Descriptive Analysis and Data Visualization

Introduction

Descriptive analysis is the first substantive stage of business data analysis. After data have been collected, researchers must transform unstructured observations into an accurate, interpretable account of what the data show. Managers use descriptive analysis to understand customers, employees, operations, markets and financial performance before making decisions or applying more advanced statistical models.

Descriptive analysis does not establish causation. It summarizes observed data through distributions, averages, measures of variability, tables, charts and dashboards. Inferential statistics go further by using sample data to draw conclusions about a wider population. Effective business research requires a clear distinction between these purposes. A high average customer-satisfaction score, for example, describes respondents in the sample; it does not automatically prove that a new service caused satisfaction to increase.

This chapter explains the complete analytical workflow from data preparation to interpretation, visualization, hypothesis testing and software implementation. It also emphasizes ethical communication: a technically correct analysis can still mislead when the sample, scale, denominator or visual design is obscured.

Learning Objectives

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

1. Explain coding, editing and data entry as stages of data preparation.
2. Distinguish descriptive from inferential statistics.
3. Construct and interpret distributions, central-tendency measures and dispersion measures.
4. Use cross-tabulations, charts and dashboards to identify and communicate patterns.
5. Explain the concepts, types and procedures of hypothesis testing.
6. Use SPSS or R appropriately for business data analysis.
7. Evaluate analytical claims for accuracy, transparency and misleading representation.

1. From Raw Data to an Analytical Dataset

1.1 Data preparation

Data preparation is the systematic process of converting collected information into a form suitable for analysis. It includes checking completeness, defining variables, coding responses, editing errors, entering data, documenting decisions and protecting confidentiality. Poor preparation produces what is sometimes called “garbage in, garbage out”: sophisticated statistical software cannot correct an invalid measurement or systematically biased sample.

A research dataset normally contains rows representing cases—such as customers, firms or transactions—and columns representing variables. Each variable should have a clear name, definition, measurement level, permissible values and missing-data code. A data dictionary records these decisions. For example, loyalty_score might be defined as a respondent’s rating from 1 (very unlikely to repurchase) to 7 (very likely to repurchase).

1.2 Coding

Coding assigns symbols or numbers to responses so that they can be stored and analyzed. In a closed question, “Full-time,” “Part-time” and “Not employed” might be coded 1, 2 and 3. The numbers are labels, not necessarily quantities. Calculating an average of employment-status codes would have no meaningful interpretation.

Binary variables often use 0 and 1, such as 1 = renewed a subscription and 0 = did not renew. For multi-response questions, separate indicator variables may be created: uses_mobile_app, uses_branch, and uses_web, each coded 1 for selected and 0 for not selected. Open-ended responses require thematic coding. Researchers should develop categories, define inclusion rules, train coders where possible and assess disagreements.

Coding must distinguish valid values from missing values. A response of 99 may mean “not answered,” not ninety-nine. Missing values can be coded as system missing or as documented user-missing values. Treating missing codes as real observations can seriously distort means, percentages and relationships.

1.3 Editing

Editing detects and resolves errors while preserving the original evidence. Range checks identify impossible values, such as an age of 240 or a five-point rating coded 8. Consistency checks identify contradictions, such as a respondent marked “no employees” but reporting a payroll cost. Duplicate records, implausibly rapid completion times and patterned responses may indicate data-quality problems.

Editing decisions should be documented. A researcher should not silently change an observation because it appears inconvenient. Appropriate actions include correcting an obvious transcription error from the original questionnaire, marking an uncertain value as missing, contacting the respondent where feasible, or retaining the value and reporting a sensitivity analysis. The raw dataset should be preserved as read-only, while a separate cleaned dataset records transformations.

1.4 Data entry and quality control

Data entry may be manual, imported from an online survey, extracted from enterprise systems or generated through an application programming interface. Controls include validation rules, required fields, range limits, double entry for critical studies and random comparison with source documents. A reproducible workflow records who changed what, when and why.

Before analysis, inspect frequencies, minimum and maximum values, missingness and unusual cases. The unit of analysis must also be verified. Mixing transaction-level records with customer-level records can cause some customers to receive disproportionate weight. Researchers must decide whether weights, aggregation or deduplication are required.

