Says what characters to recognize as valid codes in your data file, together with CODERANGE=. If XWIDE=1 (the standard), use one column/character per legitimate code. If XWIDE=2, use two columns/characters per valid code. Characters in your data not included in CODES= or CODERANGE= are given the MISSCORE= value. The highest category number 32767. CODES= can be up to 4,000 characters long.  If CODES= and CODERANGE= are both specified, then the data are matched to CODES= first.  CODES= can be rescored. CODERANGE= cannot be rescored.

 

Initially, Winsteps assumes that all items share the same response structure.

If there are 2 response codes in CODES=, then this is the Rasch dichotomous model

If there are 3 or more response codes in CODES=, then this the Andrich rating-scale model.

To override those,

If ISGROUPS=0, then this is the Masters partial-credit model. Each item is modeled to have its own rating-scale structure.

If ISGROUPS = AABBACDDCAADEEA, then the items are grouped to share rating scales.

The "A" group are items 1,2,5, 10,11,15 - they share the same rating scale.

The "B" group are items 3,4 - they share the same rating scale.

The "C" group ....

 

See also these Examples written with CODERANGE=

 

Example 1: A test has four response choices. These are "1", "2", "3", and "4". All other codes in the data file are to be treated as "item not administered". Each response uses 1 column in your data file. Data look like: 134212342132.3212343221

   XWIDE=1  one character wide (the standard)

   CODES=1234 four valid 1-character response codes

 

Example 2: There are four response choices. Each response takes up 2 columns in your data file and has leading 0's, so the codes are "01", "02", "03" and "04". Data look like: 0302040103020104040301

   XWIDE=2  two characters wide

   CODES=01020304 four valid 2-character response codes

 

Example 3: There are four response choices entered on the file with leading blanks, so that codes are " 1", " 2", " 3", and " 4". Data look like: 3 2 4 2 1 3 2

XWIDE=2  two characters wide

CODES=" 1 2 3 4" " required: blanks in 2-character responses

 

Note: when XWIDE=2 or more, both CODES= and the data value are left-aligned before matching, so both " 1" and "1 " in CODES= match both " 1" and "1 "in your data file.

 

Example 4: Your data is a mixture of both leading blanks and leading 0's in the code field, e.g. "01", " 1", " 2", "02" etc. The numerical value of a response is calculated, where possible, so that both "01" and " 1" are analyzed as 1.

Data look like:  02 1 20102 1 2 01

XWIDE=2  two characters wide

CODES=" 1 2 3 401020304" two characters per response

 

Example 5: Your valid data are 1,2,3,4,5 and your missing data codes are 7,8,9 which you want reported separately on the distractor tables.

 CODES = 12345789

 NEWSCORE = 12345XXX ; missing values scored with non-numeric values

 

Example 6: The valid responses to an attitude survey are "a", "b", "c" and "d". These responses are to be recoded "1", "2", "3" and "4". Data look like: adbdabcd

 CODES   =abcd four valid response codes

 NEWSCORE=1234 new values for codes

 RESCORE=2  rescore all items

 

Typically, "abcd" data implies a multiple choice test. Then KEY1= is used to specify the correct response. But, in this example, "abcd" always mean "1234", so that the RESCORE= and NEWSCORE= options are easier to use.

 

Example 7: Five items of 1-character width, "abcd", then ten items of 2-character width "AA", "BB", "CC", "DD". These are preceded by person-id of 30 characters. Data look like:

George Washington Carver III  dabcdBBAACCAADDBBCCDDBBAA

 FORMAT=(30A1,5A1,10A2) Name 30 characters, 5 1-chars, 10 2-chars

 XWIDE =2   all converted to 2 columns

 CODES ="a b c d AABBCCDD" "a" becomes "a "

 NEWSCORE="1 2 3 4 1 2 3 4 " response values

 RES EDFILE= edit data file CORE=2   rescore all items

 NAME1=1   name starts column 1 of reformatted record

 ITEM1=31   items start in column 31

 NI=15   15 items, all XWIDE=2

 

Example 8: Items are to rescored according to Type A and Type B. Other items to keep original scoring.

