Mastering Market Research: Measurement, Methods, and Scales

Posted on Feb 26, 2025 in Psychology and Sociology

Understanding Measurement in Market Research

Nominal: Classification without judgment (e.g., gender, mode).
Ordinal: Ordered classification (e.g., ranking sodas). Distance between ranks is unknown (median).
Interval: Equal spacing between values (e.g., Fahrenheit, “on a scale of…”). Scales must remain within the same category (mean).
Ratio: Fixed and meaningful zero point (e.g., income, age). Allows for all mathematical operations.

Demographics: Age, income, education.
Lifestyle: Activities, interests, opinions.
Awareness/Knowledge:
- Unaided recall (best).
- Aided recall.
- Recognition.
Attitudes: Feelings about a brand (“How do you feel?”).
Motivation: Reasons behind actions.
Purchase Intentions & Behaviors: Foot traffic, click rates, etc.

Avoid open-ended questions and disclosing the brand (to prevent *auspices bias*).

Observation: Less versatile, sometimes forced (if observing over the shoulder). Independent of a person’s willingness to participate.
Questionnaire: Versatile for various types of data.

Auspices Bias: Knowing the company name and giving favorable answers.
History Effect: External events influencing results.
Maturation Effect: Participants changing over time.
Mortality: Participants dropping out (e.g., only half return surveys).
Selection Bias: Trying to match the selection of stores in a test market.
Instrument Variation: Changes in the measurement instrument.
Experimenter Bias: Researcher influencing results.
Testing Effects: Participants improving due to repeated testing (e.g., IQ scores).

Reliability:
- Split-half measures reliability but could allow marketers to only use favorable data.
- Cronbach’s alpha is better (range: -1 to 1).
Validity:
- Pragmatic Validity: Extent to which our measurement correlates to certain criterion variables.
- Predictive Validity: Measurement or test used to predict an outcome/criterion.
- Content Validity: (Non-statistical) Did our research cover the correct subject (e.g., a spelling bee)? Did performance exceed your expectations?

Likert Scale: “Strongly Disagree” to “Strongly Agree.” Efficient, ordinal.
Semantic Differential: Two labels at opposite ends of a scale (e.g., “Low Quality” to “High Quality”). Ordinal, but it’s harder to argue for equal intervals.
Rank Order: Ranking items (e.g., “Rank sodas 1-4”). Can be problematic; ordinal, and it’s unknown if levels are equal. Ties may not be understood as possible.
Stapel Scale: One word or phrase (e.g., “High Quality”) rated from -5 to +5. Ordinal, but in practice, it may be treated as interval.
Constant/Fixed Sum: Allocate points/money, etc., to attributes.

Number of Items: More items can increase reliability.
Flipping Items: Reversing the scale direction for some items can mitigate response set bias, but be cautious of confusion.
Scale Points: Fewer than five scale points can limit variance.
Odd vs. Even: Odd-numbered scales have a midpoint; even-numbered scales avoid “no opinion” answers.
“Don’t Know” Option: Be careful with “don’t know” options, as they can be an easy way out for respondents.

Cannot include all topics.
Avoid high drop-off rates.
Determine what data you *need*.
Structured/Undisguised: Use closed-ended and unmasked questions. Easier for the respondent.
Content:
- Do we need to ask this?
- Will they know the answer?
- Avoid vague questions (e.g., “When did you last buy…?”).
- Avoid sensitive topics.
Form of Question:
- Open-ended vs. Closed-ended. Open-ended questions have high variance and are hard to analyze.
Wording:
- Be simple.
- Avoid ambiguity.
- Avoid leading questions.
- Avoid implicit alternatives or assumptions.
- Avoid generalizations.
- Avoid double-barreled questions.
- Be specific.
Sequencing: Order questions logically. Start with broad, general questions, then move to more specific ones.