In kthorstmann/consr: Compute Consensus For Likert Scales

This package is based on an article by Tastle & Wiermann, (2006)[^1]. It implements their functions and returns results for items and dataframes alike.

Installing the package

To install the package, run this code:

install.packages("devtools")
library(devtools)
devtools::install_github("kthorstmann/consr")
library(consr)

Example - a single item

First, lets consider the example given by Tastle & Wiermann on p. 537, equation 2:

library(consr)
item <- c(rep(1, 19), rep(2, 16), rep(3, 26), rep(4, 29), rep(5, 10))

With this package, you can create Figure 1, p. 537 simply with:

freq_item(item, plot = TRUE)

In freq_item, additional arguments can be passed on to barplot using .... The function also returns a table with the distributions of values per category.

Secondly, you can simply compute the consensus by running

consensus(item=item, range = NULL)

The output contains the following values:

• n The number of valid answers on that item
• min The minimum answer on that item
• max The maximum answer on that item
• consensus The consensus
• dissensus The dissensus
• n_missing The absolute number of missing values on that item, defined as NA
• perc_missing The relative number of missing values on that item, defined as NA

Now, lets consider an item that is not as clean and has missings as well as values that are not part of the likert scale:

item.dirty <- c(rep(1, 15), rep(2, 8), rep(3, 26), rep(4, 29), rep(5, 5), rep(NA, 5), rep(-2, 9), rep(-3, 4))

This item neads to be cleaned first, i. e. we should first get an overview of how many values are actually in each category:

freq_overview(item.dirty, return.df = TRUE)

Using freq_overview will give us an overview over each answer on the item. Using the range argument will tell the function, what range the liker scale actually has and thereby, which of the values are valid. With the argument return.df set to TRUE, the function will return a data.frame.

A vector is returned with each answer on the item as two categories. First, we see C_min3_abs, which means that -3 (min, since - is not allowd in names of items), was 4 times present in the data frame. Under the entry C_min3_perc, we see that this constitues about 3.960396% of the entries in the item item.dirty.

Actually, this gives just an overview and before computing the consensus, each item should be prepared and recoded in a proper way. To make recoding very easy, consr contains a function called prep_item that prepares an item for further processing.

prep_item(item.dirty, range = c(1, 5))

Using this function again with the range argument, all values that are not within the range are recoded to missings with the standard formula NA. This can be useful, for example, when online surveys return different values for different types of missings.

Example 2 - a data frame

Usually, one wants to process complete dataframes. consr contains functions to do this, usually indicated with an _df at the end.

Let us take a look at a normal data frame one might receive from a survey:

n = 10 # (max = 26 for now)
var_names <- paste0("var_", letters[1:10])
entries <- matrix(sample(-2:2, n*10, replace = TRUE),
                  nrow = n, ncol = 10)
entries[sample(1:100, 15)] <- NA # throw in some NA
entries[sample(1:100, 10)] <- -77 # throw in some missings, coded as -77
data <- as.data.frame(entries)
colnames(data) <- var_names
subj_id <- paste0("id_", sample(letters, n, replace = FALSE)) # get character subj ids
data <- cbind(subj_id, data, stringsAsFactors = FALSE)
data["var_j"] <- data["var_j"] / 2.3 # get one non-integer

This data frame contains multiple variables, some of them not being answers to likert-type scales (var_j, subj_id). The data frame contains also entries such as -77, which are a missing, but have not been recoded yet.

First, let us exclude the variables that are not likert type scales, using the function integer_subset_df:

data_int <- integer_subset_df(data)
data_int

This returns an data set with only integers. A warning message tells you which variables were considered as such.

You can then move on to prepare the data frame for consensus analyses, using prep_item_df:

data_pre <- prep_item_df(data_int, range = c(-2, 2))
data_pre

Using this data set, you can run the consensus analyses. Note that there is no freq_overview version for a data frame. This is due to different length of arguments that are returned by this function, depending on the item. You can use something like purrr::map(data, freq_overview) to get an overview for the whole data frame.

consensus_df(data_pre)

Note that you can also do this in one single step:

consensus_df(data, range = c(-2, 2))

The range argument

The range argument is nearly part of all functions, since it indicates which are the valid vlaues in any given item. The range is entered as a vector and has only two arguments (though, if you pass on more than that, the min and max values will be taken).

If you ever have trouble using this argument, recode your data properly before computing the consensus and set range = NULL, which is the default. It will then be guessed from your data frame for each item.

[^1]: Tastle, W. J., & Wierman, M. J. (2007). Consensus and dissention: A measure of ordinal dispersion. International Journal of Approximate Reasoning, 45(3), 531–545. http://doi.org/10.1016/j.ijar.2006.06.024.

kthorstmann/consr documentation built on May 20, 2019, 7:04 p.m.