Statistical Analysis for Market Research

Introduction

Statistical analysis converts market observations into evidence for managerial decisions. In brand management, researchers may need to estimate market size, compare brand perceptions, identify customer segments, assess relationships among attitudes and behaviours, predict purchase intention, or evaluate whether a campaign produced a meaningful change. Statistical analysis does not replace managerial judgement; it improves judgement by making uncertainty, patterns and competing explanations more explicit.

Market research data are commonly obtained through surveys, experiments, customer databases, social listening, retail audits and panel studies. These sources may contain categorical responses, ordered ratings, numerical measures and repeated observations. The appropriate statistical method depends on the research question, measurement scale, sampling design, distribution of the data and consequences of error.

A statistically significant result is not automatically important to a brand. A small effect may be statistically detectable in a very large sample but commercially negligible. Conversely, a strategically important effect may fail to reach conventional significance in a small sample. Professional interpretation therefore combines statistical evidence, effect size, uncertainty, research design and managerial relevance.

Learning objectives

After studying this chapter, the reader should be able to:

1. Distinguish descriptive statistics from inferential statistics.
2. Calculate and interpret measures of central tendency, dispersion and position.
3. Explain sampling error, estimation and confidence intervals.
4. Formulate and evaluate hypotheses using parametric and nonparametric tests.
5. Construct and interpret cross-tabulations, correlation and regression models.
6. Apply analysis of variance and select suitable post-hoc comparisons.
7. Assess assumptions, effect sizes, statistical power and managerial significance.
8. Communicate findings ethically and translate statistical results into brand decisions.

1. The logic of statistical analysis

Statistical analysis usually follows a sequence:

Research question → operationalisation → sampling and data quality → descriptive analysis → inferential analysis → interpretation → decision.

A research question should specify the population, variables and comparison or relationship of interest. For example, “Do environmentally concerned consumers show stronger preference for Brand A?” is more useful than “What do customers think?” The variables might include environmental concern, brand preference, age, income and purchase frequency.

A population is the complete set of units about which conclusions are intended. A sample is the subset actually observed. A parameter describes a population, whereas a statistic describes a sample. For example, the population mean brand-attitude score is a parameter; the sample mean is a statistic used to estimate it.

Measurement level affects method selection:

The distinction is not purely mechanical. A single Likert item is ordinal, while a reliable multi-item scale is often analysed as approximately continuous when categories are numerous, distributions are reasonable and the interpretation is cautious.

2. Descriptive statistics

Descriptive statistics organise observed data without generalising beyond them.

2.1 Frequency distributions

For a categorical variable, frequency is the number of observations in a category. A percentage is:

Missing observations should be reported separately. Percentages based only on valid responses may sum to 100%, whereas percentages based on the entire sample may not.

A frequency table should identify the variable, categories, valid sample size and treatment of missing values. Graphs such as bar charts are suitable for categorical data; histograms are suitable for quantitative data.

2.2 Central tendency

The arithmetic mean is:

where x_i is observation i and n is the number of observations. The mean is sensitive to extreme values and is most informative for quantitative data whose distribution is not severely skewed.

The median is the middle ordered observation, or the average of the two middle observations when n is even. It is robust to outliers and is often preferable for income, expenditure and highly skewed usage measures. The mode is the most frequent category and is especially useful for nominal data.

2.3 Dispersion

The range is maximum minus minimum, but it depends heavily on extreme observations. The sample variance is:

The sample standard deviation is:

The denominator n − 1 provides an unbiased estimator of population variance under standard random-sampling assumptions. Standard deviation is expressed in the same units as the variable. The interquartile range, Q₃ − Q₁, describes the central 50% and is resistant to extreme values.

The coefficient of variation is:

It is useful for comparing relative variability across positive measures with meaningful zero points. It should not be used when the mean is near zero or when the scale lacks a meaningful ratio interpretation.

Worked example: descriptive analysis

Suppose five respondents report monthly category purchases of 2, 3, 3, 4 and 8 units. The mean is 4 units and the median is 3 units. Deviations from the mean are −2, −1, −1, 0 and 4; squared deviations sum to 22. Thus:

The mean exceeds the median because the observation of 8 is high. A manager should therefore report both the typical value, represented by the median, and the variability rather than relying on the mean alone.

2.4 Standard scores and distributions

A standard score is:

where μ and σ are the population mean and standard deviation. In a sample, sample estimates may be used descriptively. A z-score indicates how many standard deviations an observation lies from the mean. It is useful for identifying unusual observations, although an unusual value is not necessarily erroneous.

