Csdrhrt

The long title would've been "create cols/var 1) named as a function of specific responses in previous cols/var and 2) that contain new values as function of previous responses positions in an array (best-worst scaling application)"

I'm performing some best-worst scaling (maxdiff) analyses and I need to convert the data from the raw responses data file into a data file suitable for BWS analyses.

My responses data file is dfex and the end result should be dfbw.
Basically, the example starting datafile is self-explanatory, i.e., there are two questions q01 and q02, and the respondents indicate which of the available alternatives to these questions are "best" or "worst" ("most useful" and "least useful", here).

The final data file, suited for BWS analysis, should indicate for each question, which alternative, out of four possible per each question, was chosen. For instance, in case 1, question 1, the respondent indicated 1st alternative as the "most useful" and 3rd alternative as the "least useful".

I currently doing this via some seriously stupid manual conversions of scores but I'm certain that much more elegant solutions are available (I'm just not that advanced in R and I'm under a lot of pressure).

Could you, please, point me in the right direction (i.e., how to automate this process and make it more smoothly) and, if possible, please indicate your reasons for choosing using a certain package or particular approach versus another (e.g., a custom function versus a combination of functions in a package/s)?

    # This is my starting data frame

    dfex <- data.frame(stringsAsFactors=FALSE,

          q01_01 = c("most useful", "least useful", "least useful", NA),

          q01_02 = c(NA, "most useful", "most useful", "most useful"),

          q01_03 = c("least useful", NA, NA, "least useful"),

          q01_04 = c(NA, NA, NA, NA),

          q02_01 = c("least useful", "least useful", NA, "least useful"),

          q02_02 = c(NA, NA, NA, NA),

          q02_03 = c(NA, "most useful", "most useful", "most useful"),

          q02_04 = c("most useful", NA, "least useful", NA))



     # which now I'm processing manually

    dfex <- dfex %>% 

      mutate(b01_01 = case_when(q01_01 == "most useful" ~ 1)) %>% 

      mutate(b01_02 = case_when(q01_02 == "most useful" ~ 2)) %>% 

      mutate(b01_03 = case_when(q01_03 == "most useful" ~ 3)) %>% 

      mutate(b01_04 = case_when(q01_04 == "most useful" ~ 4)) %>% 

      mutate(b02_01 = case_when(q02_01 == "most useful" ~ 1)) %>% 

      mutate(b02_02 = case_when(q02_02 == "most useful" ~ 2)) %>% 

      mutate(b02_03 = case_when(q02_03 == "most useful" ~ 3)) %>% 

      mutate(b02_04 = case_when(q02_04 == "most useful" ~ 4)) %>%

      mutate(w01_01 = case_when(q01_01 == "least useful" ~ 1)) %>% 

      mutate(w01_02 = case_when(q01_02 == "least useful" ~ 2)) %>% 

      mutate(w01_03 = case_when(q01_03 == "least useful" ~ 3)) %>% 

      mutate(w01_04 = case_when(q01_04 == "least useful" ~ 4)) %>% 

      mutate(w02_01 = case_when(q02_01 == "least useful" ~ 1)) %>% 

      mutate(w02_02 = case_when(q02_02 == "least useful" ~ 2)) %>% 

      mutate(w02_03 = case_when(q02_03 == "least useful" ~ 3)) %>% 

      mutate(w02_04 = case_when(q02_04 == "least useful" ~ 4))



    # create (manually) cols 

    dfex %>% 

      select(c(b01_01:b01_04)) %>% 

      rowSums(., na.rm = T) -> dfex$B1

    dfex %>% 

      select(c(b02_01:b02_04)) %>% 

      rowSums(., na.rm = T) -> dfex$B2

    dfex %>% 

      select(c(w01_01:w01_04)) %>% 

      rowSums(., na.rm = T) -> dfex$W1

    dfex %>% 

      select(c(w02_01:w02_04)) %>% 

      rowSums(., na.rm = T) -> dfex$W2



    # getting the desired BW datafile after selecting only cols of interest

    dfbw <- dfex %>% select(B1:W2)

PS: Because of all the attention given to tidyverse I thought this package collection might be suitable for my problem (?).

The tidyverse collection of packages is great, once you've learned how to format your data, and that it's usually never necessary to a) write loops of any kind or b) do more work coding once you get more data.

Start by making your data frame "tidy". Each observation should be one row, so you want a column for the question ID, a column for the alternative, and one for the answer. This is what gather does: it makes your "wide" data "long". We further remove the "q" from the question ID, as it's not needed.

Also, add a column that gives you the subject ID – it's basically the row number in each group of question/alternatives. This is done with a combination of group_by and mutate.

library(tidyverse)



dfex.tidy = dfex %>% 

  gather(question, answer, starts_with("q")) %>% 

  separate(question, into=c("question", "alternative")) %>% 

  mutate(question = str_replace_all(question, "q", "")) %>% 

  group_by(question, alternative) %>% 

  mutate(subject = row_number())

This gives you:

# A tibble: 32 x 4

# Groups:   question, alternative [8]

   question alternative answer       subject

   <chr>    <chr>       <chr>          <int>

 1 01       01          most useful        1

 2 01       01          least useful       2

 3 01       01          least useful       3

 4 01       01          NA                 4

 5 01       02          NA                 1

 6 01       02          most useful        2

 7 01       02          most useful        3

 8 01       02          most useful        4

 9 01       03          least useful       1

10 01       03          NA                 2 

… and so on. As you can see, each answer is a row. (In fact, non-answers are also part of this data frame but it's not required to remove those rows.)

Now, we filter out all missing answers, and re-code them from "most/least useful" to 1 and 2 so we know their intrinsic order.

We then group by subject and question. The chosen "most useful" alternative will be the first alternative when ordering the results by the answer value (1 or 2). Vice-versa, the "least useful" alternative will be the first alternative when ordering the results by the descending answer value:

dfex.results = dfex.tidy %>% 

  filter(!is.na(answer)) %>% 

  # recode answer to 1 or 2

  mutate(answer = ifelse(answer == "most useful", 1, 2)) %>% 

  group_by(subject, question) %>% 

  summarize(

    best = first(alternative, order_by = answer),

    worst = first(alternative, order_by = desc(answer))

  )

This gives you:

  subject question best  worst

    <int> <chr>    <chr> <chr>

1       1 01       01    03   

2       1 02       04    01   

3       2 01       02    01   

4       2 02       03    01   

5       3 01       02    01   

6       3 02       03    04   

7       4 01       02    03   

8       4 02       03    01

This data should be easy to work with and scales for any number of questions, alternatives, and answers, as long as your input data stays in the same format. That said, you should really try to collect your input data in a tidy manner already, if possible.

You may further tidy the data in order to get the best or worst alternative for a given subject and question:

dfex.results %>% 

  gather(type, alternative, c(best, worst))



# A tibble: 16 x 4

# Groups:   subject [4]

   subject question type  alternative

     <int> <chr>    <chr> <chr>      

 1       1 01       best  01         

 2       1 02       best  04         

 3       2 01       best  02         

 4       2 02       best  03