Skip to contents

Completing panel spans

completing-spans.Rmd

panel_complete() turns every segmented organization into a seamless one: it fills the interior gaps so each organization is observed continuously across its own span. It is the friendly wrapper over panel_impute() – it fills gaps for all boundary types (not just panel-spanning ones) and interpolates by default.

A toy panel with two different kinds of gap (and a filler F so the middle years belong to the panel):

df <- data.frame(
  EIN2     = c("A", "A", "C", "C", "F", "F", "F", "F"),
  TAX_YEAR = c(2018, 2021, 2019, 2021, 2018, 2019, 2020, 2021),
  revenue  = c(0, 30, 100, 200, 5, 5, 5, 5)
)
attr(panel_describe(df, print = FALSE), "classification")[, c("EIN2", "panel_type", "panel_spell")]
#>   EIN2 panel_type panel_spell
#> 1    A persistent   segmented
#> 2    C    entrant   segmented
#> 3    F persistent    seamless

A spans the window with a two-year hole (persistent + segmented); C enters in 2019 and skips 2020 (entrant + segmented).

Complete every span 

done <- panel_complete(df, vars = "revenue")     # method = "interpolate", all types
done[done$imputed_row, c("EIN2", "TAX_YEAR", "revenue")]
#>   EIN2 TAX_YEAR revenue
#> 2    A     2019      10
#> 3    A     2020      20
#> 6    C     2020     150

Both organizations are now gap-free: A gets an interpolated 10 and 20 across 2019–2020, and C gets 150 in 2020. Re-describing confirms every span is seamless:

attr(panel_describe(done, print = FALSE), "classification")[, c("EIN2", "panel_type", "panel_spell")]
#>   EIN2 panel_type panel_spell
#> 1    A persistent    seamless
#> 2    C    entrant    seamless
#> 3    F persistent    seamless

The difference from panel_impute()

panel_impute() defaults to filling only persistent organizations, so the entrant C would be left with its hole:

imp <- panel_impute(df, vars = "revenue")        # default types = "persistent"
any(imp$EIN2 == "C" & imp$imputed_row)           # C is NOT filled
#> [1] FALSE

panel_complete() is the same engine with researcher-friendly defaults: fill everyone’s interior gaps, and interpolate rather than average. Choose a different rule with method = "mean" | "locf" | "nocb" (see Imputing missing years), and remember that completion only fills interior gaps – it never extrapolates past an organization’s first or last observed year. To reach the panel edges instead, trim to a rectangle with panel_balance().