Skewness, outliers, floor effects and ceiling effects should be examined using plots and substantive knowledge. Removing an observation solely because it is inconvenient is poor practice. Researchers should verify data entry, document exclusion rules and conduct sensitivity analyses.

3. Estimation and confidence intervals

A sample statistic provides a point estimate, but sampling variability means that another sample would produce a different result. An interval estimate communicates plausible values for the population parameter.

For a population mean with known population standard deviation, a confidence interval is:

where x̄ is the sample mean, σ is the population standard deviation, n is sample size, and z_(α/2) is the standard-normal critical value. In most market research applications σ is unknown, so the t distribution is used:

Here s is the sample standard deviation and the degrees of freedom are n − 1. The standard error of the mean is s ÷ √n. The interval is narrower when variability is lower or sample size is larger.

For a sample proportion p̂, a large-sample approximate interval is:

This approximation is less reliable when the sample is small or the proportion is close to 0 or 1; exact or improved intervals may then be preferable.

A 95% confidence interval does not mean there is a 95% probability that the fixed population parameter lies in this particular interval. Under repeated sampling and correct assumptions, approximately 95% of intervals constructed by the procedure would contain the parameter.

Worked example: confidence interval

A sample of 100 customers gives a mean brand-trust score of 6.2 on a seven-point scale, with s = 1.0. Using an approximate 95% critical value of 1.96:

The estimated population mean is approximately between 6.00 and 6.40 under the sampling and measurement assumptions. The interval does not establish that every customer has a score in this range.

4. Hypothesis testing

Hypothesis testing evaluates whether observed data are sufficiently inconsistent with a specified null model. The null hypothesis, H₀, commonly states that there is no difference, association or effect. The alternative hypothesis, H₁, states the research-relevant departure.

A significance level α is selected before analysis, often 0.05. The p-value is the probability, assuming H₀ and the model are true, of obtaining a result at least as extreme as the observed result. A p-value is not the probability that H₀ is true, nor does it measure effect size.

The general workflow is:

1. Define H₀ and H₁, including whether the test is one- or two-sided.
2. Select the test based on design, measurement and assumptions.
3. Set α before examining the result.
4. Calculate the test statistic and p-value.
5. Report confidence intervals and effect size.
6. Interpret statistical and managerial significance together.

A Type I error rejects a true H₀; its long-run probability is controlled approximately by α under the model. A Type II error fails to reject a false H₀. Statistical power is 1 − β, where β is the Type II error probability. Power increases with sample size, effect magnitude and measurement precision.

“Fail to reject H₀” should not be written as “prove no effect.” Evidence for practical equivalence requires an equivalence or non-inferiority design with a pre-specified meaningful margin.

Common mean comparisons

A one-sample t-test compares a sample mean with a reference value:

where μ₀ is the hypothesised population mean. A paired t-test compares two measurements from the same units by analysing differences dᵢ:

An independent-samples t-test compares means from two independent groups. Welch’s version is generally preferred when variances differ because it does not require equal population variances.

Assumptions include independent observations, an appropriate quantitative outcome and, for small samples, reasonably normal differences or group distributions. Large samples often make the mean robust to moderate non-normality, but dependence and serious outliers remain important problems.

5. Cross-tabulation and categorical analysis

A cross-tabulation displays the joint distribution of two categorical variables. Row percentages answer questions such as “Within each customer segment, what percentage prefers Brand A?” Column percentages answer a different question and must be labelled clearly.

The Pearson chi-square statistic is:

where Oᵢⱼ is the observed count in cell i,j and Eᵢⱼ is the expected count under independence. For a table with r rows and c columns:

Degrees of freedom are (r − 1)(c − 1). Expected counts should generally be sufficiently large; sparse tables may require category consolidation or an exact test. Chi-square significance indicates evidence of association, not causation.

Cramér’s V measures association strength:

It ranges from 0 to 1. Interpretation depends on context, table size and subject matter; universal cut-offs should be avoided.

6. Correlation

Correlation describes the direction and strength of a linear association between two quantitative variables. Pearson’s correlation is:

The value ranges from −1 to +1. Values near +1 indicate a strong positive linear association; values near −1 indicate a strong negative linear association; zero indicates no linear association, not necessarily no relationship.

Pearson correlation assumes paired observations, approximate linearity and no influential outliers. Spearman’s rank correlation is appropriate for ordinal data or monotonic relationships that are not adequately linear. Neither correlation establishes causation. A third variable, reverse causality, common method bias or selection effects may explain an association.

Always inspect a scatterplot. A correlation coefficient can conceal curvilinear patterns, clusters or a single influential case.

