Design-weighted survey aggregates for the prevalence targets
Source:R/trend_fit.R
aggregate_survey_prev.RdCollapses one survey dataset (or one bootstrap resample of it) to weighted sums by year, age, sex and IMD quintile: the total design weight, the design weight carried by current smokers, and the respondent count. Summing these over any set of cells and dividing gives the pooled design-weighted prevalence for that set exactly, which is what the survey-sourced calibration targets are built from. Storing sums rather than proportions is what makes that exact: a cell that is empty in a resample contributes nothing to either sum, which is the correct pooled estimator, whereas a missing proportion would need a decision about how to average over it.
Usage
aggregate_survey_prev(
data,
keep_ages,
keep_years,
state_var = "smk.state",
age_var = "age",
year_var = "year",
sex_var = "sex",
imd_var = "imd_quintile",
weight_var = "wt_int",
current_level = "current"
)Arguments
- data
One survey dataset or resample.
- keep_ages, keep_years
Integer vectors - the cells to keep. Years in keep_years that the survey does not cover are simply absent from the output; the caller decides whether that is expected.
- state_var, age_var, year_var, sex_var, imd_var, weight_var
Column names.
- current_level
Character - the value of the state variable that counts as a current smoker.