.235530 ; 0.113603 ; 0.982670 ; 1.274063}.
>> * overall design matrix including means of public and gpa.compute x = {{1
>> ; 1; 1}, make(3, 1, 52.775), {1, 0; 0, 1; 0, 0}}.compute lp_gen = exp(x *
>> b_gen).
>> compute lp_voc = exp(x * b_voc).
>> compute lp_aca = {1; 1; 1}.
>> compute p_gen = lp_gen/(lp_aca + lp_gen + lp_voc).
>> compute p_voc = lp_voc/(lp_aca + lp_gen + lp_voc).
>> compute p_aca = lp_aca/(lp_aca + lp_gen + lp_voc).
>> compute p = {p_gen, p_aca, p_voc}.
>> print p /title 'Predicted Probabilities for Outcomes 1 2 3 for ses 1 2 3
>> at mean of write'.
>> End Matrix.*
>>
>>

Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email me.

View this message in context: http://spssxdiscussion.1045642.n5.nabble.com/Syntaxtp5718048p5718080.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.
=====================
To manage your subscription to SPSSXL, send a message to
[hidden email] (not to SPSSXL), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSXL
For a list of commands to manage subscriptions, send the command
INFO REFCARD


Peter, Use of matrix algebra is an efficient way of estimating predicted probabilities from a generalized logit model. Let's forget about the matrix code for a moment...
Consider the following linear equation (eta_j) for a multinomial logistic regression:
eta_j = b0_j + b1_j*(x1) + ... + b2_j*(xk) With three levels of the response (y=1,2,3) and three independent variables (x1, x2, x3), we would have two linear equations:
eta_1 = b0_1 + b1_1*(x1) + b2_1*(x2) + b3_1*(x3) eta_2 = b0_2 + b1_2*(x1) + b2_2*(x2) + b3_2*(x3) ...and employing the inverse logit link function would produce the probabilities:
prob_1 = exp(eta_1) / (1 + exp(eta_1) + exp(eta_2)) prob_2 = exp(eta_2) / (1 + exp(eta_1) + exp(eta_2)) prob_3 = 1 / (1 + exp(eta_1) + exp(eta_2)) Now, let's apply what's stated above to your example using COMPUTE:
compute lp_gen_write_mean_ses1 = 1.689354  0.057928*(52.775) + 1.162832*(1) + 0.629541*(0). compute lp_gen_write_mean_ses2 = 1.689354  0.057928*(52.775) + 1.162832*(0) + 0.629541*(1). compute lp_gen_write_mean_ses3 = 1.689354  0.057928*(52.775) + 1.162832*(0) + 0.629541*(0).
compute lp_voc_write_mean_ses1 = 4.235530  0.113603*(52.775) + 0.982670*(1) + 1.274063*(0). compute lp_voc_write_mean_ses2 = 4.235530  0.113603*(52.775) + 0.982670*(0) + 1.274063*(1). compute lp_voc_write_mean_ses3 = 4.235530  0.113603*(52.775) + 0.982670*(0) + 1.274063*(0).
execute. *...and employing the inverse logit link function will produce the nine probabilities. compute p_gen_write_mean_ses1 = exp(lp_gen_write_mean_ses1) / (1 + exp(lp_gen_write_mean_ses1) + exp(lp_voc_write_mean_ses1)).
compute p_gen_write_mean_ses2 = exp(lp_gen_write_mean_ses2) / (1 + exp(lp_gen_write_mean_ses2) + exp(lp_voc_write_mean_ses2)). compute p_gen_write_mean_ses3 = exp(lp_gen_write_mean_ses3) / (1 + exp(lp_gen_write_mean_ses3) + exp(lp_voc_write_mean_ses3)).
compute p_voc_write_mean_ses1 = exp(lp_voc_write_mean_ses1) / (1 + exp(lp_gen_write_mean_ses1) + exp(lp_voc_write_mean_ses1)). compute p_voc_write_mean_ses2 = exp(lp_voc_write_mean_ses2) / (1 + exp(lp_gen_write_mean_ses2) + exp(lp_voc_write_mean_ses2)).
compute p_voc_write_mean_ses3 = exp(lp_voc_write_mean_ses3) / (1 + exp(lp_gen_write_mean_ses3) + exp(lp_voc_write_mean_ses3)). compute p_aca_write_mean_ses1 = 1 / (1 + exp(lp_gen_write_mean_ses1) + exp(lp_voc_write_mean_ses1)).
compute p_aca_write_mean_ses2 = 1 / (1 + exp(lp_gen_write_mean_ses2) + exp(lp_voc_write_mean_ses2)). compute p_aca_write_mean_ses3 = 1 / (1 + exp(lp_gen_write_mean_ses3) + exp(lp_voc_write_mean_ses3)). execute.