 IREFER = AAAAAAAABBBBBBBBCCCCCCC ; 3 item types

 CODES   = 1234 Original codes in the data file

 IVALUEA = 1223 Recode Type A items

 IVALUEB = 1123 Recode Type B items

 IVALUEC = 1234 Recode Type * item. Can be omitted

 

Example 9: The valid responses are percentages in the range 00 to 99.

 XWIDE = 2  two columns each percent

 ; for data with leading zeroes,

 CODES = 0001020304050607080910111213141516171819+

  +2021222324252627282930313233343536373839+

  +4041424344454647484950515253545556575859+

  +6061626364656667686970717273747576777879+

  +8081828384858687888990919293949596979899

or, for data with blanks,

 CODES =" 0 1 2 3 4 5 6 7 8 910111213141516171819+

  +2021222324252627282930313233343536373839+

  +4041424344454647484950515253545556575859+

  +6061626364656667686970717273747576777879+

  +8081828384858687888990919293949596979899"

 

also recommended:

 ISRANGE = *

 1 0 99 ; defines the full range of the rating scale, regardless of the data

 *

 SFUNCTION=4 ; defines a smooth function for the Rasch-Andrich thresholds

 

Example 10: Codes are in the range 0-254.

XWIDE=3 ; 3 characters per response: 

 ; for data with leading zeroes,

CODES="000001002003004005006007008009010011012013014015016017018019020021022023+

      +024025026027028029030031032033034035036037038039040041042043044045046047+

      +048049050051052053054055056057058059060061062063064065066067068069070071+

      +072073074075076077078079080081082083084085086087088089090091092093094095+

      +096097098099100101102103104105106107108109110111112113114115116117118119+

      +120121122123124125126127128129130131132133134135136137138139140141142143+

      +144145146147148149150151152153154155156157158159160161162163164165166167+

      +168169170171172173174175176177178179180181182183184185186187188189190191+

      +192193194195196197198199200201202203204205206207208209210211212213214215+

      +216217218219220221222223224225226227228229230231232233234235236237238239+

      +240241242243244245246247248249250251252253254"

;or for data with blanks

CODES="  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23+

      + 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47+

      + 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71+

      + 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95+

      + 96 97 98 99100101102103104105106107108109110111112113114115116117118119+

      +120121122123124125126127128129130131132133134135136137138139140141142143+

      +144145146147148149150151152153154155156157158159160161162163164165166167+

      +168169170171172173174175176177178179180181182183184185186187188189190191+

      +192193194195196197198199200201202203204205206207208209210211212213214215+

      +216217218219220221222223224225226227228229230231232233234235236237238239+

      +240241242243244245246247248249250251252253254"

 

also recommended:

 ISRANGE = *

 1 0 254 ; defines the full range of the rating scale, regardless of the data

 *

 SFUNCTION=4 ; defines a smooth function for the Rasch-Andrich thresholds

 

Example 11: The data include negative values, -10, -9, through to positive codes 9, 10. Winsteps can only analyze data that are positive integers, so the data would need to be rescored:

XWIDE=3 ; each observation is 3 characters wide in the data file

; the next line is so that I can be sure that each code is 3 characters wide

;         123123123123123123123123123123123123123123123123123123123123123

CODES   ="-10-9 -8 -7 -6 -5 -4 -3 -2 -1 0  1  2  3  4  5  6  7  8  9  10 "

NEWSCORE="0  1  2  3  4  5  6  7  8  9  10 11 12 13 14 15 16 17 18 19 20 "

also recommended:

 ISRANGE = *

 1 0 20 ; defines the ful range of the rating scale, regardless of the data

 *

 SFUNCTION=4 ; defines a smooth function for the Rasch-Andrich thresholds

 

Example 12: Codes in the range 0 - 1000.

 ; CODES= is impractical. Use CODERANGE= 0-1000