Marketing Research

2023-2024

Statistics Review
Assessing Relationships
between Variables

Carla Carvalho Martins

 Marketing Research
Curricular Unit Presentation 2:
Final Presentation
+ Test

Communicating
Research Findings
Presentation 1:
Pitch Survey fieldwork; Quantitative
Analysis I, II and III
Qualitative Research
Techniques I, II; Analysis
Qualitative Research: Nature,
Approaches
Secondary Research

Marketing Research Problems from Management

Decision Problems; Research Design

Marketing Research Process

Introduction to Marketing Research

BIVARIATE STATISTICS – ESTABLISHING RELATIONSHIPS BETWEEN VARIABLES

We can categorize statistical procedures for analysing relationships among variables into two major groups:

A. GROUP COMPARISONS
Assessing and quantify differences between two or more groups or situations (the independent variable is necessarily
categorical).

B. CORRELATION (AND REGRESSION*) ANALYSES

Assessing the strength and direction of the associations between two or more variables, and, in many cases, to make
predictions based on the observed data.

They may be used alone or complement each other.

* Regression is out of the scope of these classes

3
EXAMPLES OF GROUP COMPARISONS - when the dependent variable (DV) is metric
Do women evaluate customer service quality differently from men?
(=Gender influences customer service quality evaluation?)
Independent variable – categorical nominal = GENDER
Dependent variable – metric = PERCEIVED CUSTOMER SERVICE QUALITY

4
EXAMPLES OF GROUP COMPARISONS - when the DV is ordinal
Is one gender more educated than the other?
(=Gender impacts education?)
Independent variable – categorical nominal = GENDER
Dependent variable – categorical ordinal = EDUCATION

5
EXAMPLES OF GROUP COMPARISONS - when the DV is nominal
Is one gender more prone to cancel the credit card than other?
(=Gender influences the likeliness of credit card cancellation?)
Independent variable – categorical nominal (2 categories) = EDUCATION
Dependent variable – categorical nominal = CANCELATION OF CARD

6
EXAMPLES OF GROUP CORRELATIONS – with two metric variables - Pearson Correlation Coefficient
How is global attitude towards the credit card related with customer service perceived quality?

It varies from -1 to 1

Pearson Correlation Coefficient

7
EXAMPLES OF GROUP CORRELATIONS – two ordinal variables (or one metric and one ordinal) –
Spearman Correlation Coefficient and Kendall Tau-b
How is customer income related with customer education?

They vary from -1 to 1

Spearman Coefficient

Kendall Tau-B

8
EXAMPLES OF GROUP CORRELATIONS – two nominal variables (or one nominal and one ordinal)
Phi coefficient and Cramer’s V
How is customer income related with customer education?

They vary from 0 to 1

9
STATISTICAL INFERENCE

Statistical inference
Statistical procedures through which inferences about a population are made based on certain statistics calculated
from a sample of data drawn from that population.

In statistical inference, we wish to make statements not merely about the particular subjects observed but also,
more importantly, about the larger population of subjects from which the study participants were drawn.

10
STATISTICAL INFERENCE

Parameter
Population Measure of a certain attribute in the target
The collection of all the elements or objects that population.
comprise the universe for the purpose of the A parameter denotes the true value that would be
research problem. obtained if a census rather than a sample was
undertaken.

Sample Statistic
A subgroup of the elements of the population Measure of a certain attribute in a sample. The
selected for participation in the study. sample statistic is used as an estimate of the
population parameter.

11
STATISTICAL INFERENCE

Statistical inference uses calculations based on probability theory to make

conclusions.

Statisticians studied the characteristics of sampling distributions

A sampling distribution is a probability distribution of a statistic obtained through

a large number of samples drawn from a specific population.

Knowing the characteristics of these sampling distributions enable us to find the

probability of getting a sample (of a given size n) with a certain statistics value,
given a hypothetical value of the parameter (population).

