Overview

• Reverse Coding

  • What?
  • Why?
  • How?

• Composites

  • What?
  • Why?
  • How?

Reverse Coding - What & Why?

• When we design questionnaires & scales we have the choice of phrasing items positively or negatively

• A positively-phrased item could be:

• Whereas a negatively-phrased item could be:

• Both Qs measure the same underlying construct, but a high score on these two items reflects opposing feelings (i.e., high like or high dislike)

Reverse Coding - What & Why?

• Reverse Coding (or reverse scoring) is where we change the numeric values assigned to negatively-phrased items so that a high score reflects the same type of response across all items in our questionnaire

• With reverse coding, we're not changing participants' data, but just the numeric values we have assigned to the responses for a specific item

Reverse Coding - What & Why?

• Let's say we have the two example items below in our questionnaire, where strongly disagree is scored as 1 and strongly agree is scored as 5:

• Assuming our participant feels consistently positive about R, then they would get a score of 5 for Q1 & a score of 1 for Q2

• So if we calculated their mean score across these two items, we'd get 3, which doesn't reflect their positive feelings towards R i.e., it looks as though they feel pretty neutral towards R, instead of reflecting their R love

Reverse Coding - What & Why?

• But we can reverse code/score the negatively-phrased item by flipping around the scoring

• Now, the first item is still scored as strongly disagree = 1 and strongly agree = 5, but our second item is now scored in the opposite direction where strongly disagree = 5 and strongly agree = 1

• Now if we calculated the mean score across these items for our R LovR participant, they would score a 5, which accurately reflects their consistently positive feelings towards R

Any Questions?

Reverse Coding - How?

• There are multiple ways we can reverse code our variables in R:

  • If we have numeric data, we can use simple maths or we can use the dplyr::recode() function - just like in the skills lab last week..

  • If we have factor data, we can use the forcats::relevel() function

  • If we have character data, we can use dplyr::recode()

• We can, but we don't need to reverse code items before we do a factor analysis (FA), but we must reverse code negative items before we do a reliability analysis (RA)

• We also need to reverse code items before creating composites - but more on that later!

Reverse Coding - How?

Step 1: Look at the factor loadings:

Reverse Coding - How?

Step 2: Decide which, if any, variables are negatively related

Reverse Coding - How?

Step 2: Decide which, if any, variables are negatively related

Reverse Coding - How?

Step 3: Recode our items

# method 1 - using simple maths on a 5-point scale 
data <- data |>
  dplyr::mutate(column_name = 6-column_name)Copy Code

# method 2 - using recode
data <- data |> 
  dplyr::mutate(column_name = dplyr::recode(
    column_name, `1` = 5, `2` = 4, `3` = 3, `4` = 2, `5` = 1))Copy Code

Step 4: Check recoding

Use the table() function to compare frequencies before and after recoding to check it has worked

Creating Composites - What & Why?

• Composite Scores are usually created by calculating the mean of a set of items from a scale or questionnaire

• We can calculate composite scores from all items on a scale

  • We could have a measure of "Love for R" made up of all 50 survey items

• Or we can create composites from different subsets of items that make up different factors/subscales

  • Maybe our measure of "Love for R" can be broken down into 3 factors/subscales of "Love for Stats", "Love for Computers", and "Coolness"

• We look at the results of our factor analysis to decide how items can be combined into a composite (i.e., if we have one overall score, or different subscales), or we can use the scoring instructions for any pre-validated scales that we're using

Creating Composites - What & Why?

• For example, we could have 4 items of 'r_love' that we can use to create a composite measure by calculating the mean of those items for each participant

• We can then use these composites to represent a given construct in further analyses/models

Any Questions?

Creating Composites - How?

• We can use a combination of different functions to create a composite score:

  • dplyr::mutate(), dplyr::rowwise(), mean(), c() & c_across()

• Before creating composites we should also reverse score any negative items because the scoring differs between positive and negative items and the mean of them would not be accurate

• We often have to make some decisions around how we choose to handle missing data when creating composites

  • I.e., whether we'll calculate the mean regardless of any missing data, whether we'll calculate it for only complete cases, or whether we'll allow some missing data

  • This is entirely up to the researcher, or the scale creator (there may be specific recommendations you need to follow for scales created by others)

Creating Composites - How?

# method 1 - using columns next to eachother
comp_scores <- data |> 
  dplyr::rowwise() |>
  dplyr::mutate(column_name1_comp = mean(c_across(item1:item3)),
                column_name2_comp = mean(c_across(item4:item5)))Copy Code

# method 2 - using columns NOT next to eachother
data <- data |>
    dplyr::rowwise() |>
    dplyr::mutate(column_name1_comp = mean(c(item1, item2, item3)),
                  column_name2_comp = mean(c(item4, item5)))Copy Code

That's all - happy wrangling!

Artwork by @allison_horst