bws.dataset: Creating a data set suitable for case 1 best-worst scaling...

In support.BWS: Tools for Case 1 Best-Worst Scaling

Creating a data set suitable for case 1 best-worst scaling analysis

Description

This function creates a data set used for bws.count() in support.BWS and clogit() in survival.

Usage

bws.dataset(data = NULL, response.type = 1,
            choice.sets, design.type = 1,
            item.names = NULL, row.renames = TRUE,
            id = NULL, response = NULL,
            model = "maxdiff", delete.best = FALSE,
            version = NULL, respondent.dataset = NULL)

Arguments

data	A data frame including responses to BWS questions (i.e., a survey/respondent data set).
response.type	A value describing the format of the response variables: 1 if the response variables follow a row number format, and 2 if they follow an item number format.
choice.sets	A data frame or matrix containing choice sets.
design.type	A value describing how to design the choice sets: 1 if the design assigned to choice.sets is a two-level OMED, and 2 if it is a BIBD.
item.names	A vector containing the names of items: if it takes NULL, default item names (i.e., ITEM1, ITEM2, ...) are used in the resultant data set.
row.renames	A logical variable describing whether or not the row names of a data set created by this function are changed. If TRUE, integer values are assigned to the row names starting from 1. If FALSE, the row names are not changed.
id	A character showing the name of the respondent identification number variable used in the respondent data set.
response	A vector containing the names of response variables in the respondent data set, showing the best and worst items selected in each BWS question.
model	A character showing a type of data set created by this function: "maxdiff" for the maxdiff model; "marginal" for the marginal model; and "sequential" for the marginal sequential model.
delete.best	A logical value denoted by TRUE when deleting an item selected as the best in the worst choice set (that is, using a marginal sequential model) or FALSE when not doing so. The argument is deprecated. Please use the argument model instead.
version	A character showing the name of the version variable used in the respondent dataset and choice sets.
respondent.dataset	A data frame containing a respondent data set. The argument is deprecated. Please use the argument data instead.

Details

The respondent data set, in which each row corresponds to a respondent, has to be organized by users. The data set must contain the id variable in the first column, denoting the respondent's identification number, and the response variables in the subsequent columns, each indicating which items are selected as the best and the worst for each question. Although the names of the id and response variables are up to the discretion of the user, the response variables must be constructed such that the best alternates with the worst by question. For example, when there are seven BWS questions, the variables are B1 W1 B2 W2 ... B7 W7; here, Bi and Wi show which items are selected as the best and the worst in the ith question.

There are two types of data format related to response variables: one is a row number format, and the other is an item number format. In the former, the row numbers of the items selected as the best and the worst are stored in the response variables. In the latter, item numbers are stored in the response variables.

The arguments choice.sets, design.type, and item.names are the same as those in the bws.questionnaire() function. However, item.names can take NULL (default), when default item names (i.e., ITEM1, ITEM2, ...) are used in the resultant data set. Further, the order of questions in the choice sets has to be the same as that in the respondent data set.

The argument version is set when two or more versions of the choice sets are used. The version variable must be included in the respondent dataset and choice sets. The variable denotes the serial integer number of versions starting from 1.

Note that this function in version 0.2-0 and later versions of the package can create a data set for the marginal (sequential) model as well as that for the maxdiff model: "maxdiff" is assigned to the argument model when the maxdiff model is used for the analysis; "marginal" is assigned when the marginal model is used; and "sequential" is assigned when the marginal sequential model is used. Furthermore, the argument delete.best is deprecated: please use the argument model = "sequential" instead.

Value

This function returns a data set in data frame format for the maxdiff model or one for the marginal (sequential) model. The data set for the maxdiff model contains the following variables:

ID	A respondent's identification number, assigned according to the id variable in the respondent data set.
Q	A serial number of BWS questions.
PAIR	A serial number of possible pairs of the best and worst items for each question.
BEST	An item number treated as the best in the possible pairs of the best and worst items for each question.
WORST	An item number treated as the worst in the possible pairs of the best and worst items for each question.
RES.B	An item number selected as the best by respondents.
RES.W	An item number selected as the worst by respondents.
ITEMj	State variables related to the possible pairs of the best and worst items: 1 if item j is treated as the best in the possible pair, -1 if item j is treated as the worst in the possible pair, and 0 otherwise. These variables are used as independent variables in the model formula of the clogit function in survival when analyzing responses to BWS questions.
RES	Responses to BWS questions: TRUE if a possible pair of the best and worst items is selected by respondents and FALSE otherwise. This variable is used as a dependent variable in the model formula of the clogit function in survival when analyzing responses to BWS questions.
STR	A stratification variable identifying each combination of respondent and question. This variable is also used in formula of the clogit function with the strata function.