Example of a sampling distribution – sampling distribution of the

mean statistic
Source: Field, A. (2005), Discovering Statistics using SPSS
STATISTICAL INFERENCE

In inference statistics, based on a statistic obtained in a sample, taken from that population,
researchers infer for the whole population, with a certain degree of confidence (by convention, 95%
or 99%) . Inference is done through one or both of the following procedures.

a. Statistical \ Hypotheses tests (with p values)

p value = plausibility of the null hypothesis in hypothesis testing. It is the probability of getting the statistic obtained in
data, if the parameter assumed a given value, which is hypothesized in the null hypothesis.
The lower the p value, the lower the chance that the null hypothesis is true.
When we test the association between variables, the null hypothesis state that there is no association between them. So,
to find support for the existence of a statistical association between variables, researchers always look for low p values.

b. Confidence intervals
A range of plausible values for an unknown parameter that has an associated confidence level that the true parameter is in
the proposed range.
STATISTICAL INFERENCE

Statistical test
Statistical procedures (based on the mathematical knowledge about sampling distributions), that enable testing hypotheses about the
value of a parameter.
Conducting a statistical test, entails running the following steps:

STEP 0: Choose a confidence level for your conclusions (for the rejection of the null hypothesis).
Convention: 95% (the most frequent), 99%, or, occasionally, 90%.
1st STEP: Set the Hypothesis, decomposing it in two:
H0 - null hypothesis
H1 – alternative hypothesis
2nd STEP: Compute the appropriate statistic and find p value
Choose the apropriate statistical test and check p value
3rd STEP: Conclude
Accept or reject the H0 and conclude about the population parameter.
Statistical test
Step 0: Establishing the confidence level

Unfortunately, when working with samples, there is no such thing as a 100% confidence level, since this would require
a sample size as large as the entire population.

The most commonly selected confidence levels are 95% (0.95) and 99% (0.99). In most cases the confidence level used
by default is 95% . The 99% confidence level is usually reserved for pharmaceutical testing and other fields where the
consequences of an incorrect conclusion are more severe.

To a given level of confidence, corresponds a given significance level (alpha)

Alpha =α = probability of committing a type I error (probability of rejecting the null hypothesis, when that is, in fact true)
α = 1 – confidence level (1 minus the level of confidence). Therefore:
- A 95% level of confidence corresponds to an α of 0.05
- A 99% level, corresponds to an α of 0.01
Statistical test
Step 1: Setting the Hypothesis (decomposed in two)

Hypotheses are always about the population, i.e., about parameters - unknown characteristics of variables or relations between variables in
the population. (We know the sample statistics. So we do not need to hypothesize about it).

Theoretically, each Hypothesis is decomposed in two hypotheses

• H0: Null hypothesis

Is the hypotheses that is, actually, going to be, statistically, tested.
Always states an equality.

• H1: Alternative Hypothesis

Always states de opposite of the H0.
It is, generally, the hypotheses “of interest” to the researcher.
H1 is not directly tested. The acceptance of H1 is dependent on the rejection of H0 (which is the hypothesis being tested). H1 is assumed
or not assumed, depending on the result of the test of the H0. Only if the H0is rejected – we assume the alternative hypothesis
Statistical test
Step 2 - Compute the appropriate statistics and find p value

When doing statistical testing, the statistical software computes the statistic (measure in the sample studied) along with
the corresponding p value (based on the mathematical knowledge about sampling distributions).

p value is the probability of getting a given sample statistic (the statistic found in the sample under study) or an even more
extreme value, if the null hypothesis was true in the population.
Statistical test
Step 3: Conclude

If the p value is very low, i.e., below or equal to the significance level adopted (5% or less, when using the 95% confidence level), we will
assume that the null hypothesis is not true (reject H0) and assume the alternative (confirm H1).

If the p value is above the significance level, we consider the nonexistence of enough evidence to reject the null hypothesis. It would be
incorrect, however, to conclude that since the null hypothesis is not rejected, it can be accepted as valid. In classical hypothesis testing, there
is no way to determine whether the null hypothesis is true.

Summing
• If p > α → assume H0 (not enough evidence to reject it)
up, for any statistical test
(no matter the parameter • If p<= α → Reject the H0. Assume H1
tested) (with the probability of p of being making an error, i.e, of rejecting the null hypothesis when that is actuallly true)
INFERENCE PROCEDURES TO TESTS RELATIONSHIPS BETWEEN VARIABLES

A. GROUP COMPARISONS STATISTICAL TESTS

B. CORRELATION (AND REGRESSION) STATISTICAL TESTS

19
A. GROUP COMPARISONS STATISTICAL TESTS

The primary objective of statistical tests for group comparisons is to determine whether differences between two or more
groups or situations regarding a variable of interest, that were found in samples, are statistically significant.

