This is NOT the missing-value code in your data. All codes NOT in CODES= are missing value codes. Use this control specification when you want missing data to be treated as valid responses. Winsteps and Missing Data: No Problem!
Winsteps processes one observation at a time. For each observation, Xni by person n on item i, it computes an expectation Eni, based on the current person measure estimate Bn and the current item measure Di and, if relevant, the current rating (or partial credit) scale structure (calibrations) {Fk}. Pnik is the probability of observing category k for person n on item i.
In this computation it skips over, omits, ignores "missing" data.
It then compares sum(Xni) with sum(Eni) for each person n, and adjusts Bn.
It then compares sum(Xni) with sum(Eni) for each item i, and adjusts Di
It then compares the count of (Xni=k) with the sum (Pnik) for each k, and adjusts Fk
These sums and counts are only over the observed data. There is no need to impute missing data.
There are no pairwise, listwise or casewise deletions associated with missing data.
MISSCORE= says what to do with characters that are not valid response codes, e.g. blanks and data entry errors. Usually any characters not in CODES= are treated as missing data, and assigned a value of -1 which mean "ignore this response." This is usually what you want when such responses mean "not administered". If they mean "I don't know the answer", you may wish to assign missing data a value of 0 meaning "wrong", or, on a typical attitude survey, 3, meaning "neutral" or "don't know".
MISSCORE=255 is the same as MISSCORE=-1
MISSING=0 is the same as MISSCORE=0 meaning that all codes in the data not listed in CODES= are to be scored 0.
Non-numeric codes included in CODES= (without rescoring/recoding) or in NEWSCORE= or IVALUE= are always assigned a value of "not administered", -1.
Example 0a: In my data file, missing data are entered as 9. I want to score them 0, wrong answers. Valid codes are 0 and 1.
CODES = 01 ; do not specify a 9 as valid
MISSCORE = 0 ; specifies that all codes not listed in CODES=, e.g., 9's. are to be scored 0.
Example 0b: In my data file, missing data are entered as 9. I want to ignore them in may analysis. Valid codes are 0 and 1.
CODES = 01 do not specify a 9 as valid
; the following line is the standard, it can be omitted.
MISSCORE = -1 specifies that all codes not listed in CODES=, e.g., 9's.
are to be treated as "not administered"
Example 1: Assign a code of "0" to any responses not in CODES=
MISSCORE=0 missing responses are scored 0.
Example 2: In an attitude rating scale with three categories (0, 1, 2), you want to assign a middle code of "1" to missing values
MISSCORE=1 missing responses scored 1
Example 3: You want blanks to be treated as "wrong" answers, but other unwanted codes to be ignored items, on a questionnaire with responses "Y" and "N".
CODES="YN " blank included as valid response
NEWSCORE=100 new response values
RESCORE=2 rescore all items
MISSCORE=-1 ignore missing responses (standard)
Example 4: Your optical scanner outputs an "@" if two bubbles are marked for the same response. You want to ignore these for the analysis, but you also want to treat blanks as wrong answers:
CODES ="1234 " blank is the fifth valid code
KEY1 =31432432143142314324 correct answers
MISSCORE=-1 applies to @ (standard)
Example 5: Unexpected codes are scored "wrong", but 2's to mean "not administered".
CODES = 012
NEWSCORE= 01X ; X is non-numeric, matching 2's ignored
MISSCORE= 0 ; all non-CODES= responses scored 0
Example 6: You have a long 4-option MCQ test with data codes ABCD. Most students do not have the time to complete all the items. This requires a two-stage item analysis:
1. Estimate the item difficulties without allowing missing data to bias them
Missing data = -1 ; not administered. Perform the analysis estimate the item difficulties. Save them with IFILE=if.txt
2. Estimate the person measures lowering the estimates if there is missing data
Missing data = 0 ; incorrect. Anchor the item difficulties with IAFILE=if.txt to estimate the person measures and report them.
So, in this Example:
Stage 1. Item calibration:
Deliberately skipped responses are coded "S" and scored incorrect. The student could not answer the question.
Not-reached items are coded "R" and scored "not administered". This prevents easy items at the end of the test being calibrated as "very difficult".
CODES="ABCDS"
KEY1="CDBAD....."
MISSCORE=-1
IFILE=ITEMCAL.TXT ; write out the item calibrations
Stage 2. Person measurement:
The convention with MCQ tests is that all missing responses are scored incorrect when measuring the persons.
IAFILE=ITEMCAL.TXT ; anchor on the Stage 1 item calibrations
CODES="ABCDS"
KEY1="CDBAD....."
MISSCORE=0 ; all missing data are scored incorrect
