Paired Comparisons with Standard Rasch Software

Rasch computer programs that can handle incomplete data (such as responses made to computer-adaptive tests) can also analyze paired comparisons, such as consumer preferences and sports competitions.

If you are doing a paired comparison analysis that models the comparisons directly, then no adjustment to the standard errors or measures is necessary.

For instance, using a Facets one-facet model for paired comparisons:
Facets=2 ; two facet positions in the data
Entered = 1,1 ; both facet positions are for the same facet
models = ?, -?, D ; elements of the one facet oppose each other.

The Facets estimates are exactly correct without adjustments.

More robust, especially if you are doing a paired comparison analysis with ties (draws, equalities) and using a Facets one-facet model for paired comparisons:
 
Facets=2 ; two facet positions in the data
Entered = 1,1 ; both facet positions are for the same facet
models = ?, -?, R2, 0.5 ; elements of the one facet oppose each other. Each observation is twice in the data file. 2=win, 1=tie, 0=loss.
Data=
; every observation twice in the data:
4, 5, 2 ; 4 plays 5 and wins
5, 4, 0 ; 5 plays 4 and loses
4, 6, 1 ; 4 plays 6 and ties
6, 4, 1 ; 6 plays 4 and ties

Set up your data matrix so that each row corresponds to a player (team, competitor, brand, procedure, etc.) and each column an occasion (match, game, aspect of quality, consumer, etc.). The rows can be used for occasions and the columns for players, if more convenient.

There are as many columns as there are occasions (matches, etc.). In each column make only two entries, the results: enter a "1" in the winner's row and a "0" in the loser's. The column total is always 1. Each row total is that player's total number of wins.

The test length is the number of occasions. The resulting measures for the columns, the occasions, can be ignored, but large column misfit flags unexpected outcomes. The measures for the rows, the players, are in logits, but doubled by this estimation method. The reported standard errors are 1.4 times too large.

Chess Matches at the Venice Tournament, 1971. Each column is one match.
1D.0..1...1....1.....1......D.......D........1.........1.......... Browne
0.1.D..0...1....1.....1......D.......1........D.........1......... Mariotti
.D0..0..1...D....D.....1......1.......1........1.........D........ Tatai
...1D1...D...D....1.....D......D.......D........1.........0....... Hort
......010D....D....D.....1......D.......1........1.........D...... Kavalek
..........00DDD.....D.....D......D.......1........D.........1..... Damjanovic
...............00D0DD......D......1.......1........1.........0.... Gligoric
.....................000D0DD.......D.......1........D.........1... Radulov
............................DD0DDD0D........0........0.........1.. Bobotsov
....................................D00D00001.........1.........1. Cosulich
.............................................0D000D0D10..........1 Westerinen
.......................................................00D1D010000 Zichichi
Figure 1. Match results of a chess tournament

Here is confirmation that the reported logits need to be divided by two. Player n of ability Bn wins Rn times against Player m of ability Bm, who wins Rm times. By direct application of the Rasch model (or the Bradley-Terry-Luce paired comparison model):

Bn - Bm = loge(Rn/Rm) (1)

Whenever persons n and m meet, there results an occasion column, o, with two observations in it, one "1" and one "0", summing to 1. Thus all occasions on which n and m meet produce columns with the same raw score of 1 across the same two rows, n and m, and so have the same occasion difficulty, Do. Because any wins for player m against player n appear in the data matrix as losses for player n on occasions with difficulty Do, an ability estimate for player n, B'n, is given by

B'n - Do = loge(Rn/Rm) (2)

Similarly, because each win for player n against player m appears in the data matrix as a loss for player m on an occasion with difficulty Do, an ability estimate for player m, B'm, is given by

B'm - Do = loge(Rm/Rn) (3)

Yielding,

Do = (B'n + B'm)/2 (4)

and

B'n - B'm = 2*loge(Rn/Rm) = 2*(Bn - Bm) (5)

confirming that the reported estimates, B'n and B'm are twice as large as the direct Rasch estimates, Bn and Bm.

If your Rasch software supports measure rescaling, an adjustment of 0.5 can be made, but this adjustment produces an S.E. that is 71% of its correct value.

Figure 1 shows the match results from a chess tournament. The columns represent chess matches in the order they were reported in the magazine, "Chess". Drawn matches have been ignored. A halving adjustment produces the measures in Table 1. Browne has the highest ability measure with 7 wins in 8 non-drawn matches. Hort's and Zichichi's results show misfit because Hort unexpectedly lost to Zichichi.

Wins Non-Draw
Matches
Adj.
Measure
Adj.
S.E.
OUTFIT
Mn-Sq
Players
7
4
6
4
5
2
2
3
3
2
2
1
8
5
8
6
7
4
6
7
9
8
9
5
2.78
2.25
1.76
1.55
1.47
.25
-.36
-.98
-1.40
-1.85
-2.02
-2.70
.83
1.00
.70
.77
.74
1.08
.86
.76
.57
.71
.66
.83
.32
9.90
.35
.47
.37
.15
.17
.60
.31
.30
8.67
.57
Browne
Hort
Mariotti
Kavalek
Tatai
Damjanovic
Radulov
Gligoric
Cosulich
Westerinen
Zichichi
Bobotsov

Table 1. Measures for chess players from non-drawn matches.

