bwVI: Black White Visual Impairment data set

In swihart/gnlrim: Generalized Non-Linear Random Intercept Models

Black White Visual Impairment data set

Description

Visual impairment dataset acquired from supplementary materials of https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4203426/

Usage

bwVI

Format

A data frame with 10398 rows and 4 variables for 5199 unique patients:

id

subject id

black

1 if subject was black; 0 if white

variable

eye1 for left eye, eye2 for right eye

value

1 if the eye is impaired, 0 if healthy

Details

References

- Tielsch JM, Sommer A, Katz J, Quigley H, Ezrine S. Socioeconomic status and visual impairment among urban Americans. Archives of ophthalmology. 1991;109:637–641.

- Liang KY, Zeger SL. Longitudinal data analysis using generalized linear models. Biometrika. 1986;73:13–22.

- Swihart, B. J., Caffo, B. S. and Crainiceanu, C. M. (2014) A unifying framework for marginalised random-476 intercept models of correlated binary outcomes. International Statistical Review, 82, 275–295

Source

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4203426/

Examples

## Example 3.5 from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4203426/
attach(bwVI)
head(bwVI)
o.value <- value ## lesson learned!  got an error when I used "value"
(LLB <-
    gnlrim::gnlrim(y=cbind(o.value, 1-o.value),
                   mu = ~ plogis(Intercept + black*b_p + rand1),
                   pmu = c(Intercept = -4.88, b_p=0.134),
                   pmix=c(var=2.03),
                   p_uppb = c(  50,   9, 20.00),
                   p_lowb = c( -50,  -9,  0.05),
                   distribution="binomial",
                   nest=id,
                   random=c("rand1"),
                   mixture="logit-bridge-var",
                   ooo=TRUE,
                   compute_hessian = FALSE,
                   compute_kkt = FALSE,
                   trace=1,
                   method='nlminb',
    )
)
# > head(bwVI)
# id black variable value
# 1  1     1     eye1     1
# 2  1     1     eye2     1
# 3  2     1     eye1     1
# 4  2     1     eye2     1
# 5  3     1     eye1     1
# 6  3     1     eye2     1
# > o.value <- value ## lesson learned!  got an error when I used "value"
# > (LLB <-
#      +     gnlrim::gnlrim(y=cbind(o.value, 1-o.value),
#                           +                    mu = ~ plogis(Intercept + black*b_p + rand1),
#                           +                    pmu = c(Intercept = -4.88, b_p=0.134),
#                           +                    pmix=c(var=2.03),
#                           +                    p_uppb = c(  50,   9, 20.00),
#                           +                    p_lowb = c( -50,  -9,  0.05),
#                           +                    distribution="binomial",
#                           +                    nest=id,
#                           +                    random=c("rand1"),
#                           +                    mixture="logit-bridge-var",
#                           +                    ooo=TRUE,
#                           +                    compute_hessian = FALSE,
#                           +                    compute_kkt = FALSE,
#                           +                    trace=1,
#                           +                    method='nlminb',
#                           +     )
#    + )
# [1] 3
# Intercept       b_p       var
# -4.880     0.134     2.030
# [1] 3119.63
# fn is  fn
# Looking for method =  nlminb
# Function has  3  arguments
# par[ 1 ]:  -50   <? -4.88   <? 50     In Bounds
# par[ 2 ]:  -9   <? 0.134   <? 9     In Bounds
# par[ 3 ]:  0.05   <? 2.03   <? 20     In Bounds
# Analytic gradient not made available.
# Analytic Hessian not made available.
# Scale check -- log parameter ratio= 1.561315   log bounds ratio= 0.7447275
# Method:  nlminb
# 0:     3119.6299: -4.88000 0.134000  2.03000
# 1:     2791.2694: -4.06317 0.507976  2.46924
# 2:     2743.1744: -3.16525 0.142932  2.71519
# 3:     2729.2153: -3.48196 -0.400397  3.49268
# 4:     2710.6152: -3.83306 0.175929  4.23063
# 5:     2709.7961: -3.77051 0.184101  4.28020
# 6:     2709.2022: -3.80107 0.130731  4.33171
# 7:     2708.0638: -3.76409 0.123961  4.48770
# 8:     2702.5219: -3.97565 0.0273767  5.52687
# 9:     2700.4295: -4.24106 0.170311  6.49712
# 10:     2700.0148: -4.28328 0.109940  6.91920
# 11:     2699.9724: -4.30180 0.104920  7.08190
# 12:     2699.9702: -4.30653 0.106829  7.11938
# 13:     2699.9701: -4.30680 0.107369  7.12115
# 14:     2699.9701: -4.30677 0.107430  7.12092
# Post processing for method  nlminb
# Successful convergence!
#   Save results from method  nlminb
# $par
# Intercept        b_p        var
# -4.3067653  0.1074302  7.1209235
#
# $message
# [1] "relative convergence (4)"
#
# $convcode
# [1] 0
#
# $value
# [1] 2699.97
#
# $fevals
# function
# 18
#
# $gevals
# gradient
# 51
#
# $nitns
# [1] 14
#
# $kkt1
# [1] NA
#
# $kkt2
# [1] NA
#
# $xtimes
# user.self
# 752.63
#
# Assemble the answers
# Intercept       b_p      var   value fevals gevals niter convcode kkt1
# nlminb -4.306765 0.1074302 7.120923 2699.97     18     51    14        0   NA
# kkt2  xtime
# nlminb   NA 752.63

swihart/gnlrim documentation built on Oct. 18, 2023, 8:29 p.m.