Tentamenstof Marketing Research Methods
Week 1 - experimental design & data preparation
Academic research
Academic research = studying a subject in detail to discover new information or reach a
new understanding → producing new knowledge
→ follows a logical “hourglass” structure, starting broad, narrowing down to data collection
and analysis, then broadening again for discussion
The research flow
Problem (action-oriented) → RQ (information-oriented) → literature (theoretical background)
→ conceptual model (structured overview) → propositions (qualitative) or hypotheses
(quantitative)
Conceptualization = drawing boundaries around terms to make them tangible and
measurable to remove vagueness and ambiguity
→ conceptual model defines relationship between key constructs
1. Concepts and dimensions = what you measure
2. Relations between them:
- Dependent/independent variable
- Antecedents/outcomes
- Moderating variable = change the strength/direction of X → Y
- Mediating variables = explain why X affects Y (the underlying process)
Operationalization = translating a concept into something measurable to make abstract
concepts measurable and allow replication across studies (define how variables are
quantitatively measured, using existing validated scales and indicators)
Measurement level = determines which statistical test can be used
1. Nominal = labels and tags with no order
2. Ordinal = Ordering categories with direction
3. Interval = equal distance between categories, but no true zero
4. Ratio = all interval scale properties with a true zero point
Experimental research design
Research design types
- Exploratory = discover ideas and insights → interviews, qualitative methods
- Descriptive = describe market characteristics → surveys, panels
- Causal = determine cause-effect relationship → experiments
Causality conditions
1. Concomitant variation = X and Y vary together with no third variable
correlated
, 2. Time order = X precedes Y, must happen before the effect
3. Absence of other causes = control for alternative explanations (achieved via
random assignment)
Experiments = manipulation of one or more IVs and measurement of their effect on one or
more DVs while controlling all other variables
→ strong causal inference, but low external validity
1. What IVs (manipulations) will be used, and how?
2. How will we measure the DVs?
3. How will we control extraneous variables?
4. Who are the test units (participants)?
5. Will we use a within-subjects or between-subjects design?
Experimental design
1. Within-subjects design = repeated measures; each participant experiences all levels
of the IV
→ controls for individual differences, increases statistical power and requires fewer
participants, but threats to validity (maturation and instrumentation effects)
2. Between-subjects design = each participant experiences only one level of the IV
→ avoids contamination across conditions, but requires more participants (>75 per
cell) and randomization needed to ensure fairness
Internal validity = whether the IV truly cause the effect on the DV
External validity = generalizability of results to other settings, populations or stimuli
→ higher internal validity means lower external validity
Experiment types
- True experiments = fully controlled, random assignment, allows process
measurement
- Field experiment = random assignment in real-world settings; less control
- Quasi-experiments = no random assignment; lower internal validity
Fully randomized design = testing the effect of a single experimental variable
Full factorial design = testing multiple IVs that may interact with each other
Handling missing data
- Listwise deletion = remove entire cases
- Pairwise deletion = exclude only in analyses where data is missing (SPSS default)
- Imputation = replace with mean value
, Week 2 - PCA, Factor Analysis & Reliability Analysis
Why do we need data reduction?
Marketing concepts (brand image, attitude, preferences etc) often measured using
multiple-item scales (questions)
→ Too many overlapping variables cause
- Multicollinearity = IVs highly correlated
- Complexity (hard to interpret results)
- Noise (redundant information)
Methods to reduce data complexity while keeping as much information as possible
1. PCA = solely data reduction, not explaining; find uncorrelated variables
(components) that explain maximal variance in the data → often used to create
perceptual maps
2. EFA = interpretation; looking for a small number of dimensions that capture a large
part of the variance, while trying to make the dimensions interpretable in terms of
the original variables
3. Reliability analysis = testing the reliability of the found underlying dimension by
measuring the internal consistency of a known set of items in each dimension
→ implemented after PCA/EFA
Standardization = rescaling variables so that they all have the same unit of measurement: a
mean of 0 and a standard deviation of 1 (z-score transformation)
→ puts all variables on the same scale, ensuring they contribute equally to the analysis.