Score from 11 matches Adj.
Measure
Adj.
S.E.
OUTFIT
Mn-Sq
Players
17
15
14
14
13
11
10
9
8
8
7
6
1.09
.68
.56
.56
.33
-.01
-.17
-.34
-.51
-.51
-.69
-.88
.36
.33
.32
.32
.31
.31
.31
.31
.31
.31
.33
.34
1.02
.96
1.54
.83
.80
.37
.91
.52
1.00
1.15
.89
1.90
Browne
Mariotti
Hort
Tatai
Kavalek
Damjanovic
Gligoric
Radulov
Bobotsov
Cosulich
Westerinen
Zichichi

Table 2. Measures for chess players from all matches.

Paired comparison with ties (draws, "no preference", etc.) are equally simple. Merely enter "2" for win, "1" for tie, "0" for loss. Each column (occasion) now has a score of 2. Rescoring the chess data matrix from "1D0" to "210" yields the adjusted measures in Table 2. A reasonable adjustment to the initially reported logits is: logit measures halved, standard errors divided by 1.4. Comparing Table 2 with Table 1, the 5 times that Hort drew with weaker players have dragged him down below Mariotti. His loss to Zichichi is no longer as surprising, so the OUTFIT mean-squares are more reasonable.

John M. Linacre

Available Rasch software is listed at www.rasch.org/software.htm.


Paired Comparisons with Standard Rasch Software. Linacre J.M. … Rasch Measurement Transactions, 1997, 11:3 p. 584-5.




Rasch Publications
Rasch Measurement Transactions (free, online) Rasch Measurement research papers (free, online) Probabilistic Models for Some Intelligence and Attainment Tests, Georg Rasch Applying the Rasch Model 3rd. Ed., Bond & Fox Best Test Design, Wright & Stone
Rating Scale Analysis, Wright & Masters Introduction to Rasch Measurement, E. Smith & R. Smith Introduction to Many-Facet Rasch Measurement, Thomas Eckes Invariant Measurement: Using Rasch Models in the Social, Behavioral, and Health Sciences, George Engelhard, Jr. Statistical Analyses for Language Testers, Rita Green
Rasch Models: Foundations, Recent Developments, and Applications, Fischer & Molenaar Journal of Applied Measurement Rasch models for measurement, David Andrich Constructing Measures, Mark Wilson Rasch Analysis in the Human Sciences, Boone, Stave, Yale
in Spanish: Análisis de Rasch para todos, Agustín Tristán Mediciones, Posicionamientos y Diagnósticos Competitivos, Juan Ramón Oreja Rodríguez

To be emailed about new material on www.rasch.org
please enter your email address here:

I want to Subscribe: & click below
I want to Unsubscribe: & click below

Please set your SPAM filter to accept emails from Rasch.org

www.rasch.org welcomes your comments:

Your email address (if you want us to reply):

 

ForumRasch Measurement Forum to discuss any Rasch-related topic

Go to Top of Page
Go to index of all Rasch Measurement Transactions
AERA members: Join the Rasch Measurement SIG and receive the printed version of RMT
Some back issues of RMT are available as bound volumes
Subscribe to Journal of Applied Measurement

Go to Institute for Objective Measurement Home Page. The Rasch Measurement SIG (AERA) thanks the Institute for Objective Measurement for inviting the publication of Rasch Measurement Transactions on the Institute's website, www.rasch.org.

Coming Rasch-related Events
May 17 - June 21, 2024, Fri.-Fri. On-line workshop: Rasch Measurement - Core Topics (E. Smith, Winsteps), www.statistics.com
June 12 - 14, 2024, Wed.-Fri. 1st Scandinavian Applied Measurement Conference, Kristianstad University, Kristianstad, Sweden http://www.hkr.se/samc2024
June 21 - July 19, 2024, Fri.-Fri. On-line workshop: Rasch Measurement - Further Topics (E. Smith, Winsteps), www.statistics.com
Aug. 5 - Aug. 6, 2024, Fri.-Fri. 2024 Inaugural Conference of the Society for the Study of Measurement (Berkeley, CA), Call for Proposals
Aug. 9 - Sept. 6, 2024, Fri.-Fri. On-line workshop: Many-Facet Rasch Measurement (E. Smith, Facets), www.statistics.com
Oct. 4 - Nov. 8, 2024, Fri.-Fri. On-line workshop: Rasch Measurement - Core Topics (E. Smith, Winsteps), www.statistics.com
Jan. 17 - Feb. 21, 2025, Fri.-Fri. On-line workshop: Rasch Measurement - Core Topics (E. Smith, Winsteps), www.statistics.com
May 16 - June 20, 2025, Fri.-Fri. On-line workshop: Rasch Measurement - Core Topics (E. Smith, Winsteps), www.statistics.com
June 20 - July 18, 2025, Fri.-Fri. On-line workshop: Rasch Measurement - Further Topics (E. Smith, Facets), www.statistics.com
Oct. 3 - Nov. 7, 2025, Fri.-Fri. On-line workshop: Rasch Measurement - Core Topics (E. Smith, Winsteps), www.statistics.com

 

The URL of this page is www.rasch.org/rmt/rmt113o.htm

Website: www.rasch.org/rmt/contents.htm