The data set for the marginal (sequential) model contains the variables ID, Q, RES.B, RES.W, and STR mentioned above and the following variables:

ALT	A serial number of alternatives for each question.
BW	A state variable that takes the value of 1 for the possible best items and -1 for the possible worst items.
Item	An item number treated as the possible best or worst items for each question.
RES	Response to BWS questions: TRUE if a possible best or worst item is selected by respondents and FALSE otherwise. This variable is used as a dependent variable in the model formula of the clogit function in survival when analyzing responses to BWS questions.
ITEMj	State variables that takes the value of 1 for the possible best items and -1 for the possible worst items. These variables are used as independent variables in the model formula of the clogit function in survival when analyzing responses to BWS questions.

Furthermore, the resultant data set includes respondent characteristic variables when the respondent data set has those variables.

Author(s)

Hideo Aizaki

See Also

support.BWS-package, oa.design, find.BIB, isGYD, clogit, strata

Examples

## Not run: 
## load packages
require(DoE.base) # include oa.design() used to generate a two-level OMED
require(crossdes) # include find.BIB() used to generate a BIBD
require(survival) # include clogit() used to analyze responses

if(getRversion() >= "3.6.0") RNGkind(sample.kind = "Rounding")

## example 1: BWS using a two-level OMED
## suppose that ten respondents answered twelve BWS questions valuing nine items

# create a two-level OMED with nine factors
set.seed(123) # set seed for random number generator
des1 <- oa.design(nfactors = 9, nlevels = 2)
des1 # resultant design with twelve rows, nine columns, and level values of 1 and 2

# set item names, in which the order of elements corresponds to 
#  the order of columns in des1
items1 <- LETTERS[1:9] # item names are "A", "B", ..., "I"

# create questions for BWS
bws.questionnaire(
 choice.sets = des1,
 design.type = 1, # OMED
 item.names = items1)

# set a respondent data set in a row number format
res1 <- data.frame(
 ID  = c(1, 2, 3, 4, 5, 6, 7, 8, 9, 10), # id variable
 B1  = c(1, 1, 1, 1, 3, 5, 1, 1, 1, 1),  # best item in question 1
 W1  = c(5, 5, 3, 5, 4, 1, 4, 2, 4, 3),  # worst item in question 1
 B2  = c(1, 3, 1, 4, 1, 2, 3, 2, 1, 2),  # best item in question 2
 W2  = c(5, 5, 5, 5, 3, 5, 1, 4, 4, 5),  # worst item in question 2
 B3  = c(1, 2, 1, 2, 4, 1, 1, 3, 1, 3),  # best item in question 3
 W3  = c(4, 3, 3, 3, 3, 4, 4, 2, 3, 4),  # worst item in question 3
 B4  = c(2, 1, 3, 5, 2, 3, 1, 1, 2, 5),  # best item in question 4
 W4  = c(4, 4, 5, 3, 5, 5, 3, 5, 4, 3),  # worst item in question 4
 B5  = c(2, 3, 3, 2, 2, 2, 2, 1, 3, 2),  # best item in question 5
 W5  = c(4, 4, 4, 4, 3, 4, 4, 4, 4, 3),  # worst item in question 5
 B6  = c(2, 1, 1, 3, 3, 3, 1, 1, 1, 1),  # best item in question 6
 W6  = c(1, 2, 3, 2, 1, 2, 3, 2, 2, 3),  # worst item in question 6
 B7  = c(3, 3, 1, 1, 3, 6, 1, 2, 1, 7),  # best item in question 7
 W7  = c(9, 6, 8, 4, 8, 2, 6, 5, 4, 6),  # worst item in question 7
 B8  = c(2, 1, 2, 2, 2, 1, 1, 3, 1, 1),  # best item in question 8
 W8  = c(3, 3, 3, 3, 4, 4, 4, 4, 3, 4),  # worst item in question 8
 B9  = c(2, 1, 3, 1, 4, 2, 3, 4, 1, 1),  # best item in question 9
 W9  = c(3, 2, 2, 3, 3, 3, 2, 2, 3, 4),  # worst item in question 9
 B10 = c(1, 1, 1, 1, 1, 1, 1, 4, 3, 3),  # best item in question 10
 W10 = c(4, 2, 2, 4, 4, 3, 4, 2, 4, 4),  # worst item in question 10
 B11 = c(2, 1, 3, 3, 3, 2, 1, 2, 2, 4),  # best item in question 11
 W11 = c(1, 4, 4, 1, 1, 4, 4, 4, 1, 1),  # worst item in question 11
 B12 = c(2, 2, 1, 1, 1, 1, 3, 2, 1, 2),  # best item in question 12
 W12 = c(3, 3, 2, 3, 3, 2, 2, 3, 3, 3))  # worst item in question 12