2. Descriptive and Inferential Statistics

Descriptive statistics summarize the observed sample. They include frequency distributions, percentages, means, medians, modes, ranges, variances, standard deviations, percentiles, cross-tabulations and graphical displays. Their purpose is organization and communication.

Inferential statistics use sample evidence to estimate population characteristics or evaluate claims about a population. They include confidence intervals, hypothesis tests, correlation tests, regression, analysis of variance and other models. Inference depends on sampling design and assumptions. A p-value from a convenience sample does not eliminate selection bias, and statistical significance does not establish managerial importance.

The distinction can be illustrated by a survey of 400 customers. The statement “62% of respondents would recommend the service” is descriptive. The statement “the population proportion is likely between 57% and 67%” is inferential because it generalizes beyond the observed respondents and requires assumptions about sampling and estimation.

3. Distributions and Measures of Location

A distribution shows how observations are allocated across values or categories. A frequency distribution lists each value or interval and its count. Relative frequency is calculated as:

where p_i is the proportion in category or value i, f_i is its frequency and n is the total number of valid observations. Percentage equals 100 × p_i. Categories should be mutually exclusive and collectively exhaustive where appropriate.

The mean is the arithmetic average:

Here x_i is observation i and n is the number of valid observations. The mean is appropriate for quantitative data when values are meaningfully additive, but it is sensitive to extreme observations.

The median is the middle ordered value. If n is odd, it is the observation at position (n+1)/2. If n is even, it is the average of the observations at positions n/2 and n/2 + 1. The median is often preferable for income, transaction value and waiting time when distributions are skewed.

The mode is the most frequently occurring value or category. It is the only measure of central tendency directly applicable to nominal categories. A distribution may have more than one mode or no unique mode.

For example, five delivery times of 2, 3, 3, 4 and 13 days have mean 5 days, median 3 days and mode 3 days. The mean is pulled upward by the unusually long delivery. Reporting only the mean would hide the typical customer experience.

Percentiles divide ordered data into location points. The 50th percentile is the median; the 25th and 75th percentiles are commonly called the first and third quartiles. Percentile definitions vary slightly across software, so the method should be documented when precision matters.

4. Dispersion, Shape and Outliers

Measures of dispersion show how widely observations vary. The range is maximum minus minimum, but it depends heavily on extreme values. The interquartile range is:

where Q₁ and Q₃ are the first and third quartiles. It summarizes the middle 50% and is resistant to outliers.

The sample variance is:

The sample standard deviation is s = √s². The denominator n − 1 is used when sample data estimate population variability; it provides the usual unbiased estimate of population variance under random sampling. Standard deviation is expressed in the same units as the original variable, whereas variance is expressed in squared units.

A coefficient of variation can compare relative dispersion when the variable is ratio-scaled and its mean is positive:

It should not be used when the mean is zero, near zero or conceptually unsuitable. Shape also matters. Positive skewness means a long right tail, as commonly occurs with income or claim size. Negative skewness has a longer left tail. Outliers may be genuine, errors, unusual but important cases or evidence of multiple segments. They should be investigated rather than automatically deleted.

A useful rule is to report both a resistant summary, such as median and IQR, and a conventional summary, such as mean and standard deviation, when the distribution is substantially skewed. A histogram, box plot or percentile table should support the interpretation.

5. Cross-Tabulations and Relationships Between Variables

A cross-tabulation, or contingency table, summarizes the joint distribution of two categorical variables. It contains cell frequencies and may display row percentages, column percentages or total percentages. The denominator must be stated. For example, a row percentage answers “within each customer segment, what proportion renewed?” whereas a column percentage answers “among renewers, what proportion belonged to each segment?” These are different questions.

Suppose renewal results are:

The renewal rate is 60% for Standard customers and 80% for Premium customers. The difference identifies an association in this sample, not proof that premium status causes renewal. Segment size, tenure, income and service quality may confound the relationship.

For two quantitative variables, scatterplots reveal direction, form, strength and unusual cases. Correlation summarizes linear association but does not prove causality and may be distorted by outliers or restricted range. Cross-tabulations and plots should precede formal tests because they reveal coding mistakes and violations of assumptions.

