Market Research Analytics and Insight Communication

Introduction

Market research creates value only when collected evidence is converted into a sound business decision. A questionnaire may generate thousands of responses, but unclean data, inappropriate analysis, or unclear communication can lead an entrepreneur to launch the wrong product, target the wrong customers, or misunderstand demand. Market research analytics is therefore the disciplined process of preparing, summarising, analysing and interpreting market data. Insight communication is the equally important process of presenting findings so that decision-makers understand what the evidence means and what should be done next.

This chapter focuses on practical analytics suitable for startup and small-business research. It covers data cleaning and coding, descriptive statistics, cross-tabulation, correlation, basic hypothesis interpretation, visualisation, the distinction between information and insight, and the preparation of actionable reports and presentations.

Learning Objectives

By the end of this chapter, students should be able to:

1. Explain the stages involved in preparing market research data.
2. Code categorical and numerical variables appropriately.
3. Calculate and interpret frequencies, percentages, means, medians and measures of variation.
4. Construct and interpret cross-tabulations.
5. Explain correlation and its limitations.
6. Interpret a basic hypothesis test without overstating the evidence.
7. Select suitable charts for different types of data.
8. Distinguish information, findings, insights and recommendations.
9. Produce a concise, ethical and actionable research report or presentation.

1. From Raw Responses to Evidence

A research dataset normally contains rows representing cases, such as respondents or transactions, and columns representing variables, such as age, location, purchase frequency or satisfaction. Raw data may contain missing answers, spelling differences, duplicate records, invalid values and inconsistent formats. Analysis should not begin until the dataset has been inspected and documented.

A practical workflow is:

1. Define the purpose of the analysis.
2. Create a data dictionary.
3. Check completeness and validity.
4. Remove or correct duplicate records.
5. Standardise formats and categories.
6. Code variables.
7. Analyse and visualise the data.
8. Interpret results in relation to the research question.
9. Communicate limitations and recommended action.

The data dictionary

A data dictionary records what each variable means and how it is measured. It should include the variable name, question wording, type, permitted values, coding rules and missing-value treatment.

A data dictionary prevents analysts from forgetting what numerical codes represent. Codes are labels, not automatically quantities. For example, satisfaction codes from 1 to 5 have an order, but the distance between categories may not be exactly equal.

Data cleaning

Common checks include:

• Range checks: age cannot be negative; a five-point satisfaction item should normally contain values from 1 to 5.
• Consistency checks: a respondent who states that they have never purchased a product should not also report a purchase amount for their last purchase unless the questionnaire permits this.
• Duplicate checks: identical response IDs or repeated submissions should be investigated.
• Missing-data checks: blank, “not applicable,” “prefer not to say” and “unknown” should not be treated as the same condition.
• Standardisation: “Nairobi,” “nairobi” and “NBI” may need to be mapped to one category.

Cleaning decisions must be documented. Deleting inconvenient observations merely to obtain a preferred result is unethical and damages validity. When missing data are substantial, the report should state how many cases were affected and whether the remaining sample may be biased.

2. Measurement and Coding

The measurement scale affects the analysis that is appropriate.

• Nominal: categories without order, such as payment method.
• Ordinal: ordered categories, such as low, medium and high satisfaction.
• Interval: numerical differences are meaningful, but zero is not a true absence; many attitude scales are treated approximately this way.
• Ratio: numerical differences and ratios are meaningful, with a true zero; examples include revenue, units sold and customer age.

Coding converts responses into a form suitable for analysis. A nominal variable such as preferred channel may be coded 1, 2 and 3, but calculating its average is meaningless. A binary variable may be coded 1 = selected and 0 = not selected. Multiple-response questions require special care: each option can become a separate binary variable.

Reverse-coded questionnaire items must be recoded before creating a scale. For a five-point item, the reverse score can be calculated as:

Here, x is the original response, x′ is the reversed response and k is the maximum scale value. For a five-point scale, a response of 1 becomes 5 and a response of 4 becomes 2. This transformation assumes that the scale categories are consistently numbered.

3. Descriptive Statistics

Descriptive statistics summarise what was observed. They do not, by themselves, prove why something occurred or guarantee that the sample represents the whole market.

