STV with vote weighting

[abissellsiders]

Hi: This is a request, not an issue per se.

The stv() function is excellent. Common STV vote providers OpenSlides and OpaVote include weighted voting. However, the stv() function does not. I could not find any R packages which produce weighted STV outputs. If this is possible, it would be very much appreciated!

(Duplicating rows by weight [for example, a weight of 3 gets 3 rows] is feasible, but greatly slows calculation when the weights are large.)

[hanase]

This might be doable - thanks for the suggestion, @abissellsiders . Can I ask what is your use case? What would be the most convenient way to input the weights - an additional column in the dataset of votes?

[abissellsiders]

Hi @hanase, thank you for your quick reply and your excellent package!

Use case: An STV election for a large member organization in the United States which has "proxies" (delegates able to vote for other delegates who could not attend). In effect, this yields weights 1 (for normal delegates), 2 (for 1 proxy vote), 3 (for 2 proxy votes), etc. It is easy to achieve this weighting in stv() for a small number of integer weights:

library(vote)
library(data.table)

#####
# unweighted
#####
# read data
data(food_election)
# print
food_election
# stv results
stv(food_election)

#####
# weighted
#####
# copy data
food_election2 = data.table(food_election)
# suppose voters 1-5 have weight 3, and voters 6-15 weight 1
food_election2[, weight := c(rep(3,5), rep(1,15))]
# print
food_election2
# to run STV, we must duplicate rows and delete "weight" column
rows_repeat_maker = function(weights) {
  r = unlist(mapply(
    FUN = rep,
    1:length(weights),
    weights))
  return(r)
}
rows_repeat = rows_repeat_maker(food_election2[["weight"]])
food_election2_rep = food_election2[rows_repeat, ]
# remove weight column
food_election2_rep[, weight := NULL]
# print
food_election2_rep
# stv results
stv(food_election2_rep)
# this shifts the winner to "Strawberries", the #2 pick for most voters in 1-5

However, if the weight is either not an integer (or is a large integer), then the number of rows created must be very large. For example, suppose we had fractional weights between 0 and 1. We could approximate this by multiplying each weight by 1000 and duplicating rows (so ~0.333 gets 333 rows), but at the cost of increasing runtime by ~500 times.

IMO: The most convenient option would be a column of weights and an argument weight=NULL in stv().

[hanase]

Interesting - a very useful application! I have now implemented the functionality for the stv() function in the branch "weights". There is an additional argument "weight.column" which is the name of the weight column. So in your case, you would call it with stv(food_election2, weight.column = "weight"). Please test the branch if it is working as it should. In the next days, I'll put the same functionality into the other voting methods.

[hanase]

@abissellsiders The weighting is now available for all methods and is on the main branch. If I don't hear any complains in the next few days I'll push it to CRAN.