6. Visualizing Business Data

A chart should answer a defined question. Bar charts compare categories; histograms display the distribution of quantitative values; line charts show change over ordered time; scatterplots show relationships between quantitative variables; box plots compare distributions across groups; and maps display spatial patterns when location is analytically relevant.

Good practice includes a meaningful title, clear units, labelled axes, an informative legend, readable scales and a stated denominator where percentages are used. Time should normally progress left to right. Three-dimensional effects, unnecessary decoration, excessive colours and truncated axes can exaggerate differences. A bar chart whose vertical axis begins at 90% can make a two-percentage-point change appear enormous. Truncating a scale is sometimes defensible for a line chart, but it must be clearly marked and should not replace the underlying values.

Dashboards combine key performance indicators, charts, filters and explanatory context. A managerial dashboard should have a defined audience, a small number of decision-relevant measures, consistent definitions, visible data dates, targets and exception alerts. It should distinguish actual performance from forecast, avoid mixing incompatible denominators and show data quality warnings. A dashboard is not a data dump: every visual should support a decision or question.

Accessibility requires adequate contrast, text alternatives, direct labels and colour choices that do not rely only on red and green. Interactive dashboards should preserve an audit trail so that users can identify the filters and calculations behind a displayed result.

7. Hypothesis Testing: Concepts, Types and Procedures

A statistical hypothesis is a claim about a population parameter or relationship. The null hypothesis, H₀, usually states no difference, no association or a specified value. The alternative hypothesis, H₁ or H_a, represents the research claim.

Tests may be two-tailed or one-tailed. A two-tailed alternative detects departures in either direction, such as μ ≠ μ₀. A one-tailed alternative specifies a direction, such as μ > μ₀ or μ < μ₀. Directional tests should be chosen before seeing the results; changing direction after inspection inflates false-positive risk.

The significance level, α, is the chosen maximum probability of rejecting a true null hypothesis under the testing framework. A Type I error is rejecting a true H₀; a Type II error is failing to reject a false H₀. Statistical power is the probability of detecting a specified effect when it exists. Power generally increases with larger samples, larger effects, lower variability and a less stringent significance threshold, although lowering α reduces protection against Type I error.

A p-value is the probability, assuming H₀ and the model assumptions are true, of obtaining a result at least as extreme as the observed result in the relevant direction. It is not the probability that H₀ is true, nor the probability that the result occurred by chance in an unrestricted sense. A small p-value indicates incompatibility with the null model, not practical importance.

A confidence interval gives a range of plausible parameter values under a specified procedure. It communicates both direction and precision and is often more informative than a significance decision. Statistical significance should be considered alongside effect size, confidence interval, practical consequences, data quality and research design.

A general testing procedure is:

1. Define the population, parameter and research question.
2. State H₀ and H_a before examining the outcome.
3. Select α and an appropriate test.
4. Check measurement, sampling and model assumptions.
5. Calculate the test statistic and p-value, or obtain software output.
6. Decide whether to reject or fail to reject H₀.
7. Report the estimate, effect size, confidence interval, p-value, sample size and limitations.

Common tests include a one-sample t-test for a mean against a reference value, an independent-samples t-test for two group means, a paired t-test for matched measurements, chi-square tests for categorical association, correlation tests and analysis of variance for more than two group means. The selection depends on the research design and measurement level. If assumptions are seriously violated, researchers may transform variables, use robust or non-parametric methods, or model the data directly. Multiple testing increases the chance of false positives and should be addressed through pre-specification or appropriate adjustment.

8. Statistical Software: SPSS and R

Software improves speed, reproducibility and the ability to examine data, but it does not replace methodological judgment. In SPSS, the Variable View defines names, labels, value labels, missing values and measurement levels. Frequencies, descriptives, explore, crosstabs, charts and compare-means procedures support the workflow. Output should be inspected rather than copied mechanically.

A typical SPSS syntax sequence might be represented as:

FREQUENCIES VARIABLES=gender region.
DESCRIPTIVES VARIABLES=spend satisfaction /STATISTICS=MEAN STDDEV MIN MAX.
CROSSTABS TABLES=segment BY renewed /CELLS=COUNT ROW.