The SPSS code you posted is simply calculating these probabilities in a much more efficient way than my approach because it uses matrix algebra. That is, one obtains the estimated probabilities of each the three response levels from all possible combinations of the two SES indicator variables (1,0; 0,1; 0,0) with the covariate set at its grand mean (52.775) by employing the design matrix.
As David suggested, if you work out the actual math the matrix code is performing, it should become selfevident.
My hope is that the inelegant, but hopefully educational code I wrote will help you achieve clarity. Still, if you are unable to connect the code I wrote above with the code you posted, write back.
Ryan
On Tue, Feb 12, 2013 at 9:51 PM, Peter Spangler <[hidden email]> wrote:
What is the function of the syntax highlighted in yellow? I am running this Multinomial Probabilities matrix
Matrix.
* intercept1 intercept2 pared public gpa.
* these coefficients are taken from the output.
compute b_gen = {1.689354 ; 0.057928 ; 1.162832 ; 0.629541}.
compute b_voc = {4.235530 ; 0.113603 ; 0.982670 ; 1.274063}.
* overall design matrix including means of public and gpa.
compute x = {{1 ; 1; 1}, make(3, 1, 52.775), {1, 0; 0, 1; 0, 0}}.
compute lp_gen = exp(x * b_gen).
compute lp_voc = exp(x * b_voc).
compute lp_aca = {1; 1; 1}.
compute p_gen = lp_gen/(lp_aca + lp_gen + lp_voc).
compute p_voc = lp_voc/(lp_aca + lp_gen + lp_voc).
compute p_aca = lp_aca/(lp_aca + lp_gen + lp_voc).
compute p = {p_gen, p_aca, p_voc}.
print p /title 'Predicted Probabilities for Outcomes 1 2 3 for ses 1 2 3 at mean of write'.
End Matrix.

Administrator

Matrix multiplication:
A=
a b
c d
e f
{a,b;c,d;e,f}
X=
x
y
{x;y}
A*X
ax+by
cx+dy
ex+fy

Matrix addition:
A=
a b
c d
e f
X=
u v
w x
y z
Matrix addition:
a+u,b+v
c+w,d+x
e+y,f+z

Ryan Black wrote
Peter,
Use of matrix algebra is an efficient way of estimating predicted probabilities from a generalized logit model. Let's forget about the matrix code for a moment...
Consider the following linear equation (eta_j) for a multinomial logistic regression:
eta_j = b0_j + b1_j*(x1) + ... + b2_j*(xk)
With three levels of the response (y=1,2,3) and three independent variables (x1, x2, x3), we would have two linear equations:
eta_1 = b0_1 + b1_1*(x1) + b2_1*(x2) + b3_1*(x3)
eta_2 = b0_2 + b1_2*(x1) + b2_2*(x2) + b3_2*(x3)
...and employing the inverse logit link function would produce the probabilities:
prob_1 = exp(eta_1) / (1 + exp(eta_1) + exp(eta_2))
prob_2 = exp(eta_2) / (1 + exp(eta_1) + exp(eta_2))
prob_3 = 1 / (1 + exp(eta_1) + exp(eta_2))
Now, let's apply what's stated above to your example using COMPUTE:
compute lp_gen_write_mean_ses1 = 1.689354  0.057928*(52.775) + 1.162832*(1) + 0.629541*(0).
compute lp_gen_write_mean_ses2 = 1.689354  0.057928*(52.775) + 1.162832*(0) + 0.629541*(1).
compute lp_gen_write_mean_ses3 = 1.689354  0.057928*(52.775) + 1.162832*(0) + 0.629541*(0).
compute lp_voc_write_mean_ses1 = 4.235530  0.113603*(52.775) + 0.982670*(1) + 1.274063*(0).
compute lp_voc_write_mean_ses2 = 4.235530  0.113603*(52.775) + 0.982670*(0) + 1.274063*(1).
compute lp_voc_write_mean_ses3 = 4.235530  0.113603*(52.775) + 0.982670*(0) + 1.274063*(0).
execute.
