Day 2 : Model Modifications

[andkov]

Modification 1

During the morning discussion we have identified the following model to be tested. Either model A or B can be modified in this way, however we start with model B.

The msm call should be modified in the following way:

• transitions to the death state (1-4, 2-4, 3-4) must be adjusted for age and sex
• backward transition (2-1) to be adjusted only for age
• forward transitions (1-2, 2-3) to be adjusted by
  • case 1 : age, sex, education, ses
  • case 2 : age, sex, education

The linking of the covariates to the specific transition is implemented in the covariates argument of the msm function:

covariates = list(
"1-2" ~  = covariates,
"2-1" `~ = age,
...
)

[andkov]

[andkov]

Results for Modification 1

The first modification of models has been estimated : html (more source code) + pdf (less source code):

Form specification ( model B)

# transition matrix
Q <- rbind( c(0, q, 0, q), 
            c(q, 0, q, q),
            c(0, 0, 0, q), 
            c(0, 0, 0, 0)) 
# misclassification matrix
E <- rbind( c( 0,  0,  0, 0),  
            c( 0,  0, .1, 0), 
            c( 0,  0,  0, 0),
            c( 0,  0,  0, 0) )

Covariate specifications:

Model 1:

# Forward transitions:
  "1-2"       = "age + male + educat"
  "2-3"       = "age + male + educat"
# Backward transitions:
  "2-1"       = "age"
# Death transitions: 
  "1-4"       = "age + male"
  "2-4"       = "age + male"
  "3-4"       = "age + male"

Model 2 :

# Forward transitions:
  "1-2"       = "age + age_bl + male + educat"
  "2-3"       = "age + age_bl + male + educat"
# Backward transitions:
  "2-1"       = "age + age_bl"
# Death transitions: 
  "1-4"       = "age + age_bl + male"
  "2-4"       = "age + age_bl + male"
  "3-4"       = "age + age_bl + male"

@GracielaMuniz @smhofer

[andkov]

Modification 2

In order to match the tables from Marioni it was decided to transform education into two dummy variables that would encode comparison between education groups.

The new model fitting in msm call should be modified in the following way:

• transitions to the death state (1-4, 2-4, 3-4) must be adjusted for age , sex , and edu_low_med',edu_low_high`
• backward transition (2-1) to be adjusted only for age
• forward transitions (1-2, 2-3) to be adjusted by
  • case 1 : age, sex, edu_low_med',edu_low_high,ses`
  • case 2 : age, sex, edu_low_med',edu_low_high`

The construction of the dummies for education:

# ref  D1  D2
#  -1  0   0
#  0   1   0
#  1   0   1

• D1 = edu_low_med - the effect of medium education (10, 11 years of edu) vs low education ( less than 10 years of education)
• D2 = edu_low_high - the effect of high education ( greater than 11 years of education) vs low education (less than 10 years of education)

[andkov]

Results from Modification 2

The second modification of models has been estimated :

• HTML (more source code)
• PDF (less source code).
• WORD (no source code. contains only intensities and hazards) :

Form specification ( model B)

# transition matrix
Q <- rbind( c(0, q, 0, q), 
            c(q, 0, q, q),
            c(0, 0, 0, q), 
            c(0, 0, 0, 0)) 
# misclassification matrix
E <- rbind( c( 0,  0,  0, 0),  
            c( 0,  0, .1, 0), 
            c( 0,  0,  0, 0),
            c( 0,  0,  0, 0) )

Covariate specifications:

Model 1:

# Forward transitions:
  "1-2"       = "age + male  + edu_low_med + edu_low_high"
  "2-3"       = "age + male  + edu_low_med + edu_low_high"
# Backward transitions:
  "2-1"       = "age"
# Death transitions: 
  "1-4"       = "age + male"
  "2-4"       = "age + male"
  "3-4"       = "age + male"

Model 2 :

# Forward transitions:
  "1-2"       = "age + male  + edu_low_med + edu_low_high"
  "2-3"       = "age + male  + edu_low_med + edu_low_high"
# Backward transitions:
  "2-1"       = "age"
# Death transitions: 
  "1-4"       = "age + male + edu_low_med + edu_low_high"
  "2-4"       = "age + male + edu_low_med + edu_low_high"
  "3-4"       = "age + male + edu_low_med + edu_low_high"

@GracielaMuniz @smhofer @annierobi