Frequencies and percentages

A frequency is the number of cases in a category. A percentage expresses that frequency relative to the valid total:

If 84 of 240 valid respondents prefer mobile purchasing, the percentage is:

The denominator should be stated. If some respondents did not answer the question, percentages based on all recruited respondents may differ from percentages based on valid responses.

Measures of central tendency

The mean is the arithmetic average:

where xᵢ is an observed value and n is the number of observations. The mean is useful for numerical data but can be strongly affected by extreme values.

The median is the middle value after observations are arranged in order. It is often preferable for income, spending and delivery time when the distribution is skewed. The mode is the most frequently occurring value or category and is particularly useful for nominal data.

For example, consider monthly customer spending of 10, 12, 12, 15 and 51 currency units. The mean is 20, while the median is 12 and the mode is 12. The mean is pulled upward by the unusually large value of 51. A responsible report might therefore present both mean and median.

Measures of dispersion

Dispersion describes how spread out observations are.

• Range: maximum minus minimum.
• Variance: average squared deviation from the mean.
• Standard deviation: the square root of variance.

For a sample, standard deviation is calculated as:

The symbol s denotes the sample standard deviation. The denominator n − 1 is used when estimating population variability from a sample. Standard deviation is expressed in the same units as the original variable. It is most informative when the mean is an appropriate summary and the distribution is not severely distorted by outliers.

A useful interpretation is comparative rather than absolute: two customer segments may have the same average spending but different standard deviations, indicating different levels of spending consistency.

4. Cross-Tabulation and Segment Analysis

A cross-tabulation, or crosstab, compares the distribution of one categorical variable across the categories of another. It is valuable for segmentation because overall averages can hide important differences.

Suppose a startup asks 200 respondents whether they would try a new delivery service:

The overall trial intention is 51%. Row percentages provide more useful segment information:

• Under 25: 48 ÷ 80 × 100 = 60%
• Age 25–44: 42 ÷ 80 × 100 = 52.5%
• Age 45 and above: 12 ÷ 40 × 100 = 30%

The apparent opportunity is strongest among respondents under 25. However, this is an association, not proof that age causes trial intention. Segment sizes, sampling method and other variables must also be considered.

Column percentages answer a different question. For example, among all respondents who said “yes,” what percentage belonged to each age group? Analysts must label percentages clearly because row and column percentages can lead to different interpretations.

For larger samples, a chi-square test may be used to assess whether two categorical variables are statistically associated. The test compares observed counts with expected counts:

Here, Oᵢ is an observed frequency and Eᵢ is the frequency expected if the variables were independent. The test assumes independent observations and sufficiently large expected counts. A statistically significant result suggests that the variables are unlikely to be independent in the population represented by the sample; it does not establish causation or practical importance.

5. Correlation and Basic Hypothesis Interpretation

Correlation measures the direction and strength of a linear relationship between two numerical variables. Pearson’s sample correlation coefficient is:

The coefficient r ranges from −1 to +1. A positive value indicates that larger values of one variable tend to accompany larger values of the other; a negative value indicates an inverse pattern. A value near zero indicates little linear association, although a nonlinear relationship may still exist.

Correlation does not demonstrate causation. Advertising exposure and sales may be positively correlated because advertising increases sales, because successful stores advertise more, or because both are influenced by location, season or business size. Outliers can also produce or conceal correlation. A scatterplot should accompany a correlation coefficient whenever possible.

Hypotheses and p-values

A hypothesis test begins with a null hypothesis, usually stating that there is no specified difference or association, and an alternative hypothesis, stating what evidence would support a difference or association. For example:

• Null hypothesis: the average satisfaction score is the same for customers using the old and new service processes.
• Alternative hypothesis: the average satisfaction scores differ.

A p-value is the probability of obtaining results at least as extreme as those observed, assuming that the null hypothesis is true and the test assumptions hold. If the p-value is below a preselected significance level, often 0.05, researchers may reject the null hypothesis. This does not mean that the probability the null hypothesis is true is 5%, nor does it prove the alternative explanation.