*...and employing the inverse logit link function will produce the nine probabilities.
compute p_gen_write_mean_ses1 = exp(lp_gen_write_mean_ses1) / (1 + exp(lp_gen_write_mean_ses1) + exp(lp_voc_write_mean_ses1)).
compute p_gen_write_mean_ses2 = exp(lp_gen_write_mean_ses2) / (1 + exp(lp_gen_write_mean_ses2) + exp(lp_voc_write_mean_ses2)).
compute p_gen_write_mean_ses3 = exp(lp_gen_write_mean_ses3) / (1 + exp(lp_gen_write_mean_ses3) + exp(lp_voc_write_mean_ses3)).
compute p_voc_write_mean_ses1 = exp(lp_voc_write_mean_ses1) / (1 + exp(lp_gen_write_mean_ses1) + exp(lp_voc_write_mean_ses1)).
compute p_voc_write_mean_ses2 = exp(lp_voc_write_mean_ses2) / (1 + exp(lp_gen_write_mean_ses2) + exp(lp_voc_write_mean_ses2)).
compute p_voc_write_mean_ses3 = exp(lp_voc_write_mean_ses3) / (1 + exp(lp_gen_write_mean_ses3) + exp(lp_voc_write_mean_ses3)).
compute p_aca_write_mean_ses1 = 1 / (1 + exp(lp_gen_write_mean_ses1) + exp(lp_voc_write_mean_ses1)).
compute p_aca_write_mean_ses2 = 1 / (1 + exp(lp_gen_write_mean_ses2) + exp(lp_voc_write_mean_ses2)).
compute p_aca_write_mean_ses3 = 1 / (1 + exp(lp_gen_write_mean_ses3) + exp(lp_voc_write_mean_ses3)).
execute.
The SPSS code you posted is simply calculating these probabilities in a much more efficient way than my approach because it uses matrix algebra. That is, one obtains the estimated probabilities of each the three response levels from all possible combinations of the two SES indicator variables (1,0; 0,1; 0,0) with the covariate set at its grand mean (52.775) by employing the design matrix.
As David suggested, if you work out the actual math the matrix code is performing, it should become selfevident.
My hope is that the inelegant, but hopefully educational code I wrote will help you achieve clarity. Still, if you are unable to connect the code I wrote above with the code you posted, write back.
Ryan
On Tue, Feb 12, 2013 at 9:51 PM, Peter Spangler < [hidden email]> wrote:
> What is the function of the syntax highlighted in yellow? I am running this Multinomial Probabilities matrix
> Matrix.
> * intercept1 intercept2 pared public gpa.
> * these coefficients are taken from the output.
> compute b_gen = {1.689354 ; 0.057928 ; 1.162832 ; 0.629541}.
> compute b_voc = {4.235530 ; 0.113603 ; 0.982670 ; 1.274063}.
> * overall design matrix including means of public and gpa.
> compute x = {{1 ; 1; 1}, make(3, 1, 52.775), {1, 0; 0, 1; 0, 0}}.
> compute lp_gen = exp(x * b_gen).
> compute lp_voc = exp(x * b_voc).
> compute lp_aca = {1; 1; 1}.
> compute p_gen = lp_gen/(lp_aca + lp_gen + lp_voc).
> compute p_voc = lp_voc/(lp_aca + lp_gen + lp_voc).
> compute p_aca = lp_aca/(lp_aca + lp_gen + lp_voc).
> compute p = {p_gen, p_aca, p_voc}.
> print p /title 'Predicted Probabilities for Outcomes 1 2 3 for ses 1 2 3 at mean of write'.
> End Matrix.
Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email me.

"Nolite dare sanctum canibus neque mittatis margaritas vestras ante porcos ne forte conculcent eas pedibus suis."
Cum es damnatorum possederunt porcos iens ut salire off sanguinum cliff in abyssum?"


It is also worth noting that the line in question: compute x = {{1 ; 1; 1}, make(3, 1, 52.775), {1, 0; 0, 1; 0, 0}}. includes the SPSS "make(s1,s2,s3)" function which literally makes a matrix with a specified number of rows (s1), a specified number of columns (s2), and a specified value for each element (s3). So, for this example, the "make(s1,s2,s3)" function is creating a column vector of three rows (3 X 1 matrix) with element values set at 52.755.