7. Regression for explanation and prediction

Simple linear regression models an outcome Y as a function of predictor X:

β₀ is the intercept, β₁ is the expected change in Y for a one-unit increase in X, and εᵢ is the error term. The fitted equation uses estimates b₀ and b₁. The slope is:

and the intercept is b₀ = ȳ − b₁x̄.

Multiple regression extends the model:

A coefficient represents the expected change in Y associated with a one-unit change in that predictor, holding the other included predictors constant. This conditional interpretation is not automatically causal.

The coefficient of determination is:

where SSE is the residual sum of squares and SST is the total sum of squares. R² is the proportion of sample outcome variation explained by the fitted model relative to a mean-only model. Adjusted R² penalises unnecessary predictors and is useful for comparing models with different numbers of predictors, but neither measure proves that a model is valid.

Important diagnostic assumptions include linearity, independent errors, constant variance, approximately normal residuals for small-sample inference and limited multicollinearity. Variance inflation factors can identify problematic predictor redundancy, though no single threshold is universally correct. Heteroscedasticity may require robust standard errors or transformation. Cross-validation or a holdout sample is preferable when prediction is the objective.

8. ANOVA and group comparisons

Analysis of variance tests whether three or more group means are all equal. The one-way ANOVA partitions variation into between-group and within-group components:

For k groups and total sample size n, degrees of freedom are k − 1 and n − k. The null hypothesis is that all population means are equal. A significant omnibus F-test indicates that at least one mean differs, but it does not identify which groups differ.

Post-hoc procedures such as Tukey’s method control familywise error for all pairwise comparisons under appropriate conditions. Planned contrasts are often more powerful when comparisons are specified in advance. Welch’s ANOVA is suitable when group variances differ. Repeated-measures ANOVA is used when the same units provide scores under several conditions, but it requires attention to dependence and, in some designs, sphericity.

Effect sizes include eta squared:

and, in some contexts, omega squared, which provides a less biased population estimate. Report group means, confidence intervals, effect size and post-hoc results rather than only an F statistic and p-value.

9. Nonparametric methods

Nonparametric methods are useful when outcomes are ordinal, distributions are strongly non-normal, samples are small or outliers make mean-based methods inappropriate. They are not assumption-free; they require suitable independence, ranking and distributional conditions.

The Mann–Whitney test evaluates rank distributions and is often described as a median comparison only when distributional shapes are similar. The Kruskal–Wallis test identifies whether at least one group’s rank distribution differs; post-hoc rank comparisons are needed to locate differences. Nonparametric tests may have lower power than suitable parametric tests when parametric assumptions are reasonably satisfied.

10. Effect sizes, power and managerial significance

Effect size expresses the magnitude of a difference or association independently, as far as possible, from sample size. For two independent means, Cohen’s d is:

where the pooled standard deviation is:

The pooled formula assumes comparable variances and independent groups. Alternatives such as Hedges’ g may reduce small-sample bias. For regression, standardised coefficients, semi-partial effects and changes in R² can be useful. For proportions, absolute percentage-point difference and relative risk may be more interpretable than a p-value.

A manager should ask:

• Is the effect large enough to change a decision?
• Is it economically or strategically meaningful?
• How precise is the estimate?
• Is the result robust across segments and specifications?
• Could bias, confounding or measurement error explain it?
• Does the proposed action justify the cost and risk?

Statistical significance is influenced by n; managerial significance depends on consequences, thresholds and context.

11. Applied case: evaluating a repositioning campaign

This is an illustrative case. A consumer electronics brand introduces a campaign positioning its headphones as premium yet sustainable. Management wants to know whether the campaign improved brand consideration and whether the effect differs by sustainability involvement.

Background and problem

A pre-campaign survey of 240 target consumers and a post-campaign survey of 260 comparable consumers measure consideration on a seven-point scale. The samples are independent, not repeated observations of the same respondents. Respondents also classify themselves into low, medium or high sustainability involvement.

Analysis approach

Researchers first inspect missing data, response patterns and distributions. They report means, standard deviations and confidence intervals. Because the samples are independent and group variances are not assumed equal, Welch’s t-test compares pre- and post-campaign consideration. They also calculate Cohen’s d and the absolute mean difference.

A cross-tabulation examines campaign awareness by involvement level, with a chi-square test and Cramér’s V. Finally, an ANCOVA-style regression predicts post-campaign consideration from campaign exposure, sustainability involvement and demographic controls. Since exposure is not randomly assigned, the regression is treated as associational rather than causal.

Suppose the pre-campaign mean is 4.8 and the post-campaign mean is 5.2, producing a difference of 0.4 points. The confidence interval for the difference is (0.12, 0.68), and the standardised effect is d = 0.24. The result may be statistically credible but modest. If a 0.5-point increase was the organisation’s pre-specified minimum meaningful improvement, the evidence does not establish that the campaign achieved the target.