Interpretation should consider:

1. Effect size: Is the difference large enough to matter commercially?
2. Confidence interval: What range of values is compatible with the data?
3. Sample size: Large samples can make trivial differences statistically significant.
4. Design: A cross-sectional survey cannot normally establish temporal causality.
5. Assumptions and data quality: Invalid coding or biased sampling can undermine the test.

Statistical significance and business significance are different. A two-point increase on a 100-point satisfaction scale may be statistically significant in a very large sample but irrelevant to a startup’s decision.

6. Visualising Market Research Findings

Visualisation should make a comparison, pattern or relationship easier to see. The chart type must match the measurement task.

Pie charts can be used sparingly when categories are few and differences are substantial; bar charts usually support more accurate comparison. Three-dimensional effects, excessive colours and truncated axes may mislead viewers.

A professional chart should have a descriptive title, labelled axes, units, a stated denominator where relevant, readable categories and a source note. The visual should emphasise the decision-relevant comparison rather than decorate the page. Charts should not imply precision greater than the data support.

7. From Information to Insight

These terms describe different levels of analytical value:

• Data: unprocessed observations, such as individual survey responses.
• Information: organised data, such as “51% selected the new service.”
• Finding: a verified pattern, such as “trial intention was higher among younger respondents.”
• Insight: an interpretation that explains relevance, such as “younger customers may be the most efficient early-adopter segment because they show high trial intention and frequent mobile purchasing.”
• Recommendation: a proposed action, such as “test a mobile-first launch offer with customers under 25 before expanding broadly.”

An insight must connect evidence to context and decision. It should identify who is affected, what appears to be happening, why it matters, and what could be tested. Analysts should distinguish observed facts from interpretations and assumptions.

A useful report structure is:

1. Executive decision: the action or decision requiring evidence.
2. Research question and method: population, sample, dates and data-collection approach.
3. Key findings: a small number of clearly labelled results.
4. Interpretation: what the patterns may mean.
5. Recommendations: specific actions, owners, timing and measures.
6. Limitations: sampling, measurement, missing data and uncertainty.
7. Technical appendix: questionnaire, coding, tables and analytical details.

Recommendations should be SMART where practical: specific, measurable, achievable, relevant and time-bound. For example, “Run a four-week pilot of mobile delivery in two university districts, targeting 500 eligible users, and compare conversion and repeat purchase with a control area” is more actionable than “improve mobile marketing.”

8. Presenting Findings to Decision-Makers

A presentation should be designed around the decision, not around every question in the questionnaire. A concise structure is:

1. Business problem and decision required.
2. Method and sample in one slide.
3. Three to five key findings.
4. Implications for customers, product and marketing.
5. Recommended action and success measures.
6. Risks, limitations and next research step.

The presenter should explain the denominator, avoid jargon, and state uncertainty honestly. A headline should communicate meaning rather than merely name a chart. “Younger respondents show the strongest trial intention” is more useful than “Trial intention by age.”

An effective recommendation links evidence to an experiment. Startups can use landing-page tests, limited geographic pilots, price tests or prototype trials. Research findings guide the test; they do not eliminate the need to learn from actual behaviour.

9. Applied Case: FreshRoute Delivery (Illustrative)

Background

FreshRoute, an illustrative startup, planned a low-cost delivery service for locally prepared meals. Management needed to decide which customers to target first and whether convenience or price should lead the launch message.

Problem and approach

The team surveyed 300 urban adults recruited through community groups and social media. The questionnaire measured age group, current ordering frequency, preferred ordering channel, acceptable delivery fee and likelihood of trying FreshRoute. The team documented the convenience sample as a limitation and removed six duplicate records, leaving 294 valid cases.

Analysis

The overall trial intention was 52%. A crosstab showed trial intention of 64% among respondents ordering meals at least four times per month, compared with 38% among those ordering once or never. Among frequent orderers, 71% preferred mobile ordering. Median acceptable delivery fee was 3 currency units, while the mean was 4.20 because a small group selected very high fees.

A scatterplot suggested a modest positive relationship between ordering frequency and willingness to pay, with correlation r = 0.31. This was interpreted as a weak-to-moderate linear association, not evidence that frequency causes willingness to pay. A comparison of two message concepts found a higher average purchase intention for the convenience message, but the difference was not statistically significant at the selected 5% level. The sample was not random, so population generalisation remained limited.

