Simulate a long-format vital registration (VR) dataset with a latent mortality process and an explicit observation (reporting) process that can collapse after a conflict start time.
The simulator generates:
latent cause-specific mortality rates (by region, time, age, sex),
true deaths given population denominators,
observed VR counts given cause- and age-specific reporting completeness,
optional misclassification between trauma and non-trauma,
optional additional missingness mechanisms to stress test inference.
Usage
vrc_simulate(
R = 5,
T = 120,
age_breaks = c(0, 5, 15, 25, 35, 45, 55, 65, Inf),
sexes = c("F", "M"),
t0 = floor(T/2),
seed = 1,
missing = list(type = "none", block_intercept = -2.5, block_conflict_coef = 0.8,
block_facility_coef = -1, age_dropout_strength = 0, age_dropout_old_from = 6,
age_dropout_post_only = TRUE, mnar_strength = 0, missing_to_zero = FALSE),
pop_total_range = c(3e+05, 1200000),
pop_error_sd = 0.05,
displacement_post_sd = 0.02,
conflict_post_mean = 1,
conflict_post_sd = 0.6,
conflict_spike_prob = 0.08,
conflict_spike_mean = 2,
facility_pre_mean = 0.9,
facility_post_drop = 0.35,
facility_noise_sd = 0.08,
beta_conf_true = c(1, 0.2),
beta_conf_region_sd = c(0, 0),
beta_fac_mort_true = c(0, -0.1),
sigma_u_lambda_true = c(0.35, 0.25),
sigma_v_lambda_true = c(0.08, 0.05),
sigma_v_lambda_region_true = c(0, 0),
phi_true = c(30, 60),
rho0_true = c(0.7, 0.9),
kappa_post_true = c(0, -1),
gamma_conf_true = c(-0.1, -0.5),
gamma_conf_region_sd = c(0, 0),
gamma_fac_true = c(0.6, 0.8),
sigma_u_rho_true = c(0.4, 0.35),
sigma_v_rho_true = c(0.12, 0.08),
sigma_v_rho_region_true = c(0, 0),
age_penalty_non = NULL,
misclass = list(p_non_to_trauma = 0.04, p_trauma_to_non = 0.01)
)Arguments
- R
Number of regions.
- T
Number of time points (for example, months).
- age_breaks
Numeric vector of age breakpoints. Defaults to 8 age groups.
- sexes
Character vector of sex labels.
- t0
Integer time index at which conflict begins.
- seed
Random seed.
- missing
A list controlling additional missingness mechanisms. See Details.
- pop_total_range
Length-2 numeric vector giving the range of total region populations to sample from.
- pop_error_sd
Numeric. Lognormal measurement error (standard deviation on the log scale) applied to the population denominator used for inference.
- displacement_post_sd
Numeric. Random-walk volatility controlling region population scaling after conflict (approximates displacement).
- conflict_post_mean
Numeric. Mean of the conflict covariate after conflict start.
- conflict_post_sd
Numeric. Standard deviation of the conflict covariate after conflict start.
- conflict_spike_prob
Numeric in (0,1). Probability of an additional conflict spike at a given time point after conflict start.
- conflict_spike_mean
Numeric. Mean spike size (gamma-distributed).
- facility_pre_mean
Numeric. Mean facility functioning score pre-conflict (bounded to 0-1).
- facility_post_drop
Numeric. Average drop in facility functioning at conflict start.
- facility_noise_sd
Numeric. Noise SD used when generating facility functioning.
- beta_conf_true
Length-2 numeric vector. True effect of the conflict covariate on log mortality rates for trauma and non-trauma.
- beta_conf_region_sd
Length-2 numeric vector. Standard deviation of region-level deviations for the conflict effect in the mortality model.
- beta_fac_mort_true
Length-2 numeric vector. True effect of facility functioning on log mortality rates.
- sigma_u_lambda_true
Length-2 numeric vector. True SD of region random intercepts in the mortality model.
- sigma_v_lambda_true
Length-2 numeric vector. True SD of national RW1 time effects in the mortality model.
- sigma_v_lambda_region_true
Length-2 numeric vector. True SD of region-specific RW1 deviations around the national time effect.
- phi_true
Length-2 numeric vector. NB2 size (overdispersion) parameters for trauma and non-trauma deaths.
- rho0_true
Length-2 numeric vector. Baseline (pre-conflict) reporting completeness for trauma and non-trauma.
- kappa_post_true
Length-2 numeric vector. Post-conflict shift on the logit reporting scale for trauma and non-trauma.
- gamma_conf_true
Length-2 numeric vector. True effect of conflict on reporting completeness (logit scale).
- gamma_conf_region_sd
Length-2 numeric vector. Standard deviation of region-level deviations for the conflict effect in the reporting model.
- gamma_fac_true
Length-2 numeric vector. True effect of facility functioning on reporting completeness (logit scale).
- sigma_u_rho_true
Length-2 numeric vector. True SD of region random intercepts in the reporting model.
- sigma_v_rho_true
Length-2 numeric vector. True SD of national RW1 time effects in the reporting model.
- sigma_v_rho_region_true
Length-2 numeric vector. True SD of region-specific RW1 deviations around the national reporting time effect.
- age_penalty_non
Optional numeric vector of length equal to the number of age groups. If provided, gives the monotone age-selective reporting penalty for non-trauma post-conflict.
- misclass
A list with elements
p_non_to_traumaandp_trauma_to_nongiving misclassification probabilities among recorded deaths.
Value
A list with elements:
df_full: full long-format dataset (may containNAiny).df_obs: subset of rows with observedy.truth: latent arrays and parameters used for simulation.meta: dimension metadata.
Details
The missing argument allows additional missingness beyond the reporting
process. This is useful for stress testing identifiability.
Supported patterns:
type = "none": no additional missingness.type = "block": region-time block dropout.type = "age_selective": older-age dropout (typically for non-trauma).type = "mnar": missing-not-at-random dropout driven by true death burden.type = "combined": apply all three.
If missing$missing_to_zero = TRUE, missing cells are converted to zeros
(tests robustness to bad pipelines). Otherwise missing values become NA.