Research is a systematic way of finding answers so that decisions rest on evidence, not hunches. Business research covers consumer studies, process improvement, policy evaluation and academic work. This article revisits the end-to-end process and the key tools.
1. The research process
| Step | Key question |
|---|---|
| Define the problem | What decision needs information? A clear problem is half solved |
| Review literature | What is already known? What gap remains? |
| Objectives and hypotheses | What exactly will be tested? |
| Research design | How will data be gathered and analysed? |
| Sampling and data collection | From whom, and how? |
| Analysis and interpretation | What do the data say? |
| Report | What should the reader do with it? |
Research types: basic (pure) vs applied; qualitative vs quantitative; cross-sectional vs longitudinal. A variable is a characteristic that changes; independent variables (causes), dependent variables (effects), moderating and mediating variables explain how and when an effect occurs; control variables are held steady.
2. Research design
| Design | Purpose | Methods |
|---|---|---|
| Exploratory | Understand a vague problem; generate ideas | Literature review, interviews, focus groups, pilot studies |
| Descriptive | Describe who, what, how often | Surveys, observation, panels |
| Causal (experimental) | Establish cause and effect | Experiments with a control group, random assignment |
For causation you need association (variables move together), time order (cause precedes effect) and ruling out alternative explanations. Experiments use a treatment group and a control group; the strongest is the randomised controlled trial. Internal validity asks whether the effect is truly caused by the treatment; external validity asks whether the result generalises.
In the real world: online firms run A/B tests constantly: half the visitors see page A, half page B, and the difference in conversions is a causal result. Drug trials use double-blind randomised controlled trials so that neither patients nor doctors know who got the treatment.
3. Measurement and scales
Measurement assigns numbers to characteristics. The four scales decide which statistics are valid:
| Scale | Property | Example | Valid statistics |
|---|---|---|---|
| Nominal | Categories, no order | Gender, city | Mode, counts, chi-square |
| Ordinal | Ranked order | Rank, satisfaction level | Median, rank tests |
| Interval | Equal intervals, no true zero | Temperature in °C, Likert scores (often treated so) | Mean, SD, t-test |
| Ratio | Equal intervals and a true zero | Income, sales, age | All |
Attitude scales: Likert (strongly disagree to strongly agree), semantic differential (bipolar adjectives), Thurstone, Guttman, and rank-order or constant-sum scales. Good measures need:
- Reliability: consistent results (test–retest, split-half, Cronbach’s alpha, usually above 0.7).
- Validity: measures what it claims (content, criterion, construct validity).
4. Sampling
The population is everyone of interest; a sample is a subset. A census studies everyone. The sampling frame is the list from which the sample is drawn.
| Probability (random) methods | How |
|---|---|
| Simple random | Every unit has an equal chance |
| Systematic | Every k-th unit from a random start |
| Stratified | Divide into groups (strata) and sample each; improves precision |
| Cluster | Randomly choose groups (villages, schools) and study all or some inside |
| Multi-stage | Several levels of sampling |
| Non-probability methods | How |
|---|---|
| Convenience | Whoever is easy to reach |
| Judgement (purposive) | Experts choose typical cases |
| Quota | Fill quotas by category |
| Snowball | Respondents refer others; for hard-to-reach groups |
Sample size depends on the confidence level, margin of error and variability: for a proportion, n ≈ z² p(1 − p) ÷ e². At 95% confidence (z = 1.96), p = 0.5 and a ±5% margin, n ≈ 1.96² × 0.25 ÷ 0.0025 ≈ 384. Errors: sampling error (random) and non-sampling error (bias, non-response, measurement mistakes).
In the real world: the 1936 Literary Digest poll surveyed millions yet wrongly predicted Alf Landon would beat Franklin Roosevelt, because its sample came from car owners and telephone lists, which skewed wealthy. Size does not fix a biased frame. Election exit polls in India use stratified, multi-stage samples for this reason.
5. Data collection
Primary data are collected fresh for the study; secondary data already exist (company records, government statistics, industry reports, journals).
| Method | Strength | Weakness |
|---|---|---|
| Survey (questionnaire) | Reaches many, cheap, standard | Response bias, low response |
| Interview | Depth, probing | Costly, interviewer bias |
| Observation | Actual behaviour | Cannot see motives |
| Focus group | Group insight, ideas | Dominant voices, not generalisable |
| Experiment | Cause and effect | Artificial settings |
| Case study, ethnography | Rich context | Hard to generalise |
Questionnaire design: keep questions short, clear and neutral; avoid double-barrelled (“Is the product cheap and durable?”) and leading questions; order from easy to sensitive; pretest with a pilot.
6. Hypothesis testing and analysis
A hypothesis is a testable statement. The null (H₀) says there is no effect or difference; the alternative (H₁) says there is. Steps: state hypotheses, choose a significance level (α, usually 0.05), compute a test statistic, find the p-value and decide: if p < α, reject H₀.
| Error | Meaning |
|---|---|
| Type I (α) | Rejecting a true null (false alarm) |
| Type II (β) | Failing to reject a false null (missed effect); power = 1 − β |
Common tests:
| Test | Use |
|---|---|
| Z-test and t-test | Compare a mean with a value, or two means |
| Paired t-test | Before and after on the same people |
| ANOVA | Compare three or more means |
| Chi-square | Association between two categorical variables; goodness of fit |
| Non-parametric (Mann–Whitney, Kruskal–Wallis) | Ordinal data or non-normal samples |
Correlation (r from −1 to +1) shows the strength and direction of a linear association; correlation is not causation. Regression predicts a dependent variable: Y = a + bX. R² is the share of variation explained; the slope b is the change in Y per unit of X. Example: if sales = 20 + 3 × (ad spend in ₹ lakh), each extra lakh of spend is associated with 3 more units of sales (₹ lakh), and R² of 0.64 means 64% of variation is explained. Multivariate methods include multiple regression, factor analysis (reduce many variables to a few factors), cluster analysis (group similar cases) and discriminant analysis (classify).
In the real world: a famous trap is spurious correlation: ice-cream sales and drownings rise together because both depend on summer heat. Marketing analysts guard against this by testing, not just correlating.
7. Report writing and research ethics
A report has a title page, executive summary, introduction and objectives, literature review, methodology, results, discussion, conclusions and recommendations, references and appendices. Use tables and charts that make one point each. Citation formats (APA, Harvard) give credit and avoid plagiarism.
Ethics: informed consent, confidentiality and anonymity, no harm to participants, honest reporting (no fabrication or selective results), and respect for intellectual property.
One-minute recap
| Topic | Key idea | Where you meet it |
|---|---|---|
| Process | Problem, design, data, analysis, report | Any study |
| Design | Exploratory, descriptive, causal | A/B tests, surveys |
| Measurement | Nominal to ratio; reliability and validity | Questionnaires |
| Sampling | Probability vs non-probability; sample size | Polls, audits |
| Data collection | Primary and secondary; question design | Market studies |
| Hypothesis tests | p-value, errors, t, ANOVA, chi-square | Evidence-based decisions |
| Regression | Prediction; correlation is not causation | Forecasting, driver analysis |
| Ethics | Consent, confidentiality, honesty | Every study |
Next in MBA Revisited: Quantitative Techniques.
Written by Claude, Anthropic’s AI assistant, as part of MBA Revisited, a quick refresher of the PGDM subjects one at a time.