Top of page
Skip to main content
Contact

Likert, semantic differential and Stapel: response formats, coding and scoring

These formats can all produce numeric columns in SPSS while asking respondents to perform different tasks. Selecting a point count before defining the construct can create data that are convenient for software but difficult for participants.

SCALE AND SURVEY DESIGNPrepared by Lê Minh5 minUpdated
CENTRAL ARGUMENT

Response format belongs to measurement design. Align content, anchors, direction and missing-data rules; assigning numbers from 1 to 5 does not by itself establish equal distances.

Start reading

Response formats should fit the concept

  1. Likert

    Agreement with a statement

  2. Stapel

    Rating around a single attribute

  3. Semantic differential

    Position between opposite meanings

Conceptual diagram. Response format alone does not determine all analysis assumptions.

Source: Lê Minh · Conceptual illustration, not software output.

1. Separate an item format from a multi-item scale

An agreement statement is a Likert-type item. A multi-item scale additionally requires justification for combining items to represent a construct (DeVellis & Thorpe, 2022). Not every 1–5 question is Likert: frequency ratings ask a different question.

“The provider honors its commitments” calls for an agreement judgment. “How many late responses occurred last month?” calls for recall over a defined period. Five response circles do not make those measurements equivalent.

Cognitive interviews and pretesting can examine understanding, recall and response selection (Krosnick & Presser, 2010). A reliability coefficient after fieldwork cannot replace those checks.

2. Compare three response tasks

Table 1. Original teaching examples, not validated scales
FormatIllustrationCheck
Likert-type agreement“The provider honors commitments”: 1 strongly disagree to 5 strongly agreeA specific statement with one main idea
Semantic differentialDifficult to use 1 2 3 4 5 6 7 Easy to useOpposite poles on the same dimension
Stapel“Reliable”: select +5…+1 or −1…−5Understand positive/negative description and distinguish missing codes

Semantic differentials use contrasting descriptors; Stapel uses a descriptor with positive and negative ratings. The traditional ten-point Stapel illustration excludes zero (Clow & James, 2014). Report and test adaptations such as adding a zero category.

“Fast” and “professional” are not opposites on one dimension. A service can be fast but unprofessional. Placing these descriptors at opposite ends prevents respondents from expressing that combination.

3. Choose categories and distinguish nonresponse

No point count is optimal merely because a particular analysis is planned. Consider respondents’ ability to discriminate, the administration mode and the measurement objective. Test labels on an actual phone rather than only on a desktop questionnaire.

Neutrality is an attitudinal position. “Do not know”, “not used” and “not applicable” represent other states. Coding them as the midpoint mixes attitude with unavailable information. Give them separate codes and declare their analytical treatment.

An ordered response is not automatically interval-scaled. Analysis depends on aggregation, distributions and the measurement model. Comparisons of ordinal CFA estimators are conditional on data characteristics, not a universal “five categories imply ML” rule (Li, 2016).

4. Reverse the intended meaning and valid codes

Suppose four 1–5 items should all indicate higher trust. A negatively directed Q3 requires Q3R = 6 − Q3 for valid responses. This is the arithmetic of this hypothetical design, not a recommendation to add negative wording to every scale.

Hypothetical example: Q1 = 4, Q2 = 5, Q3 = 2, Q4 = 3 ⇒ Q3R = 4 ⇒ mean = 4.

If 99 means “do not know”, applying 6 − 99 creates −93. Address invalid or missing codes first. The same formula cannot reverse a Stapel scale; define the transformation from its coding range and intended meaning.

Reverse coding cannot repair ambiguous wording. When pilot respondents misunderstand a negative item, revisit the instrument and its evidence.

5. SPSS example: at least three valid items out of four

The following template adopts an illustrative minimum of three valid items. MEAN.3 specifies the required valid count and RECODE explicitly maps values (IBM, n.d.). The minimum is not a universal rule for scales.

Template: replace variable names and instrument-specific rules

* Illustrative four-item scoring rule; validate original codes first.
MISSING VALUES Q1 Q2 Q3 Q4 (99).
RECODE Q3 (1=5)(2=4)(3=3)(4=2)(5=1)(ELSE=SYSMIS) INTO Q3R.
COMPUTE SCORE=MEAN.3(Q1,Q2,Q3R,Q4).
EXECUTE.

First check that Q1, Q2 and Q4 contain only 1–5 or declared missing codes. This RECODE protects Q3 only. Retain the original Q3, label Q3R, and examine frequencies before and after conversion.

A respondent with two valid items must have a missing SCORE under this rule; three valid items use a denominator of three. Check individual rows and valid-item counts. The syntax is instructional and is not claimed as output executed on a user dataset.

6. Pilot and report the measurement

  • Define construct, respondent and reference period.
  • Use one main idea per item.
  • Specify anchors and distinguish neutrality from lack of knowledge.
  • Record original codes, reverse items and final direction.
  • Declare valid-item and missing-data rules before examining results.
  • Test comprehension on the intended device and population.
  • Assess structure and validity before aggregation or comparison.

Reporting template

The construct [ ] used [number] items with [format] and anchors [ ]. Item [ ] was reversed after code validation. The [mean/sum/model] score required [ ] valid items; [do not know/not applicable] responses were handled through [rule]. Piloting and measurement evidence comprised [ ].

Continue: SERVQUAL and SERVPERF. A useful format helps respondents express their experience and analysts understand what each value represents.

References