surveillance: hhh4_simulate – R documentation

Become an expert in R — Interactive courses, Cheat Sheets, certificates and more!

hhh4_simulate

Simulate "hhh4" Count Time Series

Description

Simulates a multivariate time series of counts based on the Poisson/Negative Binomial model as described in Paul and Held (2011).

Usage

## S3 method for class 'hhh4'
simulate(object, nsim = 1, seed = NULL, y.start = NULL,
         subset = 1:nrow(object$stsObj), coefs = coef(object),
         components = c("ar","ne","end"), simplify = nsim>1, ...)

Arguments

`object`	an object of class `"hhh4"`.
`nsim`	number of time series to simulate. Defaults to `1`.
`seed`	an object specifying how the random number generator should be initialized for simulation (via `set.seed`). The initial state will also be stored as an attribute `"seed"` of the result. The original state of the `.Random.seed` will be restored at the end of the simulation. By default (`NULL`), neither initialization nor recovery will be done. This behaviour is copied from the `simulate.lm` method.
`y.start`	vector or matrix (with `ncol(object$stsObj)` columns) with starting counts for the epidemic components. If `NULL`, the observed means in the respective units of the data in `object` during `subset` are used.
`subset`	time period in which to simulate data. Defaults to (and cannot exceed) the whole period defined by the underlying `"sts"` object.
`coefs`	coefficients used for simulation from the model in `object`. Default is to use the fitted parameters. Note that the `coefs`-vector must be in the same order and scaling as `coef(object)`, which especially means `reparamPsi = TRUE` (as per default when using the `coef`-method to extract the parameters). The overdispersion parameter in `coefs` is the inverse of the dispersion parameter `size` in `rnbinom`.
`components`	character vector indicating which components of the fitted model `object` should be active during simulation. For instance, a simulation with `components="end"` is solely based on the fitted endemic mean.
`simplify`	logical indicating if only the simulated counts (`TRUE`) or the full `"sts"` object (`FALSE`) should be returned for every replicate. By default a full `"sts"` object is returned iff `nsim=1`.
`...`	unused (argument of the generic).

Details

Simulates data from a Poisson or a Negative Binomial model with mean

μ_it = λ_it y_i,t-1 + φ_it ∑_j w_ji y_j,t-1 + ν_it

where λ_{it}>0, φ_{it}>0, and ν_{it}>0 are parameters which are modelled parametrically. The function uses the model and parameter estimates of the fitted object to simulate the time series.

With the argument coefs it is possible to simulate from the model as specified in object, but with different parameter values.

Value

If simplify=FALSE: an object of class "sts" (nsim = 1) or a list of those (nsim > 1).

If simplify=TRUE: an object of class "hhh4sims", which is an array of dimension c(length(subset), ncol(object$stsObj), nsim), where the third dimension is dropped if nsim=1 (yielding a matrix). The originally observed counts during the simulation period, object$stsObj[subset,], are attached for reference (used by the plot-methods) as an attribute "stsObserved", and the initial condition y.start as attribute "initial".

Author(s)

Michaela Paul and Sebastian Meyer

References

Paul, M. and Held, L. (2011) Predictive assessment of a non-linear random effects model for multivariate time series of infectious disease counts. Statistics in Medicine, 30, 1118–1136

Examples

data(influMen)
# convert to sts class and extract meningococcal disease time series
meningo <- disProg2sts(influMen)[,2]

# fit model
fit <- hhh4(meningo, control = list(ar = list(f = ~ 1),
            end = list(f = addSeason2formula(S = 1, period = 52)),
            family = "NegBin1"))
plot(fit)

# simulate from model
simData <- simulate(fit, seed=1234)

# plot simulated data
plot(simData, main = "simulated data", xaxis.labelFormat=NULL)

# consider a Poisson instead of a NegBin model
coefs <- coef(fit)
coefs["overdisp"] <- 0

simData2 <- simulate(fit, seed=123, coefs = coefs)
plot(simData2, main = "simulated data: Poisson model", xaxis.labelFormat = NULL)

# consider a model with higher autoregressive parameter
coefs <- coef(fit)
coefs[1] <- log(0.5)

simData3 <- simulate(fit, seed=321, coefs = coefs)
plot(simData3, main = "simulated data: lambda = 0.5", xaxis.labelFormat = NULL)

surveillance

Temporal and Spatio-Temporal Modeling and Monitoring of Epidemic Phenomena

v1.19.1

GPL-2

Authors

Michael H<f6>hle [aut, ths] (<https://orcid.org/0000-0002-0423-6702>), Sebastian Meyer [aut, cre] (<https://orcid.org/0000-0002-1791-9449>), Michaela Paul [aut], Leonhard Held [ctb, ths], Howard Burkom [ctb], Thais Correa [ctb], Mathias Hofmann [ctb], Christian Lang [ctb], Juliane Manitz [ctb], Andrea Riebler [ctb], Daniel Saban<e9>s Bov<e9> [ctb], Ma<eb>lle Salmon [ctb], Dirk Schumacher [ctb], Stefan Steiner [ctb], Mikko Virtanen [ctb], Wei Wei [ctb], Valentin Wimmer [ctb], R Core Team [ctb] (A few code segments are modified versions of code from base R)

Initial release

2021-03-30