Ryan Matrix multiplication:A=a bc de f{a,b;c,d;e,f}X=xy{x;y}A*Xax+bycx+dyex+fyMatrix addition:A=a bc de fX=u vw xy zMatrix addition:a+u,b+vc+w,d+xe+y,f+zRyan Black wrotePeter,
Use of matrix algebra is an efficient way of estimating predicted
probabilities from a generalized logit model. Let's forget about the
matrix code for a moment...
Consider the following linear equation (eta_j) for a multinomial logistic
regression:
eta_j = b0_j + b1_j*(x1) + ... + b2_j*(xk)
With three levels of the response (y=1,2,3) and three independent
variables (x1, x2, x3), we would have two linear equations:
eta_1 = b0_1 + b1_1*(x1) + b2_1*(x2) + b3_1*(x3)
eta_2 = b0_2 + b1_2*(x1) + b2_2*(x2) + b3_2*(x3)
...and employing the inverse logit link function would produce the
probabilities:
prob_1 = exp(eta_1) / (1 + exp(eta_1) + exp(eta_2))
prob_2 = exp(eta_2) / (1 + exp(eta_1) + exp(eta_2))
prob_3 = 1 / (1 + exp(eta_1) + exp(eta_2))
Now, let's apply what's stated above to your example using COMPUTE:
compute lp_gen_write_mean_ses1 = 1.689354  0.057928*(52.775) +
1.162832*(1) + 0.629541*(0).
compute lp_gen_write_mean_ses2 = 1.689354  0.057928*(52.775) +
1.162832*(0) + 0.629541*(1).
compute lp_gen_write_mean_ses3 = 1.689354  0.057928*(52.775) +
1.162832*(0) + 0.629541*(0).
compute lp_voc_write_mean_ses1 = 4.235530  0.113603*(52.775) +
0.982670*(1) + 1.274063*(0).
compute lp_voc_write_mean_ses2 = 4.235530  0.113603*(52.775) +
0.982670*(0) + 1.274063*(1).
compute lp_voc_write_mean_ses3 = 4.235530  0.113603*(52.775) +
0.982670*(0) + 1.274063*(0).
execute.
*...and employing the inverse logit link function will produce the nine
probabilities.
compute p_gen_write_mean_ses1 = exp(lp_gen_write_mean_ses1) / (1 +
exp(lp_gen_write_mean_ses1) + exp(lp_voc_write_mean_ses1)).
compute p_gen_write_mean_ses2 = exp(lp_gen_write_mean_ses2) / (1 +
exp(lp_gen_write_mean_ses2) + exp(lp_voc_write_mean_ses2)).
compute p_gen_write_mean_ses3 = exp(lp_gen_write_mean_ses3) / (1 +
exp(lp_gen_write_mean_ses3) + exp(lp_voc_write_mean_ses3)).
compute p_voc_write_mean_ses1 = exp(lp_voc_write_mean_ses1) / (1 +
exp(lp_gen_write_mean_ses1) + exp(lp_voc_write_mean_ses1)).
compute p_voc_write_mean_ses2 = exp(lp_voc_write_mean_ses2) / (1 +
exp(lp_gen_write_mean_ses2) + exp(lp_voc_write_mean_ses2)).
compute p_voc_write_mean_ses3 = exp(lp_voc_write_mean_ses3) / (1 +
exp(lp_gen_write_mean_ses3) + exp(lp_voc_write_mean_ses3)).
compute p_aca_write_mean_ses1 = 1 / (1 + exp(lp_gen_write_mean_ses1) +
exp(lp_voc_write_mean_ses1)).
compute p_aca_write_mean_ses2 = 1 / (1 + exp(lp_gen_write_mean_ses2) +
exp(lp_voc_write_mean_ses2)).
compute p_aca_write_mean_ses3 = 1 / (1 + exp(lp_gen_write_mean_ses3) +
exp(lp_voc_write_mean_ses3)).
execute.
The SPSS code you posted is simply calculating these probabilities in a
much more efficient way than my approach because it uses matrix algebra.
That is, one obtains the estimated probabilities of each the three
response levels from all possible combinations of the two SES indicator
variables (1,0; 0,1; 0,0) with the covariate set at its grand mean
(52.775) by employing the design matrix.
As David suggested, if you work out the actual math the matrix code is
performing, it should become selfevident.