Assume awareness is 42% among low-involvement, 58% among medium-involvement and 71% among high-involvement respondents. A significant chi-square result would indicate an association between involvement and awareness, not that involvement caused awareness. Managers might respond with broader reach tactics for low-involvement consumers while using sustainability proof points for high-involvement consumers.

Outcome and learning points

The appropriate conclusion is cautious: consideration increased after the campaign, but the size of the increase was small to moderate and the design cannot isolate campaign causality from time trends, competitor activity or sample composition. A stronger follow-up would use a randomized exposure experiment, a matched control market or a longitudinal panel.

Key lessons are that: (1) the research design constrains causal claims; (2) confidence intervals clarify precision; (3) effect sizes assess practical magnitude; (4) segmentation should be theoretically justified rather than discovered opportunistically; and (5) a statistically significant result should be translated into a decision threshold.

12. Assumptions, data quality and research governance

Statistical software can calculate a result even when the data are unsuitable. Before testing, researchers should verify sampling eligibility, duplicate records, impossible values, missingness, coding, weighting and dependence among observations. Clustered samples, such as customers nested within stores, violate ordinary independence and may require multilevel models or cluster-robust standard errors.

Multiple testing increases the chance of false positives. Researchers should pre-specify primary outcomes, distinguish confirmatory from exploratory analyses and use procedures such as false-discovery-rate or familywise-error control when appropriate. HARKing—presenting an unexpected finding as if it had been predicted—weakens scientific credibility.

Survey weights may be required when the sample does not represent the target population. Weighted estimates should be accompanied by an explanation of the weighting scheme and its effect on precision. Nonresponse bias cannot generally be corrected merely by increasing sample size.

Ethically, market researchers must protect confidentiality, minimise collection of personal data, obtain appropriate consent, and comply with applicable privacy and data-protection law. Sensitive segmentation variables should not be used to discriminate unfairly. Automated or AI-supported analysis requires human review, documented preprocessing, reproducibility checks and safeguards against fabricated outputs or opaque algorithmic decisions. Analysts should preserve audit trails, report limitations and avoid selective disclosure.

13. Best practices for professional reporting

A high-quality statistical report should include the research question, population and sampling method, operational definitions, sample size, missing-data treatment, descriptive statistics, chosen method, assumptions, estimate, confidence interval, p-value where relevant, effect size and limitation. Tables should show denominators and clearly label percentages. Visualisations should avoid distorted axes, unnecessary three-dimensional effects and unsupported causal language.

A concise managerial conclusion might state: “Consideration was higher post-campaign by 0.40 points on a seven-point scale, 95% CI [0.12, 0.68], with a small standardised effect (d = 0.24). The result supports an association with campaign timing but does not establish causality. The increase is below the organisation’s 0.50-point action threshold; further controlled testing is recommended.” This is more useful than “The campaign was highly successful, p < 0.05.”

Chapter Summary

Statistical analysis in market research begins with a well-defined question and an appropriate measurement and sampling plan. Descriptive statistics summarise distributions through frequencies, means, medians and measures of dispersion. Estimation uses sample information to quantify uncertainty, while confidence intervals communicate the precision of population estimates.

Hypothesis tests evaluate data against a null model, but p-values do not measure practical importance or prove causality. Cross-tabulation and chi-square analysis address categorical associations; correlation measures linear or monotonic association; regression estimates conditional relationships and supports prediction when assumptions and validation are adequate. ANOVA compares several means, with post-hoc tests locating differences. Nonparametric methods provide alternatives for ordinal outcomes, non-normal distributions and small samples, while retaining their own assumptions.

Effect sizes, confidence intervals, power and managerial thresholds should accompany significance tests. Outliers, missing data, dependence, multiple testing, measurement error and sampling bias must be addressed explicitly. Ethical analysis requires privacy protection, transparent reporting, reproducibility and avoidance of discriminatory or misleading conclusions. The strongest market-research decisions integrate statistical evidence with research design, brand strategy and commercial consequences.

Key Terms

Descriptive statistics: Methods that summarise observed data without generalising beyond them.

Inferential statistics: Methods that use sample data to draw conclusions about a population.

Population: The complete set of units to which a study intends to generalise.

Sample: The units actually observed in a study.

Parameter: A numerical characteristic of a population.

Statistic: A numerical characteristic calculated from a sample.

Standard deviation: A measure of dispersion around the mean.

Standard error: The estimated sampling variability of a statistic.

