Skip to contents

Prepare severity estimates for SAD validation

Source: R/sad_conventional.R
sad_data.Rd

Organises visual severity estimates made with and without a standard area diagram (SAD), or with two or more SADs, into the long format used by the sad_* family, and checks the data for common problems.

Usage

sad_data(
  data,
  rater = "rater",
  image = "leaf",
  actual = "actual",
  estimate = "estimate",
  method = "method",
  wide = NULL,
  group = NULL,
  sep = "_"
)

Arguments

data

A data frame in long or wide format.

rater, image, actual

Names of the columns with the rater identifier, the image (leaf, fruit, ...) identifier and the actual severity (%).

estimate

Name of the column with the estimates (long format).

method

Name of the column with the assessment method (long format), for example "unaided" and "aided".

wide

Character vector with the estimate columns in wide format. Each column becomes one level of method. If the vector is named, the names are used as the method labels, e.g. c(UN = "unaided", SAD = "aided").

group

Optional name of a column that splits raters into groups, for example which SAD each rater used. When given, condition is method and group pasted together.

sep

Separator used to build condition from method and group.

Value

A tibble of class sad_data with columns rater, image, actual, estimate, method, group, condition and error (estimate minus actual, in percentage points), and attribute design.

Details

Each combination of method and group is a condition. The function warns when an image has more than one actual severity value, when estimates fall outside 0 to 100%, when a rater x image x condition combination is repeated, and when a rater lacks one of the methods. The design is labelled "paired" when raters assessed the images under more than one condition (the usual unaided then aided study), and "independent" otherwise.

See also

sad_ccc(), sad_compare(), sad_fit()

Examples

x <- sad_data(SADRustSoybean, rater = "rater", image = "leaf",
              wide = c(UN = "unaided", SAD = "aided"), group = "sad")
x
#> SAD validation data (paired design): 40 raters, 50 images, 4000 estimates
#> Conditions: UN_Old, UN_New, SAD_Old, SAD_New 
#> # A tibble: 4,000 × 8
#>    rater image actual estimate method group condition  error
#>    <fct> <fct>  <dbl>    <dbl> <fct>  <fct> <fct>      <dbl>
#>  1 1     1       0.25      5   UN     Old   UN_Old     4.75 
#>  2 1     1       0.25      0.6 SAD    Old   SAD_Old    0.35 
#>  3 1     2       2.5      15   UN     Old   UN_Old    12.5  
#>  4 1     2       2.5       5   SAD    Old   SAD_Old    2.5  
#>  5 1     3       7.24     10   UN     Old   UN_Old     2.76 
#>  6 1     3       7.24      7   SAD    Old   SAD_Old   -0.240
#>  7 1     4       7.31     20   UN     Old   UN_Old    12.7  
#>  8 1     4       7.31     10   SAD    Old   SAD_Old    2.69 
#>  9 1     5       9.07     15   UN     Old   UN_Old     5.93 
#> 10 1     5       9.07     18   SAD    Old   SAD_Old    8.93 
#> # ℹ 3,990 more rows