PAIRED= correction for paired comparison data = No

Paired comparison data is entered as only two observations in each row (or each column). The raw score of every row (or column) is identical. In the simplest case, the "winner" receives a '1', the "loser" a '0', and all other column (or rows) are left blank, indicating missing data.

 

Example 1: Data for a chess tournament is entered. Each row is a player. Each column a match. The winner is scored '2', the loser '0' for each match. For draws, each player receives a '1'. See Bradley-Terry model.

PAIRED=YES ; paired comparisons

  CODES=012  ; valid outcomes

  NI=56  ; number of matches

 

Example 2: From a research paper - mcg.lboro.ac.uk/mji/files/BERJpre.pdf

" for each question used in the main study (38 in total)"

"Judges ... 250 pairwise comparisons of question responses via the comparative judgement website. 5000 pairwise judgement decisions were collected in total." = 20 judges.

 

So we follow www.rasch.org/rmt/rmt113o.htm with Winsteps:

 

1. The 38 questions are 38 Winsteps items

2. The 5000 judgements are 5000 "persons". Each person has two responses: "1" (yes/more/better) and "0" (no/less/worse)

3. Each of the 20 judges has a code in the person label, so we can summarize the judges with Winsteps Table 28

4. Since the data are pairwise, in the Winsteps control file,

PAIRED=YES