Principal component analysis (PCA)
= transforms many (to some extent) correlated variables into a smaller number of
uncorrelated components → explain the maximum possible variance in the data; linear
combinations of standardized variables (C1, C2, etc), ranked according to initial variability
they explain to keep the strongest ones
Variance explained = how much of the total variability in the data is captured by each
component
Total variability = sum of variability of the individual variables → variability of the
components is smaller, because part of the information is lost
Steps in PCA
1. Standardize the variables = to ensure each variable has equal weights
→ so that all variables have a mean = 0, variance = 1
Week 1 - experimental design & data preparation
Academic research
Academic research = studying a subject in detail to discover new information or reach a
new understanding → producing new knowledge
→ follows a logical “hourglass” structure, starting broad, narrowing down to data collection
and analysis, then broadening again for discussion
The research flow
Problem (action-oriented) → RQ (information-oriented) → literature (theoretical background)
→ conceptual model (structured overview) → propositions (qualitative) or hypotheses
(quantitative)
Conceptualization = drawing boundaries around terms to make them tangible and
measurable to remove vagueness and ambiguity
→ conceptual model defines relationship between key constructs
1. Concepts and dimensions = what you measure
2. Relations between them:
- Dependent/independent variable
- Antecedents/outcomes
- Moderating variable = change the strength/direction of X → Y
- Mediating variables = explain why X affects Y (the underlying process)
Operationalization = translating a concept into something measurable to make abstract
concepts measurable and allow replication across studies (define how variables are
quantitatively measured, using existing validated scales and indicators)
Measurement level = determines which statistical test can be used
1. Nominal = labels and tags with no order
2. Ordinal = Ordering categories with direction
3. Interval = equal distance between categories, but no true zero
4. Ratio = all interval scale properties with a true zero point
Experimental research design
Research design types
- Exploratory = discover ideas and insights → interviews, qualitative methods
- Descriptive = describe market characteristics → surveys, panels
- Causal = determine cause-effect relationship → experiments
Causality conditions
1. Concomitant variation = X and Y vary together with no third variable
correlated
, 2. Time order = X precedes Y, must happen before the effect
3. Absence of other causes = control for alternative explanations (achieved via
random assignment)
Experiments = manipulation of one or more IVs and measurement of their effect on one or
more DVs while controlling all other variables
→ strong causal inference, but low external validity
1. What IVs (manipulations) will be used, and how?
2. How will we measure the DVs?
3. How will we control extraneous variables?
4. Who are the test units (participants)?
5. Will we use a within-subjects or between-subjects design?
Experimental design
1. Within-subjects design = repeated measures; each participant experiences all levels
of the IV
→ controls for individual differences, increases statistical power and requires fewer
participants, but threats to validity (maturation and instrumentation effects)
2. Between-subjects design = each participant experiences only one level of the IV
→ avoids contamination across conditions, but requires more participants (>75 per
cell) and randomization needed to ensure fairness
Internal validity = whether the IV truly cause the effect on the DV
External validity = generalizability of results to other settings, populations or stimuli
→ higher internal validity means lower external validity
Experiment types
- True experiments = fully controlled, random assignment, allows process
measurement
- Field experiment = random assignment in real-world settings; less control
- Quasi-experiments = no random assignment; lower internal validity
Fully randomized design = testing the effect of a single experimental variable
Full factorial design = testing multiple IVs that may interact with each other
Handling missing data
- Listwise deletion = remove entire cases
- Pairwise deletion = exclude only in analyses where data is missing (SPSS default)
- Imputation = replace with mean value
, Week 2 - PCA, Factor Analysis & Reliability Analysis
Why do we need data reduction?
Marketing concepts (brand image, attitude, preferences etc) often measured using
multiple-item scales (questions)
→ Too many overlapping variables cause
- Multicollinearity = IVs highly correlated
- Complexity (hard to interpret results)
- Noise (redundant information)
Methods to reduce data complexity while keeping as much information as possible
1. PCA = solely data reduction, not explaining; find uncorrelated variables
(components) that explain maximal variance in the data → often used to create
perceptual maps
2. EFA = interpretation; looking for a small number of dimensions that capture a large
part of the variance, while trying to make the dimensions interpretable in terms of
the original variables
3. Reliability analysis = testing the reliability of the found underlying dimension by
measuring the internal consistency of a known set of items in each dimension
→ implemented after PCA/EFA
Standardization = rescaling variables so that they all have the same unit of measurement: a
mean of 0 and a standard deviation of 1 (z-score transformation)
→ puts all variables on the same scale, ensuring they contribute equally to the analysis.
Principal component analysis (PCA)
= transforms many (to some extent) correlated variables into a smaller number of
uncorrelated components → explain the maximum possible variance in the data; linear
combinations of standardized variables (C1, C2, etc), ranked according to initial variability
they explain to keep the strongest ones
Variance explained = how much of the total variability in the data is captured by each
component
Total variability = sum of variability of the individual variables → variability of the
components is smaller, because part of the information is lost
Steps in PCA
1. Standardize the variables = to ensure each variable has equal weights
→ so that all variables have a mean = 0, variance = 1
