R help - Nov 2007 - constraint matrices in vglm (VGAM package)

Hello R users,

 

I am performing a multinomial logit regression and would like to
constrain a few model coefficients to be equal.  

 

Here is my model:

multi <- vglm(case123con ~ SNP_A1+SNP_A2+age, multinomial, work.analy)

where case123con is a four level categorical variable (case 1, case 2,
case 3, control) and SNP_A1 and SNP_A2 are indicator functions (yes/no).

 

The output of this model is:

Coefficients:

                          Value Std. Error    t value

(Intercept):1        6.9798044  0.8145521  8.5688866

(Intercept):2        2.5729346  0.8733182  2.9461592

(Intercept):3        0.0129745  1.0966651  0.0118308

age:1                 -0.0468339  0.0087766 -5.3362053

age:2                -0.0306066  0.0091644 -3.3397314

age:3                -0.0058718  0.0114316 -0.5136443

SNP_A1:1          0.0284303  0.1590333  0.1787698

SNP_A1:2          0.1685160  0.1674925  1.0061104

SNP_A1:3          0.0137997  0.2052559  0.0672315

SNP_A2:1         -0.6265717  0.2694583 -2.3253010

SNP_A2:2         -0.2528873  0.2790632 -0.9062006

SNP_A2:3         -0.0789181  0.3339153 -0.2363418

 

I would like to constrain the model coefficients so that SNP_A1:1 SNP_A1:2 =
SNP_A1:3 and SNP_A2:1 = SNP_A2:2 = SNP_A2:3.  The reason I
want to do this is because I want to test the null hypothesis that these
coefficients are equal through a likelihood ratio test.  Can I do this
by defining the constraint matrices and if so, how do I go about doing
so? 

 

Any help or tips would be greatly appreciated! Thank you in advance!!

Lin



**NOTA DE CONFIDENCIALIDAD** Este correo electrÿnico, ...{{dropped:18}}

Hi Lin,

try

multi2 = vglm(case123con ~ SNP_A1+SNP_A2+age,
              multinomial(parallel = TRUE ~ SNP_A1+SNP_A2 - 1),
              work.analy)

or

multi3 = vglm(case123con ~ SNP_A1+SNP_A2+age,
              multinomial(parallel = FALSE ~ 1 + age),
              work.analy)

After your fit, typing coef(multi2, mat=TRUE) is a way of checking.
Also constraints(multi2).

Further info is at
http://www.stat.auckland.ac.nz/~yee/VGAM/doc/constraints.pdf
More generally, one can use the constraints argument to pass in a
list with explicit 'constraint matrices'; the parallel argument
here is just a more convenient way to set them up.

cheers
Thomas



 > Hello R users,
 >
 > I am performing a multinomial logit regression and would like to
 > constrain a few model coefficients to be equal. 
 >
 > Here is my model:
 >
 > multi <- vglm(case123con ~ SNP_A1+SNP_A2+age, multinomial, work.analy)
 >
 > where case123con is a four level categorical variable (case 1, case 2,
 > case 3, control) and SNP_A1 and SNP_A2 are indicator functions (yes/no).
 >
 > The output of this model is:
 >
 > Coefficients:
 >
 >                           Value Std. Error    t value
 >
 > (Intercept):1        6.9798044  0.8145521  8.5688866
 > (Intercept):2        2.5729346  0.8733182  2.9461592
 > (Intercept):3        0.0129745  1.0966651  0.0118308
 > age:1                 -0.0468339  0.0087766 -5.3362053
 > age:2                -0.0306066  0.0091644 -3.3397314
 > age:3                -0.0058718  0.0114316 -0.5136443
 > SNP_A1:1          0.0284303  0.1590333  0.1787698
 > SNP_A1:2          0.1685160  0.1674925  1.0061104
 > SNP_A1:3          0.0137997  0.2052559  0.0672315
 > SNP_A2:1         -0.6265717  0.2694583 -2.3253010
 > SNP_A2:2         -0.2528873  0.2790632 -0.9062006
 > SNP_A2:3         -0.0789181  0.3339153 -0.2363418
 >
 > I would like to constrain the model coefficients so that SNP_A1:1  >
SNP_A1:2 = SNP_A1:3 and SNP_A2:1 = SNP_A2:2 = SNP_A2:3.  The reason I
 > want to do this is because I want to test the null hypothesis that these
 > coefficients are equal through a likelihood ratio test.  Can I do this
 > by defining the constraint matrices and if so, how do I go about doing
 > so?
 >
 > Any help or tips would be greatly appreciated! Thank you in advance!!
 >
 > Lin
 >