pachamaltese pachamaltese - 12 days ago 5
R Question

plyr + simpleboot: NA in probability vector

I am using

simpleboot
package (https://cran.r-project.org/web/packages/simpleboot/index.html) to obtain confidence intervals.

This is my function:

lb_weighted_median_dplyr <- function(x,v) {
set.seed(1234)
b <- one.boot(x, weights = v, FUN = function(x,w) matrixStats::weightedMedian(x, w = v, na.rm = TRUE), R = 100, student = FALSE)
round(perc(b, 0.025), 0)
}


What the function does is to calculate the lower bound of the confidence interval when I run

ddply(wage_by_gender_2015, .(sex,region), summarise, FUN = lb_weighted_median_dplyr(wage, exp_region))


Where
wage
is a numeric column and
exp_region
is another numeric column that contains weights.

I don't have data for some regions, therefore the function fails with some regions and returns

Error in eval(substitute(expr), envir, enclos) : NA in probability vector


How can I bypass that error and obtain NA as the lower bound for a region without data?

A
dplyr
equivalent approach that also returns
NA in probability vector
is

grouped <- group_by(wage_by_gender_2015, sex, region)
dplyr::summarise(grouped, FUN = lb_weighted_median_dplyr(wage, exp_region))


Relevant sample of the data here: http://users.dcc.uchile.cl/~mvargas/casen/wage_by_gender_2015.RData

Answer
wage_by_gender_2015 <- data.frame(sex    = rep(c("male", "female"),100),
                                  region = rep(c("north", "south", "east",
                                                 "west"), 50),
                                  exp_region = abs(rnorm(100)),
                                  wage       = abs(rnorm(100))
)

wage_by_gender_2015$exp_region[10] <- NA
ddply(wage_by_gender_2015, .(sex,region), summarise, FUN = lb_weighted_median_dplyr(wage, exp_region))
 Error in sample.int(length(x), replace = TRUE, ...) :    NA in probability vector
# impute
wage_by_gender_2015$exp_region <- RRF::na.roughfix(wage_by_gender_2015$exp_region)

ddply(wage_by_gender_2015, .(sex,region), summarise, FUN = lb_weighted_median_dplyr(wage, exp_region))
    sex region FUN
1 female  south   0
2 female   west   0
3   male   east   1
4   male  north   0

As mentioned in the comment I would've used your sample data but it was missing sex.