Interpolate or carry forward missing observations in a panel. Every value this invents is flagged in a companion column, and that flag is not optional – an imputed value that travels through a pipeline looking like data is exactly the failure this package exists to prevent.
Usage
interpolate_missing(
data,
value = NULL,
method = c("linear", "locf", "none"),
max_gap = 3
)Arguments
- data
A panel with
iso3candyear.- value
Column(s) to fill (character).
NULLfills every numeric column exceptyear.- method
"linear"(default, interior gaps only),"locf"(carry the last observation forward) or"none".- max_gap
Longest run of consecutive missing years to fill. Gaps longer than this are left alone, because interpolating across a decade is not interpolation. Default
3.
Value
data with the gaps filled and, for each filled column, a logical
<column>_imputed companion. The map verbs count those columns when they
write provenance, so they keep working through any verb that preserves
columns. An "countryatlas_imputed" attribute lists the flag columns for
convenience, but nothing in the package reads it, and dplyr drops it as
it drops most attributes – rely on the columns, not the attribute. Rows
come back sorted by iso3c then year.
The hard rule
The flag cannot be turned off. world_map() reads it and refuses to draw
imputed values as though they were observed without at least noting it in the
caption. If you need values with no flag, compute them yourself – the
package will not hand you a frame where invented numbers are indistinguishable
from measured ones.
Examples
p <- data.frame(iso3c = "USA", year = 2000:2005,
gdp = c(1, NA, NA, 4, NA, 6))
interpolate_missing(p, "gdp")
#> # A tibble: 6 × 4
#> iso3c year gdp gdp_imputed
#> <chr> <int> <dbl> <lgl>
#> 1 USA 2000 1 FALSE
#> 2 USA 2001 2 TRUE
#> 3 USA 2002 3 TRUE
#> 4 USA 2003 4 FALSE
#> 5 USA 2004 5 TRUE
#> 6 USA 2005 6 FALSE