In caldwellst/augury: Provides Streamlined Methods for Data Imputation and Forecasting for WHO DDI Statistics

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

The simple methods available in augury are easy to use, but provide the same functionality of allowing a test column and returning error metrics as the more complex modeling functions available in the package. Let's use data on alcohol from the GHO to demonstrate the functionality.

library(augury)

df <- ghost::gho_data("SA_0000001688",
                      query = "$filter=Dim1 eq 'BTSX'") %>%
  billionaiRe::wrangle_gho_data(source = "WHO GHO",
                                type = "estimated") %>%
  dplyr::arrange(iso3, year)

head(df)

Here we can see that data has time series and gaps in years. We can use linear interpolation and flat extrapolation here to get data out to 2023.

df <- tidyr::expand_grid(iso3 = unique(df$iso3),
                         year = 2000:2023) %>%
  dplyr::left_join(df, by = c("iso3", "year"))

df %>%
  dplyr::filter(iso3 == "AFG",
                year >= 2010,
                year <= 2018) %>%
  dplyr::select(iso3,
                year,
                value)

Let's now use our linear interpolation and flat extrapolation on this data.

pred_df <- predict_simple(df,
                          group_col = "iso3",
                          sort_col = "year") 

pred_df %>%
  dplyr::filter(iso3 == "AFG",
                year >= 2010,
                year <= 2018) %>%
  dplyr::select(iso3, year, value)

And we can see our linear interpolation there. We can also see the flat extrapolation.

pred_df %>%
  dplyr::filter(iso3 == "AFG",
                year > 2016) %>%
  dplyr::select(iso3, year, value)

We can use the predict_average() function in much the same way, except it is most useful when we have robust series for a set of countries, and no data for others. We can then use something like the regional average to infill data for missing countries.

df <- ghost::gho_data("PHE_HHAIR_PROP_POP_CLEAN_FUELS") %>%
  billionaiRe::wrangle_gho_data(source = "WHO GHO",
                                type = "estimated") %>%
  dplyr::filter(whoville::is_who_member(iso3))

x <- whoville::who_member_states()
x[!(x %in% df$iso3)]

Above, we have 4 missing WHO member states, Lebanon, Cuba, Bulgaria, and Libya. Let's use regional averaging to fill in this data. We can use the most recent World Bank income groups from the whoville package as our relevant group.

df <- tidyr::expand_grid(iso3 = x,
                         year = 2000:2018) %>%
  dplyr::left_join(df, by = c("iso3", "year")) %>%
  dplyr::mutate(region = whoville::iso3_to_regions(iso3, region = "wb_ig"))

predict_average(df,
                average_cols = c("region", "year"),
                group_col = "iso3",
                sort_col = "year",
                type_col = "type",
                source_col = "source",
                source = "WB IG regional averages") %>%
  dplyr::filter(iso3 == "LBN")

Hope these examples have been clear and highlight some of the usefulness of these simple modelling functions.

caldwellst/augury documentation built on Oct. 10, 2024, 8:20 a.m.