My hope is that the inelegant, but hopefully educational code I wrote will
help you achieve clarity. Still, if you are unable to connect the code I
wrote above with the code you posted, write back.
Ryan
On Tue, Feb 12, 2013 at 9:51 PM, Peter Spangler <
pspangler@
> wrote:
What is the function of the syntax highlighted in yellow? I am running
this Multinomial Probabilities matrix
Matrix.
* intercept1 intercept2 pared public gpa.
* these coefficients are taken from the output.
compute b_gen = {1.689354 ; 0.057928 ; 1.162832 ; 0.629541}.
compute b_voc = {4.235530 ; 0.113603 ; 0.982670 ; 1.274063}.
* overall design matrix including means of public and gpa.
compute x = {{1 ; 1; 1}, make(3, 1, 52.775), {1, 0; 0, 1; 0, 0}}.
compute lp_gen = exp(x * b_gen).
compute lp_voc = exp(x * b_voc).
compute lp_aca = {1; 1; 1}.
compute p_gen = lp_gen/(lp_aca + lp_gen + lp_voc).
compute p_voc = lp_voc/(lp_aca + lp_gen + lp_voc).
compute p_aca = lp_aca/(lp_aca + lp_gen + lp_voc).
compute p = {p_gen, p_aca, p_voc}.
print p /title 'Predicted Probabilities for Outcomes 1 2 3 for ses 1 2 3
at mean of write'.
End Matrix.
Please reply to the list and not to my personal email.Those desiring my consulting or training services please feel free to email me.View this message in context: http://spssxdiscussion.1045642.n5.nabble.com/Syntaxtp5718048p5718095.htmlSent from the SPSSX Discussion mailing list archive at Nabble.com.=====================To manage your subscription to SPSSXL, send a message to[hidden email] (not to SPSSXL), with no body text except thecommand. To leave the list, send the commandSIGNOFF SPSSXLFor a list of commands to manage subscriptions, send the commandINFO REFCARD


I received a question offlist, so I think to be complete, I might as well just show the full X matrix: 1 52.755 1 0 1 52.755 0 1 1 52.755 0 0
The first column represents the place for the constant, the second column represents the meanvalue of the first predictor, the third column represents the value of the first SES indicator variable, and the fourth column represents the second SES indicator variable. (If there is confusion as to why a categorical predictor with three levels, "SES", is converted into two indicator variables coded 0/1, I suggest brushing up on the use of indicator/dummy coding to capture all the information from a categorical independent variable with more than two levels).
X is then multiplied by the corresponding set of estimated regression coefficients for y=1 ("b_gen") and y=2 ("b_voc"), exponentiated, and then converted to the nine probabilities using the formulas I presented in a previous post. Again, the matrix code posted by the OP is simply a more efficient way of obtaining the probabilities.
Ryan
On Sat, Feb 16, 2013 at 11:45 AM, <[hidden email]> wrote:
It is also worth noting that the line in question: compute x = {{1 ; 1; 1}, make(3, 1, 52.775), {1, 0; 0, 1; 0, 0}}.
includes the SPSS "make(s1,s2,s3)" function which literally makes a matrix with a specified number of rows (s1), a specified number of columns (s2), and a specified value for each element (s3). So, for this example, the "make(s1,s2,s3)" function is creating a column vector of three rows (3 X 1 matrix) with element values set at 52.755.
Ryan
Matrix multiplication:A=a bc de f{a,b;c,d;e,f}X=x
y{x;y}A*Xax+bycx+dyex+fyMatrix addition:A=
a bc de fX=u vw xy zMatrix addition:a+u,b+v
c+w,d+xe+y,f+zRyan Black wrotePeter,
Use of matrix algebra is an efficient way of estimating predicted
probabilities from a generalized logit model. Let's forget about the
matrix code for a moment...