Insight and recommendation

The information was that frequent orderers had higher trial intention and tended to prefer mobile ordering. The insight was that FreshRoute’s most promising early-adopter segment may be digitally comfortable, frequent meal purchasers who value reduced effort more than the lowest possible fee. The team recommended a four-week mobile-first pilot aimed at frequent orderers, with a transparent delivery fee and convenience-focused message. It would track landing-page conversion, first orders, repeat orders and contribution margin.

Outcome and learning points

The pilot would not be judged solely by survey intention. Actual behaviour would provide stronger evidence about demand. The case demonstrates the importance of cleaning records, comparing segments, using median alongside mean, avoiding causal claims, stating sampling limitations and converting findings into a measurable test.

10. Ethics, Governance and Technology

Market research analytics must protect participants and preserve trust. Researchers should collect only necessary personal data, explain the purpose of collection, secure identifiable information and restrict access. Names and contact details should be separated from analytical data where possible. Reports should use aggregated results and avoid re-identification of small groups.

Consent, voluntary participation and the right to withdraw are important principles. Analysts must not fabricate responses, suppress inconvenient findings, manipulate charts or present exploratory patterns as confirmed conclusions. Where applicable, organisations must follow data-protection law, institutional research rules and contractual requirements.

Spreadsheets are adequate for small datasets when formulas, versions and validation rules are controlled. Statistical software and dashboards improve reproducibility when analysts preserve the original data, syntax or transformation record, and a final analytical dataset. Artificial intelligence can assist with categorising open-ended responses, detecting unusual records and drafting chart descriptions, but outputs require human review. AI may reproduce bias, expose confidential data or invent interpretations. It should not replace methodological judgement or ethical accountability.

11. Common Errors and Best Practices

Frequent errors include treating codes as measurements, reporting percentages without denominators, ignoring missing data, relying on means for highly skewed variables, inferring causation from correlation, testing many relationships without caution, and burying the recommendation under technical detail.

Best practice is to:

• begin with the decision and research question;
• maintain a data dictionary and cleaning log;
• preserve raw data and create a separate analysis copy;
• report both counts and percentages for important categories;
• inspect distributions and outliers;
• use segment comparisons thoughtfully;
• combine statistical evidence with practical effect and context;
• label assumptions and limitations;
• use simple, honest charts;
• link every recommendation to evidence and a measurable next step.

Chapter Summary

Market research analytics transforms raw responses into evidence for entrepreneurial decisions. Reliable analysis begins with a data dictionary, systematic cleaning, transparent coding and careful treatment of missing data. Frequencies and percentages describe categorical responses; means, medians, modes and standard deviations summarise numerical data. Cross-tabulation reveals segment differences, while correlation describes linear association without proving causation. Hypothesis interpretation requires attention to p-values, effect size, confidence, sample design and assumptions.

Visualisations should match the analytical purpose and avoid misleading design. Information becomes insight when a verified pattern is interpreted in business context. A strong report separates data, findings, insights and recommendations, communicates uncertainty, and proposes an action that can be tested. Ethical governance requires privacy, informed participation, accurate reporting and responsible use of analytical technology.

Key Terms

Data cleaning: Detecting and correcting or documenting errors, inconsistencies and omissions in a dataset.

Data dictionary: A record describing variables, meanings, types, codes and permitted values.

Coding: Assigning labels or numerical representations to responses.

Nominal scale: A measurement scale containing categories without inherent order.

Ordinal scale: A scale whose categories have a meaningful order.

Mean: The arithmetic average of numerical observations.

Median: The middle ordered observation.

Mode: The most frequently occurring value or category.

Standard deviation: A measure of the dispersion of observations around their mean.

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

Correlation: A numerical measure of the direction and strength of linear association.

Outlier: An observation unusually distant from the rest of a dataset.

Null hypothesis: A statement representing no specified difference or association.

P-value: The probability of results at least as extreme as those observed if the null hypothesis is true.

Statistical significance: Evidence judged unlikely under a stated null hypothesis and significance level.