The appropriate tests must be chosen based on the…

… the dependent variable type of data

If Metric – use t tests, ANOVA

If Ordinal – use nonparametric tests (Mann-Whitney, Wilcoxon, Kruskall Wallis, Friedman… )

If Nominal – use nonparametric test Chi Square

… the types of groups being compared (independent or paired)

Independent samples – observations being compared are obtained from distinct and unrelated groups or subjects (e.g. difference of grades of women vs
men in a specific course)

Paired samples - observations that are somehow matched. Each observation in one group is paired or matched with a specific observation in the other
group (normally, the same group in two different conditions or treatments) (e.g. grades of the same group in two different courses)

20
 GROUP COMPARISONS STATISTICAL TESTS

Mean comparisons – used when the dependent variable is METRIC.

Mean comparison statistical tests

Type of Statistic |
Independent variable Nº of categories Test
Dependent. Type of groups Parameter under Statistical Test Effect size statistic
type (groups | situations) statistic
variable analysis

Independent | t
Between subjects
Independent samples Student t test
Two Mean Difference Cohen’s d
One Paired| Within
subjects
Paired samples Student t test t
Categorical Independent |
(nominal or ordinal) ANOVA (independent samples) F
Between subjects 𝜂2
Two or more Mean Differences
Paired| Within F
Metric subjects
Repeated Measures ANOVA (eta squared)

Independent | Multiple factor ANOVA (independent F

Between subjects samples) 𝜂2
2 or more Categorical Two or more Paired| Within
Mean Differences
Multiple factor Repeated Measures F
(nominal or ordinal) (eta squared)
subjects ANOVA
Independent and Paired Mixed ANOVA F

21
 GROUP COMPARISONS STATISTICAL TESTS

Median comparisons – used when the dependent variable is ORDINAL*.

Statistical methods used to assess and analyze differences in the central tendencies of different groups or conditions based on the median
values of the data (instead of the mean).

Median comparison statistical tests

Dependent Independent Nº of categories Statistic | Parameter Statistical Test Test
Variable Type variable type (groups | situations) under analysis statistic
Independent | U
Between subjects
Wilcoxon-Mann-Whitney
Two Median Difference
Paired| Within W
subjects
Wilcoxon (paired samples)
Categorical
Ordinal*
(nominal or ordinal) Independent | H
Between subjects
Kruskal-Wallis
Two or more Median Difference
Paired| Within 𝝌𝟐
subjects
Friedman

*Or metric, when the assumption of normality of DV data is questionable, especially if samples are small (<30 per group)

22
GROUP COMPARISONS STATISTICAL TESTS

Frequency (proportion) comparisons – used when the dependent variable is CATEGORICAL (nominal or ordinal)*.

Involve assessing and statistically analysing the differences in proportions or percentages between different groups or conditions.
Proportions represent the fraction of a total that exhibits a particular characteristic or outcome (of the dependent variable).

Frequency comparison statistical tests

Dependent Independent Nº of categories Statistic | Parameter Statistical Test Test statistic

variable type variable type (groups | situations) under analysis
Independent | Between
subjects
Chi square for association 𝝌𝟐
Two Frequency Differences
Categorical Categorical Paired| Within subjects
(situations)
McNamar's Chi-Square 𝝌𝟐
(nominal or ordinal) (nominal or ordinal)
Independent | Between
Two or more subjects Frequency Differences Chi square for association 𝝌𝟐

23
B. CORRELATION STATISTICAL TESTS
Used to test whether the correlation found between two variables in the observed data is statistically significant.

V1 V2 Statistic and Parameter Statistical test Test statistic

Metric Metric Pearson Correlation Pearson Correlation r
Spearman Correlation Spearman Correlation rS
Ordinal (or metric) Ordinal (or metric)
Kendall’s Rank Correlation Kendall’s Rank Correlation 𝝉
Pearson Phi correlation Phi 𝝓
Nominal (or ordinal) Nominal (or ordinal)
Kramer’s V correlation Cramer’s V V

24