Consider the following linear equation (eta_j) for a multinomial logistic
regression:
eta_j = b0_j + b1_j*(x1) + ... + b2_j*(xk)
With three levels of the response (y=1,2,3) and three independent
variables (x1, x2, x3), we would have two linear equations:
eta_1 = b0_1 + b1_1*(x1) + b2_1*(x2) + b3_1*(x3)
eta_2 = b0_2 + b1_2*(x1) + b2_2*(x2) + b3_2*(x3)
...and employing the inverse logit link function would produce the
probabilities:
prob_1 = exp(eta_1) / (1 + exp(eta_1) + exp(eta_2))
prob_2 = exp(eta_2) / (1 + exp(eta_1) + exp(eta_2))
prob_3 = 1 / (1 + exp(eta_1) + exp(eta_2))
Now, let's apply what's stated above to your example using COMPUTE:
compute lp_gen_write_mean_ses1 = 1.689354  0.057928*(52.775) +
1.162832*(1) + 0.629541*(0).
compute lp_gen_write_mean_ses2 = 1.689354  0.057928*(52.775) +
1.162832*(0) + 0.629541*(1).
compute lp_gen_write_mean_ses3 = 1.689354  0.057928*(52.775) +
1.162832*(0) + 0.629541*(0).
compute lp_voc_write_mean_ses1 = 4.235530  0.113603*(52.775) +
0.982670*(1) + 1.274063*(0).
compute lp_voc_write_mean_ses2 = 4.235530  0.113603*(52.775) +
0.982670*(0) + 1.274063*(1).
compute lp_voc_write_mean_ses3 = 4.235530  0.113603*(52.775) +
0.982670*(0) + 1.274063*(0).
execute.
*...and employing the inverse logit link function will produce the nine
probabilities.
compute p_gen_write_mean_ses1 = exp(lp_gen_write_mean_ses1) / (1 +
exp(lp_gen_write_mean_ses1) + exp(lp_voc_write_mean_ses1)).
compute p_gen_write_mean_ses2 = exp(lp_gen_write_mean_ses2) / (1 +
exp(lp_gen_write_mean_ses2) + exp(lp_voc_write_mean_ses2)).
compute p_gen_write_mean_ses3 = exp(lp_gen_write_mean_ses3) / (1 +
exp(lp_gen_write_mean_ses3) + exp(lp_voc_write_mean_ses3)).
compute p_voc_write_mean_ses1 = exp(lp_voc_write_mean_ses1) / (1 +
exp(lp_gen_write_mean_ses1) + exp(lp_voc_write_mean_ses1)).
compute p_voc_write_mean_ses2 = exp(lp_voc_write_mean_ses2) / (1 +
exp(lp_gen_write_mean_ses2) + exp(lp_voc_write_mean_ses2)).
compute p_voc_write_mean_ses3 = exp(lp_voc_write_mean_ses3) / (1 +
exp(lp_gen_write_mean_ses3) + exp(lp_voc_write_mean_ses3)).
compute p_aca_write_mean_ses1 = 1 / (1 + exp(lp_gen_write_mean_ses1) +
exp(lp_voc_write_mean_ses1)).
compute p_aca_write_mean_ses2 = 1 / (1 + exp(lp_gen_write_mean_ses2) +
exp(lp_voc_write_mean_ses2)).
compute p_aca_write_mean_ses3 = 1 / (1 + exp(lp_gen_write_mean_ses3) +
exp(lp_voc_write_mean_ses3)).
execute.
The SPSS code you posted is simply calculating these probabilities in a
much more efficient way than my approach because it uses matrix algebra.
That is, one obtains the estimated probabilities of each the three
response levels from all possible combinations of the two SES indicator
variables (1,0; 0,1; 0,0) with the covariate set at its grand mean
(52.775) by employing the design matrix.
As David suggested, if you work out the actual math the matrix code is
performing, it should become selfevident.
My hope is that the inelegant, but hopefully educational code I wrote will
help you achieve clarity. Still, if you are unable to connect the code I
wrote above with the code you posted, write back.
Ryan
On Tue, Feb 12, 2013 at 9:51 PM, Peter Spangler <
pspangler@
> wrote:
What is the function of the syntax highlighted in yellow? I am running
this Multinomial Probabilities matrix
Matrix.
* intercept1 intercept2 pared public gpa.
* these coefficients are taken from the output.
compute b_gen = {1.689354 ; 0.057928 ; 1.162832 ; 0.629541}.
compute b_voc = {4.235530 ; 0.113603 ; 0.982670 ; 1.274063}.
