Those expressions answer different completion rules. Choose the rule from the approved scoring specification before choosing the function.
Separate a valid zero from the missing code
This original fictional practice codebook allows numeric item responses 0 through 4. It reserves 99 for an unanswered item and defines a score as the mean of at least two valid responses. It is an arithmetic exercise, not a validated survey instrument.
| Respondent | q1 | q2 | q3 | Expected score under this rule |
|---|---|---|---|---|
| A | 0 | 2 | 99 | 1, from two valid responses |
| B | 4 | 99 | 99 | Missing, only one valid response |
| C | 1 | 3 | 2 | 2, from three valid responses |
For A, zero contributes to the mean: (0+2)/2=1. Removing it as if it were missing would leave only one valid response and change eligibility. For B, reporting 4 would ignore the two-answer minimum.
Declare user-missing values before computing the score
For a working copy with numeric variables q1, q2 and q3, the proposed syntax is:
MISSING VALUES q1 q2 q3 (99).
COMPUTE valid_items = NVALID(q1, q2, q3).
COMPUTE practice_score = MEAN.2(q1, q2, q3).
EXECUTE.
IBM documents NVALID in its missing-value functions reference. The expected valid_items values are 2, 1 and 3 for A, B and C. Before using the real data, check that values outside 0 through 4 or the declared missing codes have been identified; declaring 99 missing does not validate every other number.
IBM's MISSING VALUES command reference, checked October 24, 2026, describes user-missing specifications and their immediate effect. A later specification for the same variable replaces the earlier one. Review existing declarations so a new command does not unintentionally remove other missing codes already defined by the codebook.
IBM's missing-values expression documentation distinguishes ordinary arithmetic, MEAN and a minimum-valid-argument suffix such as MEAN.2. In the fixture, (q1+q2+q3)/3 becomes missing for A; MEAN(q1,q2,q3) would score B as 4. Neither matches all three rows of the supplied policy.
Request syntax tied to the actual codebook
Draft SPSS scoring syntax for review on a working copy.
SPSS version and numeric item variables: [ENTER]
Approved scoring rule, minimum valid items and rationale: [PASTE]
Allowed values and meaning of zero: [ENTER]
User-missing and system-missing rules: [PASTE]
Existing missing-value declarations and relevant syntax: [PASTE]
Permitted sample rows with IDs and expected scores: [PASTE]
Reverse-scored items or weighting, if any: [PASTE OR NONE]
Explain which records receive a score and which remain missing.
Preserve raw responses. Include a valid-item count, explicit handling
of the supplied missing codes, and checks for out-of-range values.
Do not invent a completion threshold, reverse-code an item, impute
responses, overwrite data or claim the scale is validated. Distinguish
proposed syntax and expected arithmetic from an actual SPSS run.
Use the scoring specification to set the threshold; ChatGPT should not choose a convenient minimum because it produces fewer missing scores.
Compare the respondent-level score and valid count with the fixture in your SPSS working copy. These are authored expectations with locally checked arithmetic, not recorded SPSS output. Keep the accepted syntax, codebook version and missing-value declarations together. The statistics guide then addresses a separate decision: which study design and interpretation the resulting scores can support.