R supports transparent, script-based analysis. A simple workflow is:

data <- read.csv("survey.csv")
summary(data$satisfaction)
mean(data$satisfaction, na.rm = TRUE)
table(data$segment, data$renewed)
prop.table(table(data$segment, data$renewed), margin = 1)

The argument na.rm = TRUE excludes missing values from the mean; it should be used knowingly because the amount and pattern of missingness must still be reported. In professional work, R scripts, package versions, data dictionaries and output files should be archived. SPSS syntax provides similar reproducibility. Analysts should check whether defaults use sample or population formulas, how missing values are handled and how confidence intervals and percentile methods are calculated.

AI-assisted analytics can help generate code, suggest visualizations or identify anomalies, but generated code must be tested against known calculations. Confidential business data should not be uploaded to an unapproved service. Human review remains essential for variable meaning, privacy, fairness and interpretation.

9. Illustrative Case: Retailer Renewal Dashboard

Background and problem

A fictional subscription retailer wants to understand a decline in annual renewal. Management requests a dashboard and a recommendation about whether the premium plan improves retention.

Analysis

The research team first freezes the raw customer file, removes duplicate customer identifiers and documents that the unit of analysis is the customer, not the invoice. It codes renewal as 1/0, checks that plan categories are valid and marks missing renewal outcomes separately. The team finds 325 valid customers: 150 Standard and 175 Premium.

The cross-tabulation shows renewal rates of 60% for Standard and 80% for Premium. Median customer tenure is 18 months, while a small number of very long-tenure accounts make the mean 31 months. A box plot identifies these accounts for investigation rather than deletion. A line chart shows that renewal fell in the most recent quarter, but the chart includes only completed renewals and clearly labels the observation period.

Approach and interpretation

A two-proportion comparison could assess whether the observed difference is unlikely under equal population renewal rates. However, the groups were not randomly assigned: premium customers may have higher spending, longer tenure or better service access. Therefore, even a statistically significant result would not establish that premium status caused renewal. The dashboard reports the descriptive rates, group sizes, confidence intervals and a warning about possible confounding. It also includes a filter for tenure and acquisition channel.

Outcome and learning points

Management launches a controlled offer in which eligible customers are randomly assigned to a premium trial or control condition. The original descriptive analysis becomes a basis for a stronger causal study rather than an overconfident conclusion. The case demonstrates that preparation, denominators, visualization, inference and research design are interconnected. A credible dashboard communicates uncertainty and limitations instead of presenting one attractive percentage as a complete explanation.

10. Ethical and Governance Considerations

Researchers must protect personal data through data minimization, access controls, de-identification where appropriate, retention limits and compliance with applicable privacy law and organizational policy. Small cells in dashboards may permit re-identification even when names are removed. Sensitive attributes require additional safeguards and careful consideration of fairness.

Selective reporting is unethical. Analysts should not hide inconvenient cases, change hypotheses after viewing results without disclosure, repeatedly test alternatives until one is significant, or use a misleading denominator. Reproducibility requires retaining raw data where lawful, transformation records, code, analytical decisions and versions of software. Managers should be told what the data cannot establish, including uncertainty, non-response, measurement error and limits to generalization.

Chapter Summary

Descriptive analysis converts prepared data into understandable evidence. Coding, editing and controlled data entry establish the quality of the analytical dataset. Distributions, means, medians, modes, percentiles and measures of dispersion summarize location, variability and shape. Cross-tabulations and charts reveal patterns, but percentages require explicit denominators and relationships do not automatically imply causation.

Inferential statistics generalize from samples through estimation and hypothesis testing. Hypothesis tests require clearly stated null and alternative hypotheses, an appropriate test, assumptions, an error framework and interpretation based on effect size and uncertainty as well as p-values. SPSS and R support efficient, reproducible analysis, but software cannot replace design knowledge or ethical judgment. Dashboards should be decision-oriented, accessible, transparent and resistant to misleading representation.

Key Terms

Coding: Assigning labels or values to responses for storage and analysis.

Data dictionary: Documentation of variable definitions, codes, labels and missing values.

Data editing: Checking and resolving errors, inconsistencies and invalid observations.

Descriptive statistics: Statistics that summarize observed data.

Inferential statistics: Methods for drawing conclusions about a population from sample data.

Frequency distribution: A display of counts or proportions across values or categories.

Mean: The arithmetic average of quantitative observations.

Median: The middle ordered observation.

Mode: The most frequently occurring value or category.

