Projects prevalences of disability with an estimated model
Source:R/predict.prevalence_model.R
predict.prevalence_model.RdCalculates prevalences by age and year from a model estimated with fitPrevalenceModel.
Arguments
- object
a model estimated with
fitPrevalenceModel- newdata
a data frame with variables 'year', 'age' (+ 'qx' for method "mortality") and, optionally, the variables defining series
- interval
if TRUE, adds the lower and upper bounds of a prediction interval ('pix_lwr' and 'pix_upr')
- level
confidence level of the prediction interval
- on_unfeasible
method "target" only: what to do if the hypothesis cannot be satisfied with the projected mortality, i.e. if it requires prevalences outside the interval 0 to 1 (all ages) or if no uniform evolution can reach the target (minimum age). "warn" (default) returns the values as they are with a warning (prevalences outside the interval 0 to 1, or NA for a uniform evolution without solution); "clip" bounds prevalences to the interval 0 to 1 (or to the closest feasible solution), with a warning; "error" stops.
- ...
not used
Value
the rows of newdata with a new variable 'pix' (projected prevalence), and 'pix_lwr' and 'pix_upr' if requested.
If the scale of the model is "log" or "level", prevalences are bounded to the interval 0 to 1 (with a warning).
Details
With methods "trend" and "shift", newdata only needs the years and ages. With method "mortality", newdata must also contain the
mortality rates 'qx' for all the ages of the model and for each year (they are used at ages \(a+k\), see fitPrevalenceModel).
Variables defining series (sex, categories) can be omitted from newdata: it is then used for all the values
of these variables in the model (for example, mortality rates by sex apply to all categories).
Rows of newdata with ages that are not in the model are dropped.
Prediction intervals (with interval = TRUE) are calculated on the scale of the estimation, from the
residual variance of the regression, and then transformed back to prevalences. They do not take into account
the uncertainty on the projected mortality rates, nor the uncertainty on the validity of the model outside the period of estimation.
They are not available for estimator = "endpoints", nor for methods "target" and "shift".
Examples
ages <- 60:100; years <- 2010:2019
obs <- expand.grid(age = ages, year = years)
obs$pix <- plogis(-6 + 0.12 * (obs$age - 60) - (0.01 + 0.0005 * (obs$age - 60)) * (obs$year - 2019))
m <- fitPrevalenceModel(obs, method = "trend")
proj <- predict(m, newdata = expand.grid(age = ages, year = c(2025, 2030)), interval = TRUE)
head(proj)
#> age year pix pix_lwr pix_upr
#> 1 60 2025 0.002328964 0.002328964 0.002328964
#> 2 61 2025 0.002617277 0.002617277 0.002617277
#> 3 62 2025 0.002941177 0.002941177 0.002941177
#> 4 63 2025 0.003305027 0.003305027 0.003305027
#> 5 64 2025 0.003713722 0.003713722 0.003713722
#> 6 65 2025 0.004172744 0.004172744 0.004172744