net_promoter: Calculate a 'Net Promoter Score' from rating scale data
In liam-c-smith/secretsauce: Top-Secret R Functions

Description Usage Arguments Details Value Examples

net_promoter() is a summary function which takes a vector of 0 to 10 rating scale data and returns a 'Net Promoter Score'. There are options to and calculate confidence intervals and treat missing values as 'Neutral' ratings.

1 2	net_promoter(x, scale = 100, include_na = FALSE, interval = FALSE, method = "wald", level = 0.95, boot_r = 2000, boot_type = "bca")

`x`	A numeric vector of integers between 0 and 10.
`scale`	A numeric scaling factor to apply to the score once it has been calculated.
`include_na`	Should missing values be treated as 'Neutral' responses?
`interval`	Should a confidence interval, as well as the score, be returned?
`method`	A string specifying the type of method used in calculating the confidence interval, see details. Must be either "wald" or "boot".
`level`	Desired confidence level.
`boot_r`	An integer specifying the number of bootstrap replications used when `method = "boot"`.
`boot_type`	A string specifying the type of calculation used to find the confidence limits when `method = "boot"`. These are the same as in `boot::boot.ci` but only `c("norm", "basic", "perc", "bca")` are implemented.

The 'Net Promoter Score' is commonly used to summarise responses to 'likelihood to recommend' questions obtained in customer surveys. The rating scale ranges from 0 to 10 but this is collapsed into 3 bins. 0 to 6 are considered 'Detractors', 7 to 8 are 'Neutral' and 9 to 10 are 'Promoters'. The score is then the difference between the proportion of 'Promoter' responses and the proportion of 'Detractor' responses.

As this score is a difference between two proportions in the same sample it takes discrete values between -1 and 1. However, when reported the score is often multiplied by 100 and treated much like a percentage. Of course, this scaling is completely arbitrary so the function provides an argument to change the scale but the default is set to 100.

Confidence Intervals:

By default, the function will return the score alone but when interval = TRUE a confidence interval will also be calculated. Either a Wald-type or bootstrapping approach will be used depending on the value of method.

For method = #' "wald" the score is assumed to be normally distributed with the variance and derived from the properties of a multinomial random variable.

For method = "boot", bootstrapping will be employed using functions from the boot package. The default approach for calculating the interval the adjusted bootstrap percentile method but other boot::boot.ci interval types can be specified.

The Wald-type intervals should be a good approximation in large samples but when the sample size is small or the score is near -1 or +1 bootstrapping is recommended.

If interval = FALSE then only the score is returned as a numeric value. Otherwise both the score and limits of the confidence interval will be returned in a length 3 numeric vector of the form c(score, lower, upper).

# Generate rating scale data
p <- 0.5 # probability of 'Promoter'
d <- 0.3 # probability of 'Detractor
prob <- c(rep(d, 7) / 7, rep(1 - p - d, 2) / 2, rep(p, 2) / 2)
ratings <- sample(0:10, 30, TRUE, prob)

# Calculate NPS
net_promoter(ratings)                   # default action
net_promoter(ratings, scale = 1)        # unadulterated score
net_promoter(ratings, interval = TRUE)  # get a confidence interval

# Wald-type confidence interval can lead to impossible limits when the score is near the
# extremes. Its better to use a percentile bootsrapping method in this case.
p <- 0.05
d <- 0.9
prob <- c(rep(d, 7) / 7, rep(1 - p - d, 2) / 2, rep(p, 2) / 2)
set.seed(1234)
ratings <- sample(0:10, 30, TRUE, prob)

net_promoter(ratings, interval = TRUE, method = "wald")
net_promoter(ratings, interval = TRUE, method = "boot")