# create a data set for the maxdiff model analysis 
#  by combining the choice sets and respondent data set
dat1 <- bws.dataset(
 respondent.dataset = res1,
 response.type = 1, # row number format
 choice.sets = des1,
 design.type = 1)   # OMED

# analyze responses to BWS questions
# counting approach
bws1 <- bws.count(dat1)
# modeling approach
# note: ITEM5 is excluded from fr1 to normalize its coefficient to zero
fr1 <- RES ~ ITEM1 + ITEM2 + ITEM3 + ITEM4 + ITEM6 + ITEM7 +
             ITEM8 + ITEM9 + strata(STR)
clg1 <- clogit(formula = fr1, data = dat1)
clg1


## example 2: BWS using a balanced incomplete block design
## suppose that ten respondents answered seven BWS questions valuing seven items

# create a BIBD with seven items, four items per question, and seven questions
set.seed(123) # set seed for random number generator
des2 <- find.BIB(trt = 7, k = 4, b = 7)
isGYD(des2)  # check whether the design is a BIBD
des2 # resultant design with seven rows, four columns, and level values ranging from 1 to 7

# set item names, in which the order of element corresponds to 
#  the order of level values in des2
items2 <- LETTERS[1:7]

# create questions for BWS
bws.questionnaire(
 choice.sets = des2,
 design.type = 2, # BIBD
 item.names = items2)

# set a respondent data set in a row number format
res2 <- data.frame(
 ID = c(1, 2, 3, 4, 5, 6, 7, 8, 9, 10), # id variable
 B1 = c(2, 1, 2, 4, 2, 2, 2, 1, 2, 1),  # best item in question 1
 W1 = c(3, 4, 3, 3, 1, 3, 3, 2, 3, 3),  # worst item in question 1
 B2 = c(4, 3, 3, 3, 3, 1, 1, 2, 1, 1),  # best item in question 2
 W2 = c(3, 2, 4, 1, 2, 3, 3, 4, 2, 4),  # worst item in question 2
 B3 = c(3, 1, 1, 1, 1, 1, 1, 1, 2, 1),  # best item in question 3
 W3 = c(1, 4, 2, 2, 4, 4, 2, 3, 3, 3),  # worst item in question 3
 B4 = c(2, 2, 1, 3, 2, 2, 2, 2, 4, 1),  # best item in question 4
 W4 = c(4, 4, 3, 4, 1, 3, 4, 1, 2, 4),  # worst item in question 4
 B5 = c(1, 3, 2, 1, 3, 2, 1, 1, 1, 1),  # best item in question 5
 W5 = c(3, 1, 4, 4, 1, 4, 3, 2, 4, 3),  # worst item in question 5
 B6 = c(2, 1, 1, 3, 2, 4, 4, 3, 3, 3),  # best item in question 6
 W6 = c(3, 2, 3, 4, 3, 2, 2, 4, 4, 2),  # worst item in question 6
 B7 = c(2, 1, 3, 1, 3, 2, 3, 3, 2, 2),  # best item in question 7
 W7 = c(4, 4, 4, 4, 4, 1, 4, 1, 4, 4))  # worst item in question 7

# create a data set for the maxdiff model analysis
#  by combining the choice sets and respondent data set
dat2 <- bws.dataset(
 respondent.dataset = res2,
 response.type = 1,   # row number format
 choice.sets = des2,
 design.type = 2,     # BIBD
 item.names = items2) # state variables are labeled using item names

# analyze responses to BWS questions
# counting approach
bws2 <- bws.count(dat2)
bws2
# the argument cl is set to 2 to generaet a data set
#  of the S3 class 'bws.count2'
bws2.2 <- bws.count(dat2, cl = 2)
plot(bws2.2, score = "bw")
barplot(bws2.2, score = "bw")
sum(bws2.2)
summary(bws2.2)
# modeling approach
# note: D is excluded from fr2 to normalized its coefficient to zero
fr2 <- RES ~ A + B + C + E + F + G + strata(STR)
clg2 <- clogit(fr2, data = dat2)
clg2
bws.sp(clg2, base = "D", order = TRUE)

## End(Not run)

support.BWS documentation built on March 31, 2023, 8:12 p.m.