Confidence interval: An interval produced by a procedure designed to contain a population parameter at a stated long-run rate.

Null hypothesis: A formal statement representing no difference, association or effect under the tested model.

p-value: The probability, under the null model, of data at least as extreme as those observed.

Type I error: Rejecting a true null hypothesis.

Statistical power: The probability of detecting a specified effect when it exists under the model.

Cross-tabulation: A table showing the joint frequencies of categorical variables.

Cramér’s V: An effect-size measure for association in a contingency table.

Correlation: A measure of association, commonly linear association, between variables.

Regression: A model relating an outcome to one or more predictors.

ANOVA: Analysis of variance used to compare several group means.

Effect size: A measure of the magnitude of a difference or relationship.

Nonparametric test: A procedure relying less heavily on distributional assumptions, often using ranks.

Review Questions

Short Answer Questions

1. Distinguish a population parameter from a sample statistic.
2. Why might the median be preferred to the mean for highly skewed expenditure data?
3. What does a 95% confidence interval communicate under repeated sampling?
4. Why does a significant chi-square test not establish causation?
5. What is the difference between statistical significance and managerial significance?

Long Answer Questions

1. Explain the assumptions, interpretation and reporting requirements of an independent-samples t-test.
2. Compare Pearson correlation and linear regression in the context of brand research.
3. Explain how one-way ANOVA partitions variation and why post-hoc tests may be necessary.
4. Discuss the uses and limitations of nonparametric methods in analysing attitude data.
5. Develop a professional framework for reporting statistical findings to senior brand managers.

Analytical / Application Questions

1. A brand’s awareness is 48% in a sample of 400 consumers. Explain how you would estimate uncertainty, identify relevant assumptions and communicate the result to management.
2. A campaign produces a statistically significant 0.2-point increase in a seven-point brand-attitude scale, with a very large sample. Evaluate whether the campaign should be expanded.
3. Design an analysis plan to test whether campaign exposure predicts purchase intention while accounting for age, category involvement and prior brand usage. Identify assumptions, possible confounding and appropriate managerial interpretation.

Multiple Choice Questions with Answers

Question 1

Which statistic is generally most resistant to an extreme outlier?

A. Mean

B. Median

C. Variance

D. Standard deviation

Correct Answer: B. Median

Question 2

In a 2 × 3 cross-tabulation, the degrees of freedom for Pearson’s chi-square test are:

A. 1

B. 2

C. 3

D. 6

Correct Answer: B. 2

Question 3

A 95% confidence interval is best interpreted as:

A. A 95% probability that the fixed parameter lies in this particular interval

B. The range containing 95% of individual observations

C. An interval-producing procedure that captures the parameter in about 95% of repeated samples

D. Proof that the null hypothesis is false

Correct Answer: C. An interval-producing procedure that captures the parameter in about 95% of repeated samples

Question 4

Which method is most appropriate for comparing three independent groups when strong non-normality and ordinal measurement are present?

A. Kruskal–Wallis test

B. Paired t-test

C. Pearson correlation

D. McNemar test

Correct Answer: A. Kruskal–Wallis test

Question 5

If a regression slope is positive, it indicates that, holding other included predictors constant in a multiple regression, the outcome is expected to:

A. Decrease as the predictor increases

B. Increase as the predictor increases

C. Become normally distributed

D. Have zero residual variance

Correct Answer: B. Increase as the predictor increases

Question 6

A statistically significant omnibus ANOVA result establishes that:

A. Every group differs from every other group

B. At least one group mean differs from another

C. The largest group has the highest mean

D. The effect is necessarily managerially important

Correct Answer: B. At least one group mean differs from another

Question 7

For five observations 2, 3, 3, 4 and 8, the median is:

A. 2

B. 3

C. 4

D. 8

Correct Answer: B. 3

Question 8

Which statement about correlation is correct?

A. Correlation alone proves causation

B. A zero Pearson correlation rules out every possible relationship

C. Correlation measures association and may be distorted by outliers

D. Correlation can only be used with nominal variables

Correct Answer: C. Correlation measures association and may be distorted by outliers

Question 9

If sample size increases while variability and confidence level remain fixed, the standard error of a sample mean generally:

A. Increases proportionally

B. Decreases because it is divided by √n

C. Becomes equal to the population mean

D. Is unaffected

Correct Answer: B. Decreases because it is divided by √n

Question 10

Which is an appropriate interpretation of a p-value?

A. The probability that the research hypothesis is true

B. The probability that the sample was randomly selected

C. The probability, under the null model, of results at least as extreme as those observed

D. The magnitude of the managerial effect

Correct Answer: C. The probability, under the null model, of results at least as extreme as those observed