Longitudinal Data, All Data Types, Conditional Specification

Can create cross-sectional and longitudinal data with this. Next step is do a more careful walk-through.

Cross Sectional Data

First load the R packages

# Requires these R packages:
library(SPR)
library(glmmTMB)
library(lme4)
library(ordinal)

Create cross sectional data by

• set number of timepoints equal to 1
• set the variance of the random intercept equal to 0
• pass a regression formula that does not contain time as a variable

#----
# Check if we can create cross-sectional data:
sim.out <- SPR::sim_dat_types(N=5000, 
                              data.type = 'Ordinal',
                              reg.formula =  formula(~Group),
                              subject.var = 0, 
                              number.timepoints = 1)
  
dat <- sim.out$dat
modt1 <- ordinal::clm(ordered(Y_ord) ~ Group, data = dat)
summary(modt1)

Longitudinal data

Looks like we can create cross-sectional data just fine. Let’s try some longitudinal data

With this function, the defaults are:

• number of timepoints is 4
• regression formula is ~ Group + Time + Group*Time
• random intercept variance (subject variance) is subject.var = 1

You can use those or change ’em.

#---
# ZINB
sim.out <- SPR::sim_dat_types(N=2000,
                         data.type = 'ZINB',
                         reg.formula = formula(~ Time))
  dat <- sim.out$dat
mod <- glmmTMB::glmmTMB(Y_zinb ~ Time + (1|USUBJID),
                ziformula = ~ 1,
                data = dat,
                family = nbinom2(link = "log"))
summary(mod)

#--------
# ZIP
sim.out <- SPR::sim_dat_types(N=2000,
                         data.type = 'ZIP',
                         reg.formula = formula(~ Time))
  dat <- sim.out$dat
mod <- glmmTMB(Y_zip ~ Time + (1|USUBJID),
                ziformula = ~ 1,
                data = dat,
                family = poisson)
summary(mod)

#--------------
# Gaussian
library(lme4)
sim.out <- SPR::sim_dat_types(data.type = 'Gaussian')
  dat <- sim.out$dat
mod1 <- lme4::lmer(Y_gaussian ~ Group + Time + Group*Time + (1|USUBJID),
             data = dat)
summary(mod1)

#---
# Gaussian - aligns with lme4
sim.out <- SPR::sim_dat_types(data.type = 'Gaussian')
  dat <- sim.out$dat
mod0 <- glmmTMB(Y_gaussian ~ Group + Time + Group*Time + (1|USUBJID),
                data = dat,
                family = gaussian)
summary(mod0)


#---
# Poisson
sim.out <- SPR::sim_dat_types(data.type = 'Poisson')
  dat <- sim.out$dat
mod1 <- glmmTMB(Y_pois ~ Group + Time + Group*Time + (1|USUBJID),
                data = dat,
                family = poisson)
summary(mod1)

#---
# Negative Binomial
sim.out <- SPR::sim_dat_types(data.type = 'NegBinom')
  dat <- sim.out$dat
mod2 <- glmmTMB(Y_nb ~ Group + Time + Group*Time + (1|USUBJID),
                data = dat,
                family = nbinom2(link = "log"))
summary(mod2)



#---
# Beta distribution
sim.out <- SPR::sim_dat_types(data.type = 'Beta')
  dat <- sim.out$dat
mod3 <- glmmTMB(Y_beta ~ Group + Time + Group*Time + (1|USUBJID),
                data = dat,
                family = beta_family(link = "logit"))
summary(mod3)



library(ordinal)
#---
# Ordinal
sim.out <- SPR::sim_dat_types(data.type = 'Ordinal')
  dat <- sim.out$dat
mod4 <- clmm(ordered(Y_ord) ~ Group + Time + Group*Time + (1|USUBJID),
             data = dat)
summary(mod4)
sim.out$thresholds
sim.out$Beta


#---
# Nominal data
sim.out <- sim_dat_types(N = 5000,
                         data.type = 'Multinom',
                         reg.formula =  formula(~ Time) )
  dat <- sim.out$dat

library(nnet)
dat$Y_nom_factor <- as.factor(dat$Y_nom)
mod <- nnet:::multinom(Y_nom_factor ~ Time, data = dat)
summary(mod)
t(coef(mod))
sim.out$Beta