Skip to contents

Takes amount from one channel and gives it to another, keeping each channel's flighting: from is scaled down by the same proportion in every period, and to gains amount in proportion to its spend in each period. The result is a new plan for simulate_plan() to compare with the old one.

Usage

reallocate(plan, from, to, amount, by = NULL)

Arguments

plan

A data frame with one row per period and a numeric column of spend per channel. The from and to columns must be complete, finite and non-negative.

from, to

Names of the columns to move spend from and to.

amount

Spend to move, non-negative. Without by, one number. With by, one number moved within every group, or a vector named by group that moves each named group its own amount and leaves the others as they are.

by

Optional grouping, as for transform_plan(): column names of plan, or a vector or data frame of vectors with one element or row per row of plan. Group names for amount are the group's values, joined with " / " when there is more than one grouping column.

Value

plan with the from and to columns changed and everything else as it was.

Details

Without by, amount is moved across the whole plan, so each group keeps its share of both channels. With by, it is moved within each group, so the total moved is amount times the number of groups, or the sum of a named amount. Moving an amount in one region only is how a reallocation is set up as a geo test.

A move larger than from's spend (in the group) is an error, and so is a move into a channel with no spend (in the group), since there is no flighting to spread the amount over. To start a dark channel, give it small weights in the periods it should run, for example plan$video <- 1e-6 for an even spread, and move the amount into it.

Examples

data(mm_weekly)
north <- mm_weekly[mm_weekly$geo == "north", ]

moved <- reallocate(north, from = "display", to = "social", amount = 2000)
c(display = sum(moved$display) - sum(north$display),
  social = sum(moved$social) - sum(north$social))
#> display  social 
#>   -2000    2000 

# Display's weeks off stay off, and social keeps its own pattern
all(moved$display[north$display == 0] == 0)
#> [1] TRUE
range(moved$social / north$social, na.rm = TRUE)
#> [1] 1.054472 1.054472

# A test in one region: only the north moves
test <- reallocate(mm_weekly, "display", "social",
                   amount = c(north = 2000), by = "geo")
tapply(test$social - mm_weekly$social, mm_weekly$geo, sum)
#> central   north   south 
#>       0    2000       0