Effect size: The magnitude of a difference or relationship.

Information: Organised data that describes a result or pattern.

Insight: A context-based interpretation that explains why a finding matters.

Recommendation: A proposed action based on evidence and business objectives.

Data governance: Policies and practices controlling data quality, security, access and ethical use.

Review Questions

Short Answer Questions

1. What is the purpose of a data dictionary?
2. Distinguish nominal, ordinal and ratio measurement scales.
3. Why might the median be more useful than the mean for customer spending?
4. What does a correlation coefficient measure?
5. State two principles of ethical market research analytics.

Long Answer Questions

1. Explain a complete workflow for preparing survey data for analysis, including the treatment of duplicates and missing values.
2. Discuss how cross-tabulation can support market segmentation, and explain why row and column percentages must be labelled.
3. Explain the difference between statistical significance and business significance.
4. Describe how an entrepreneur can move from data to information, finding, insight and recommendation.
5. Design a structure for a professional market research report and justify the role of its major sections.

Analytical / Application Questions

1. A survey records monthly spending of 8, 9, 10, 11 and 42 currency units. Calculate the mean and median, identify the likely influence of the extreme value, and recommend which measure should headline a customer profile.
2. A startup finds that 70% of frequent buyers prefer its mobile app, compared with 35% of infrequent buyers. Explain what this pattern suggests, identify two alternative explanations, and propose one follow-up test.
3. A statistically significant relationship is found between advertising exposure and purchase intention, with r = 0.22. Prepare a short managerial interpretation that states what may be concluded, what cannot be concluded, and what additional evidence should be collected.

Multiple Choice Questions with Answers

Question 1

Which activity should normally occur before calculating descriptive statistics?

A. Designing a promotional campaign

B. Cleaning and validating the dataset

C. Deleting all unusual responses

D. Converting every variable into a percentage

Correct Answer: B. Cleaning and validating the dataset

Question 2

Which measure is generally least affected by an extreme outlier?

A. Mean

B. Median

C. Range

D. Variance

Correct Answer: B. Median

Question 3

A nominal variable coded 1, 2 and 3 should not normally be analysed by calculating its mean because the codes represent:

A. Equal numerical quantities

B. Ordered and equally spaced values

C. Category labels

D. Standard deviations

Correct Answer: C. Category labels

Question 4

What is the main purpose of a cross-tabulation?

A. To compare distributions across categories

B. To prove that one variable causes another

C. To replace data cleaning

D. To calculate only numerical averages

Correct Answer: A. To compare distributions across categories

Question 5

If 60 of 150 valid respondents select an option, the percentage is:

A. 25%

B. 30%

C. 40%

D. 60%

Correct Answer: C. 40%

Question 6

A correlation of r = −0.80 indicates:

A. A strong negative linear association

B. A weak positive linear association

C. Proof that one variable causes the other

D. No possible relationship of any kind

Correct Answer: A. A strong negative linear association

Question 7

A p-value below 0.05 most appropriately means that:

A. The null hypothesis has a 5% probability of being true

B. The observed result would be relatively unlikely if the null hypothesis were true

C. The result is automatically important commercially

D. The sample must be unbiased

Correct Answer: B. The observed result would be relatively unlikely if the null hypothesis were true

Question 8

Which chart is most suitable for examining the relationship between two numerical variables?

A. Pie chart

B. Scatterplot

C. Bar chart of one category

D. Organisational chart

Correct Answer: B. Scatterplot

Question 9

Which statement is an insight rather than merely information?

A. 52% of respondents selected mobile ordering

B. There were 294 valid responses

C. Frequent mobile users may be the strongest early-adopter segment for a mobile-first pilot

D. The questionnaire contained ten questions

Correct Answer: C. Frequent mobile users may be the strongest early-adopter segment for a mobile-first pilot

Question 10

Which recommendation is most actionable?

A. Improve marketing

B. Study customers more

C. Launch everywhere immediately

D. Run a four-week mobile pilot in two target districts and compare conversion and repeat purchase

Correct Answer: D. Run a four-week mobile pilot in two target districts and compare conversion and repeat purchase