2. Data

Yuki Atsusaka and Seo-young Silvia Kim

In this vignette, we show what the typical input data look like for the rankingQ package using the identity dataset.

rankingQ assumes a dataset that contains (1) responses to ranking questions with J items and (2) a binary indicator for whether each respondent provides the correct answer to an anchor-ranking question, an auxiliary ranking question whose correct answer(s) are known to researchers. For example, the package features a dataset identity, which stores real-world survey responses to a question that asks 1,082 respondents to rank four sources of identity (partisanship, race, gender, and religion) based on their relative importance to them.

library(rankingQ)
data("identity")

Target ranking question

Ranking data are expected to be in the wide format, where multiple columns are used to represent different items and their values represent the items’ marginal ranks.

For example, in identity, the four sources of identity are stored in party, religion, gender, and race. The first respondent ranked party first, gender second, race third, and religion fourth, so the full ranking profile app_identity is "1423" given the reference choice set of party-religion-gender-race.

head(identity)
#> # A tibble: 6 × 16
#>   s_weight app_identity party religion gender  race anc_identity household
#>      <dbl> <chr>        <dbl>    <dbl>  <dbl> <dbl> <chr>            <dbl>
#> 1    0.844 1423             1        4      2     3 1234                 1
#> 2    0.886 1423             1        4      2     3 1234                 1
#> 3    2.96  3412             3        4      1     2 1234                 1
#> 4    0.987 1423             1        4      2     3 1234                 1
#> 5    1.76  4132             4        1      3     2 1324                 1
#> 6    0.469 3124             3        1      2     4 1234                 1
#> # ℹ 8 more variables: neighborhood <dbl>, city <dbl>, state <dbl>,
#> #   anc_correct_identity <dbl>, app_identity_recorded <chr>,
#> #   anc_identity_recorded <chr>, app_identity_row_rnd <chr>,
#> #   anc_identity_row_rnd <chr>

Anchor ranking question

To perform bias correction, the data must have what we call the anchor ranking question. The anchor question is an auxiliary ranking question that looks similar to the target question, whose “correct” answer is known to researchers. For example, identity has responses to the anchor question that asked respondents to rank four nested units: household, neighborhood, city, and state. These responses are included in household, neighborhood, city, and state. Here, the correct answer is assumed to be "1234". Based on these responses, we code an indicator variable (anc_correct_identity) that takes 1 if respondents offer the correct answer and 0 otherwise.

In the following dataset, the first and third respondents have provided an incorrect answer for the anchor question, whereas the rest have provided the correct answer.

identity[
  ,
  c("household", "neighborhood", "city", "state", "anc_correct_identity")
] |>
  tail()
#> # A tibble: 6 × 5
#>   household neighborhood  city state anc_correct_identity
#>       <dbl>        <dbl> <dbl> <dbl>                <dbl>
#> 1         3            1     2     4                    0
#> 2         1            2     3     4                    1
#> 3         4            3     2     1                    0
#> 4         1            2     3     4                    1
#> 5         1            2     3     4                    1
#> 6         1            2     3     4                    1