* overall design matrix including means of public and gpa.
compute x = {{1 ; 1; 1}, make(3, 1, 52.775), {1, 0; 0, 1; 0, 0}}.
compute lp_gen = exp(x * b_gen).
compute lp_voc = exp(x * b_voc).
compute lp_aca = {1; 1; 1}.
compute p_gen = lp_gen/(lp_aca + lp_gen + lp_voc).
compute p_voc = lp_voc/(lp_aca + lp_gen + lp_voc).
compute p_aca = lp_aca/(lp_aca + lp_gen + lp_voc).
compute p = {p_gen, p_aca, p_voc}.
print p /title 'Predicted Probabilities for Outcomes 1 2 3 for ses 1 2 3
at mean of write'.
End Matrix.
Please reply to the list and not to my personal email.Those desiring my consulting or training services please feel free to email me.
View this message in context: http://spssxdiscussion.1045642.n5.nabble.com/Syntaxtp5718048p5718095.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.=====================To manage your subscription to SPSSXL, send a message to
[hidden email] (not to SPSSXL), with no body text except thecommand. To leave the list, send the commandSIGNOFF SPSSXL
For a list of commands to manage subscriptions, send the commandINFO REFCARD


Okay, I have been so used to dummy coding variables without the use of syntax I didn't even think of the 0,1 function as such. So much faster!
I certainly would like to crush up on dummy variables with multiple levels. Any recommended texts?
Sent from my iPhone I received a question offlist, so I think to be complete, I might as well just show the full X matrix: 1 52.755 1 0 1 52.755 0 1 1 52.755 0 0
The first column represents the place for the constant, the second column represents the meanvalue of the first predictor, the third column represents the value of the first SES indicator variable, and the fourth column represents the second SES indicator variable. (If there is confusion as to why a categorical predictor with three levels, "SES", is converted into two indicator variables coded 0/1, I suggest brushing up on the use of indicator/dummy coding to capture all the information from a categorical independent variable with more than two levels).
X is then multiplied by the corresponding set of estimated regression coefficients for y=1 ("b_gen") and y=2 ("b_voc"), exponentiated, and then converted to the nine probabilities using the formulas I presented in a previous post. Again, the matrix code posted by the OP is simply a more efficient way of obtaining the probabilities.
Ryan
On Sat, Feb 16, 2013 at 11:45 AM, <[hidden email]> wrote:
It is also worth noting that the line in question: compute x = {{1 ; 1; 1}, make(3, 1, 52.775), {1, 0; 0, 1; 0, 0}}.
includes the SPSS "make(s1,s2,s3)" function which literally makes a matrix with a specified number of rows (s1), a specified number of columns (s2), and a specified value for each element (s3). So, for this example, the "make(s1,s2,s3)" function is creating a column vector of three rows (3 X 1 matrix) with element values set at 52.755.
Ryan
Matrix multiplication:A=a bc de f{a,b;c,d;e,f}X=x
y{x;y}A*Xax+bycx+dyex+fyMatrix addition:A=
a bc de fX=u vw xy zMatrix addition:a+u,b+v
c+w,d+xe+y,f+zRyan Black wrotePeter,
Use of matrix algebra is an efficient way of estimating predicted
probabilities from a generalized logit model. Let's forget about the
matrix code for a moment...
Consider the following linear equation (eta_j) for a multinomial logistic
regression:
eta_j = b0_j + b1_j*(x1) + ... + b2_j*(xk)
With three levels of the response (y=1,2,3) and three independent
variables (x1, x2, x3), we would have two linear equations:
eta_1 = b0_1 + b1_1*(x1) + b2_1*(x2) + b3_1*(x3)
eta_2 = b0_2 + b1_2*(x1) + b2_2*(x2) + b3_2*(x3)
...and employing the inverse logit link function would produce the
probabilities:
prob_1 = exp(eta_1) / (1 + exp(eta_1) + exp(eta_2))
prob_2 = exp(eta_2) / (1 + exp(eta_1) + exp(eta_2))
prob_3 = 1 / (1 + exp(eta_1) + exp(eta_2))
Now, let's apply what's stated above to your example using COMPUTE:
compute lp_gen_write_mean_ses1 = 1.689354  0.057928*(52.775) +
1.162832*(1) + 0.629541*(0).
compute lp_gen_write_mean_ses2 = 1.689354  0.057928*(52.775) +
1.162832*(0) + 0.629541*(1).
compute lp_gen_write_mean_ses3 = 1.689354  0.057928*(52.775) +
1.162832*(0) + 0.629541*(0).
compute lp_voc_write_mean_ses1 = 4.235530  0.113603*(52.775) +
0.982670*(1) + 1.274063*(0).
compute lp_voc_write_mean_ses2 = 4.235530  0.113603*(52.775) +
0.982670*(0) + 1.274063*(1).
compute lp_voc_write_mean_ses3 = 4.235530  0.113603*(52.775) +
0.982670*(0) + 1.274063*(0).
execute.
