Marketing Research
Methods Summary
- University of Groningen -
Hi, I'm Santi. A Graduate from the University of Groningen in
MSc Marketing of the 2016/17 class, which I graduated cum laude (all grades 8).
This was partly due to my ambition and partly due to these scripts that I wrote.
They have allowed me to memorize the important concepts effectively and
helped out friends and fellow students to obtain high grades and
eventually become successful even in their resits.
I hope this script will help you, too!
Disclaimer:
This script has been concluded from the material available on Nestor,
the indicated lecture slides may deviate from yours.
The script claims no guarantee for completion or success.
You are responsible to study those concepts that you don’t understand.
Author:
Santi Helps
,Correlation
A correlation tells us something about the extent to which some variables are related to each other,
as well as how variables are related to each other. This lets us know whether a change in one
variable is met with similar changes in another variable. A correlation can have a value ranging
from -1 to 1, where -1 means a perfect negative and 1 means a perfect positive correlation. 0
means there is no relationship at all.
- correlation is usually important first step for further research (i.e. regression)
- use only if both variables are at interval level (not nominal or ordinal)
Chi Square
A Chi Square test – also often called a cross table test – helps you to make a decision on whether
there is a relation between two variables. A Chi Square test is used when your data is nominal or
categorical. It is important to remark that a chi square is to be used when you have two categorical
variables from a single population. (i.e. ask voters for political preference & their gender -> this way
find out if gender is related to one’s voting preference).
- compare the observed and expected division of numbers of the cells
- when difference b/w observed and expected values is small -> probably no relationship
- when difference b/w observed and expected values is large -> might be a relationship
- with Chi Square test we can find out whether difference is significant or not
- Question: “is there a significant difference b/w observed & expected values over the cells?”
,Factor analysis - In General
A factor analysis can be used in two different ways:
(1) Data Reduction. Reduce a large set of data to a smaller subset of variables that are more
easily managed and measured, while retaining as much of the original information as possible.
A factor analysis provides us with individuals’ score on this different subset of variables. Most
often, you make use of an existing concept and you want to check whether your data shows
the same sub dimensions as stated in literature. Factor analysis therefore helps in verifying that
your concepts indeed consist of a fixed amount of factors, and you want to work with the sub
dimensions in subsequent regression analysis. You perform the factor analysis as a pre-
measurement.
- a priori criterion (upfront classification of number of dimensions) is best option in this case
- factor analysis in this way merely serves as a check and start of empirical research
(2) Explorative Research. If you want to make a new concept scale (i.e. for Burnout) because you
argue that existing scales are outdated, a factor analysis is of appropriate use. A factor analysis
helps us in understanding the structure of a set of variables (identifying structure). The factor
analysis has the purpose of showing the amount of sub dimensions that underlie your new
developed construct. In that case, the factor analysis will be your main statistical test and the
amount of factors that need to be defined are supposed to be left open.
- In this case, latent [not observable] root criterion (Eigenvalue > 1) is most appropriate
- This criterion states that underlying factors should not be pre-determined
- The extraction of number of factors is not an exact science. Different methods can be applied.
- A scree plot can give a good idea of how many factors can be distinguished within the analysis
In General:
- Condition for factor analysis: the minimum is to have at least five times as many observations as
there are variables (items) to be analyzed, but a 10-to-1 (ratio) is preferred.
- Uses interval data (continuous) and commonly employs Likert Scales (5 or 7-point)
- No categorical or nominal data! (Likert scale is ordinal, but assumed to be interval data)
- Can indicate convergent validity when factor loadings are > 0.70
- => This means our measures are measuring what they should
- Convergent validity: variables within a single factor are highly correlated
- Can indicate discriminant validity when correlations < 0.30 between constructs
- => This means that there is not much overlap between two constructs
- Discriminant validity: the extent to which a measure does not correlate with other constructs
from which it is supposed to differ
- Factor rotation is a method within factor analysis to improve the interpretation of the results
, Factor analysis (means finding common variance) - specific
Before beginning any analysis, we need to understand what the underlying concept of our study
means for the consumer (i.e. privacy, innovativeness) -> this should always be the first step!
Use factor & reliability analysis because often variables are interdependent (=multicollinearity) so
that we can boil these items down for good further analysis (regression, conjoint, cluster, Anova)
Construct / concept / latent variable
Innovativeness
If concept goes up,
items go up
Reflective scale (concept reflected in items)
Item Item Item
1 2 3
Ask the items (questions) in a survey to obtain information about the underlying concept.
Stages for data analysis
1. Inspection of data (items): which variables, measurement scales, coding scheme
2. Cleaning dataset: missing values, outliers..
3. Combining items into new dimensions (e.g. factor / reliability)
4. Final analysis, testing hypothesis (i.e regression analysis with the newly created dimensions)
-> Crap in, crap out! (If data preparation is crap in the beginning, the final analysis will be crap!)
Ex. 12 items (questions) for 1 factor should boil down to a couple of dimensions (i.e. 4)
No distinction is made b/w dependent (Y) and independent (X) variables! FA is usually applied to
your independent variables (X). There is no causal relation b/w the variables.
Why can we reduce the number of variables?
- Because two items are highly correlated, combine them to get:
• Parsimoniousness
• Less multicollinearity (in subsequent analysis)
• Combine variables into factors
Dimensionality = Is the # of dimensions in the underlying factor analysis
Transforming variables into factors cuts multicollinearity, but we still have too many factors left.
Therefore only choose strongest factors (although information will be lost)
Ex.
Items: X1, X2, X3, X4, X5, X6 -> can be expressed as linear combination called Factor
Factors: F1, F2, F3, F4, F5, F6 Note:
Strongest F: F1 & F2 Items = variables = survey questions
Lost info: F3, F4, F5, F6 Dimensions = factors = components