Variance: Average squared deviation, using an appropriate population or sample denominator.

Standard deviation: Square root of variance, expressed in the original units.

Interquartile range: The difference between the third and first quartiles.

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

Dashboard: An integrated display of decision-relevant indicators and visualizations.

Null hypothesis: A formal statement evaluated by a statistical test, usually representing no effect or a specified value.

Alternative hypothesis: The claim considered against the null hypothesis.

Type I error: Rejecting a true null hypothesis.

Type II error: Failing to reject a false null hypothesis.

P-value: Probability of results at least as extreme as observed under the null model and assumptions.

Statistical power: Probability of detecting a specified effect when it exists.

Review Questions

Short Answer Questions

1. Why should missing-value codes be distinguished from valid numerical values?
2. When is the median more informative than the mean?
3. What question does a row percentage in a cross-tabulation answer?
4. Distinguish descriptive from inferential statistics.
5. What is the difference between a Type I and a Type II error?

Long Answer Questions

1. Explain the stages of data preparation and propose quality controls for an online survey.
2. Compare the mean, median, mode, standard deviation and interquartile range, including suitable business applications.
3. Explain how a dashboard can communicate useful information while avoiding misleading visual representation.
4. Describe the complete procedure for conducting and reporting a hypothesis test.
5. Compare SPSS and R as tools for business research, including reproducibility and governance.

Analytical / Application Questions

1. A retailer reports average spending of 210 currency units, but the median is 95 and a few accounts exceed 10,000. Which summaries and visualizations should be reported, and why?
2. A cross-tabulation shows higher renewal among premium customers. Design an analysis that distinguishes association from a plausible causal effect.
3. Critique a chart with a vertical axis beginning at 98%, three-dimensional columns and no sample size. Explain how you would redesign and interpret it.

Multiple Choice Questions with Answers

Question 1

Which measure is generally most resistant to an extreme outlier?

A. Mean

B. Median

C. Variance

D. Range

Correct Answer: B. Median

Question 2

A row percentage in a cross-tabulation of customer segment by renewal answers which question?

A. What proportion of all customers are in each segment?

B. Within each segment, what proportion renewed?

C. Among renewers, what proportion belongs to each segment?

D. What is the average renewal time?

Correct Answer: B. Within each segment, what proportion renewed?

Question 3

What is the main purpose of descriptive statistics?

A. To prove causation

B. To summarize observed data

C. To eliminate sampling bias

D. To guarantee a significant result

Correct Answer: B. To summarize observed data

Question 4

Which statement about a p-value is correct?

A. It is the probability that the null hypothesis is true

B. It is the probability that the sample was selected correctly

C. It is calculated under the null model as the probability of results at least as extreme as observed

D. It measures practical importance directly

Correct Answer: C. It is calculated under the null model as the probability of results at least as extreme as observed

Question 5

A Type I error occurs when a researcher:

A. Fails to reject a false null hypothesis

B. Rejects a true null hypothesis

C. Uses a median instead of a mean

D. Reports a confidence interval

Correct Answer: B. Rejects a true null hypothesis

Question 6

Which visualization is most suitable for examining the relationship between two quantitative variables?

A. Scatterplot

B. Pie chart

C. Bar chart of categories

D. Frequency table of one variable

Correct Answer: A. Scatterplot

Question 7

If values are 2, 3, 3, 4 and 13, what is the median?

A. 3

B. 4

C. 5

D. 13

Correct Answer: A. 3

Question 8

Why is the denominator important when reporting percentages?

A. It determines the software brand

B. It identifies the population mean automatically

C. Different denominators answer different analytical questions

D. It removes all missing data

Correct Answer: C. Different denominators answer different analytical questions

Question 9

Which action best supports reproducible analysis?

A. Changing the raw file directly without a record

B. Saving code, transformations, definitions and software versions

C. Deleting unusual observations automatically

D. Reporting only statistically significant results

Correct Answer: B. Saving code, transformations, definitions and software versions

Question 10

Which statement best describes a dashboard?

A. A complete replacement for research design

B. A decorative collection of all available data

C. A decision-oriented display of selected indicators and visualizations

D. A method that proves causation from descriptive data

Correct Answer: C. A decision-oriented display of selected indicators and visualizations