*...and employing the inverse logit link function will produce the nine
probabilities.
compute p_gen_write_mean_ses1 = exp(lp_gen_write_mean_ses1) / (1 +
exp(lp_gen_write_mean_ses1) + exp(lp_voc_write_mean_ses1)).
compute p_gen_write_mean_ses2 = exp(lp_gen_write_mean_ses2) / (1 +
exp(lp_gen_write_mean_ses2) + exp(lp_voc_write_mean_ses2)).
compute p_gen_write_mean_ses3 = exp(lp_gen_write_mean_ses3) / (1 +
exp(lp_gen_write_mean_ses3) + exp(lp_voc_write_mean_ses3)).
compute p_voc_write_mean_ses1 = exp(lp_voc_write_mean_ses1) / (1 +
exp(lp_gen_write_mean_ses1) + exp(lp_voc_write_mean_ses1)).
compute p_voc_write_mean_ses2 = exp(lp_voc_write_mean_ses2) / (1 +
exp(lp_gen_write_mean_ses2) + exp(lp_voc_write_mean_ses2)).
compute p_voc_write_mean_ses3 = exp(lp_voc_write_mean_ses3) / (1 +
exp(lp_gen_write_mean_ses3) + exp(lp_voc_write_mean_ses3)).
compute p_aca_write_mean_ses1 = 1 / (1 + exp(lp_gen_write_mean_ses1) +
exp(lp_voc_write_mean_ses1)).
compute p_aca_write_mean_ses2 = 1 / (1 + exp(lp_gen_write_mean_ses2) +
exp(lp_voc_write_mean_ses2)).
compute p_aca_write_mean_ses3 = 1 / (1 + exp(lp_gen_write_mean_ses3) +
exp(lp_voc_write_mean_ses3)).
execute.
The SPSS code you posted is simply calculating these probabilities in a
much more efficient way than my approach because it uses matrix algebra.
That is, one obtains the estimated probabilities of each the three
response levels from all possible combinations of the two SES indicator
variables (1,0; 0,1; 0,0) with the covariate set at its grand mean
(52.775) by employing the design matrix.
As David suggested, if you work out the actual math the matrix code is
performing, it should become selfevident.
My hope is that the inelegant, but hopefully educational code I wrote will
help you achieve clarity. Still, if you are unable to connect the code I
wrote above with the code you posted, write back.
Ryan
On Tue, Feb 12, 2013 at 9:51 PM, Peter Spangler <
pspangler@
> wrote:
What is the function of the syntax highlighted in yellow? I am running
this Multinomial Probabilities matrix
Matrix.
* intercept1 intercept2 pared public gpa.
* these coefficients are taken from the output.
compute b_gen = {1.689354 ; 0.057928 ; 1.162832 ; 0.629541}.
compute b_voc = {4.235530 ; 0.113603 ; 0.982670 ; 1.274063}.
* overall design matrix including means of public and gpa.
compute x = {{1 ; 1; 1}, make(3, 1, 52.775), {1, 0; 0, 1; 0, 0}}.
compute lp_gen = exp(x * b_gen).
compute lp_voc = exp(x * b_voc).
compute lp_aca = {1; 1; 1}.
compute p_gen = lp_gen/(lp_aca + lp_gen + lp_voc).
compute p_voc = lp_voc/(lp_aca + lp_gen + lp_voc).
compute p_aca = lp_aca/(lp_aca + lp_gen + lp_voc).
compute p = {p_gen, p_aca, p_voc}.
print p /title 'Predicted Probabilities for Outcomes 1 2 3 for ses 1 2 3
at mean of write'.
End Matrix.
Please reply to the list and not to my personal email.Those desiring my consulting or training services please feel free to email me.
View this message in context: http://spssxdiscussion.1045642.n5.nabble.com/Syntaxtp5718048p5718095.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.=====================To manage your subscription to SPSSXL, send a message to
[hidden email] (not to SPSSXL), with no body text except thecommand. To leave the list, send the commandSIGNOFF SPSSXL
For a list of commands to manage subscriptions, send the commandINFO REFCARD

