In itsdalmo/reporttoolDT: Work with surveys to generate reports

Foreword

Before we can generate a report, we need to "prepare" the data with the necessary information. This vignette gets into some more detail on the hidden fields in the Survey class. In this example, we'll use the same dataset as was shown in the "introduction" vignette and complete a Survey for PLS-PM modelling.

1. Read the data (seamless)

require(reporttoolDT)
sav <- read_data(system.file("extdata", "raw_data.sav", package = "reporttoolDT"))
class(sav)

2. Convert to Survey.

You can convert any data.frame to a Survey using survey(), or you can use one of the following functions which also coerces the input to a specific type of data.frame:

• survey_df(): A regular R data.frame.
• survey_dt() A data.table. Read more here.
• survey_tbl(), requires dplyr.

For this example, we can use a regular data.frame:

srv <- survey_df(sav)
class(srv)

3. Labels

When the data is read from SPSS, the labels are collected from the file. We can check the labels in several ways, but model() works best to get an overview:

model(srv)

The output from model shows us the column number, name, type in parantheses, and their labels. Let's say we wanted to change the labels to say "Loyalty" instead of just "loyal". We can do the following:

srv <- set_label(srv, q10 = "Loyalty 1", q15b = "Loyalty 2")
# To see that the labels have changed:
get_label(srv, c("q10", "q15b"))

If we wanted to change the label for several variables, we could also supply a list:

new_labels <- list(q10 = "Loyal 1", q15b = "Loyal 2")
srv <- set_label(srv, .list = new_labels)
# Same result as above.

4. Association

We also need to specify which variables are associated with which latent construct, we can specify them as follows:

srv <- set_association(srv, image = c("q4a", "q4b"))
model(srv)

As you can see from the output, q4a and q4b have a star next to them which indicates that the variables have an association. set_association() can also look for common latent associations, based on the name of the variables:

srv <- set_association(srv, .common = TRUE)
# Run model(srv) to see the result

Since all of our variables follow the naming convention, all associations have been identified.

5. Config

We also need to set the config for the survey. Most importantly, we need to set the cutoff to use for valid observations:

srv <- set_config(srv, name = "Example", segment = "B2C", cutoff = .3)

6. Marketshares

In order to be able to weight variables, we also need to specify the marketshare for each company in our study. After the mainentity association is set, we can do this using set_marketshare():

srv <- set_marketshare(srv, CompanyA = .3, CompanyB = .5, CompanyC = .2)
entities(srv)

Above, I have used the entities() function which gives you an overview of the entities in the data, and their marketshare if it is set. For the column "Valid", the values are NA because we have not calculated the percentage of missing values on variables associated with latents.

7. Latents (mean)

At this point we have two choices, calculate latent scores as a mean to do a "topline", or prepare it for PLS modelling using the PLS wizard, using the two functions latents_pls() and latents_mean(). The former does the following:

• Adds EM variables to the data (based on associations).
• Calculates the missing percentage (percent_missing).
• It adds a coderesp variable to the data (1 to N).
• Also, it calculates latents (mean) for each respondent.

srv_mean <- latents_mean(srv)
# model(srv_mean)

After running latents_mean() we see that the labels are empty. To fix this we could have set_translation() before adding the latents, or set_translation() and use the .auto argument for set_label():

srv_mean <- set_translation(srv_mean, .language = "english")
srv_mean <- set_label(srv_mean, .auto = TRUE)
tail(model(srv_mean))

With missing percentage calculated, we can check entities() again to see the number of valid observations:

entities(srv_mean)

8. PLS-wizard

Let's instead run latents_pls() to generate input for the PLS-wizard:

srv <- latents_pls(srv)
tail(model(srv))

Here, the EM variables and latents are not included - only percent_missing and coderesp. This (in addition to the associations we have set) is enough to create the input files for the PLS-wizard. To create the files, simply run:

# write_survey(srv, file = getwd())

The second argument file is the path to where you would like to store the Survey. write_survey() will always write a separate ...(Survey).Rdata file which contains all the hidden information for the Survey, which you can get back again by using read_survey(). If file is a directory, it will create a new directory Data and store the SPSS file, as well as Input which contains all the input files for the PLS-wizard.

More information

Introduction:

vignette("introduction", package = "reporttoolDT")

Survey-class:

vignette("survey", package = "reporttoolDT")

Other functions:

vignette("other", package = "reporttoolDT")

itsdalmo/reporttoolDT documentation built on May 18, 2019, 7:11 a.m.