These 12 CBDA statistics questions test the statistical reasoning the real exam leans on: choosing mean or median, flagging outliers with IQR fences, reading p-values against alpha, naming Type I errors, interpreting confidence intervals, sizing samples, and reading correlation and residual plots. Eight questions need a small calculation or a tiny table. Every answer is explained and tied to a section of the Guide to Business Data Analytics.
How this CBDA statistics questions set is built
The set targets the statistical techniques behind Domain 3 (Analyze Data, 16%) and Domain 4 (Interpret and Report Results, 20%). Our author who passed found the real exam more focused on statistical reasoning than on definitions, as described in our CBDA certification overview. So each question asks what a numerate analyst should conclude or do next.
The questions draw on five parts of the Guide:
- 3.9 Descriptive and Inferential Statistics: central tendency and dispersion, with a warning that the mean reacts to outliers and hides fluctuation; correlation and regression as inferential tools.
- 3.12 Hypothesis Formulation and Testing: the analyst turns a business hunch into a null and alternate hypothesis, fixes the significance level before collecting samples, and reports results with their uncertainty.
- 3.11 Exploratory Data Analysis and 2.3.3 Explore Data: distributions, collinearity and outliers are checked before modelling; boxplots spot outliers, which are investigated before anyone drops them.
- 3.18 Technical Visualizations: Box-Cox plots for skewed predictors, correlation maps for dependent inputs, residual plots for model fit.
- 2.2.2 Determine the Data Sets: sample size is decided while defining data sets.
Scenarios follow the exam format: a short business situation, then the best interpretation or next action. Difficulty sits at or slightly above the real exam. All questions are original, not exam dumps. For the wider domain, pair this set with our CBDA Analyze Data practice questions.
| Sub-topic | Questions |
|---|---|
| Central tendency and dispersion | 1, 3 |
| Outliers, skew and transformation | 2, 4 |
| Hypothesis formulation and p-values | 5, 6 |
| Type I and Type II errors | 7 |
| Confidence intervals and sample size | 8, 9 |
| Correlation, regression and model fit | 10, 11, 12 |
CBDA exam at a glance
| Fact | Detail |
|---|---|
| Format | 75 multiple-choice, scenario-based questions |
| Time | 2 hours |
| Domain 3: Analyze Data | 16% |
| Domain 4: Interpret and Report Results | 20% |
| Exam fee | US$405 (IIBA members US$250); first year of membership included |
| Passing score | Not published by IIBA |
Last verified: October 2026
Practice Questions
Question 1
Lindqvist Home Goods tracks delivery times for a new regional courier. A sample of nine orders took 2, 3, 3, 4, 4, 5, 5, 6 and 31 days. The operations director wants one number for the weekly report that describes a typical delivery.
Which figure should the analyst report?
- A. The mean of 7 days, because it uses every order in the sample
- B. The median of 4 days, and flag the 31-day order for investigation
- C. The mode, because the most frequent value is the clearest summary
- D. The standard deviation, because it shows how reliable the courier is
Question 2
Baltera Grocers is reviewing online basket values before building a demand model. The data scientist reports Q1 = €42 and Q3 = €78 and plots a boxplot using the standard 1.5 × IQR rule. Three baskets stand out: €125, €138 and €210.
What should the analyst conclude and do?
- A. Only €210 is an outlier; remove it and continue with the model
- B. All three baskets are outliers; exclude them from the training data
- C. None of the baskets are outliers, so no further review is needed
- D. €138 and €210 are outliers; check how they were recorded first
Question 3
Ostrava Parts is choosing between two packaging suppliers for a just-in-time assembly line. Over six recent deliveries, Supplier X took 10, 10, 11, 9, 10 and 10 days, and Supplier Y took 4, 16, 10, 7, 13 and 10 days. The plant manager says the suppliers are equal because both average 10 days.
How should the analyst respond?
- A. Supplier X is more predictable because its spread is far smaller
- B. The suppliers are equivalent because their means are identical
- C. Supplier Y is better because it sometimes delivers in four days
- D. No view is possible until a regression model has been built
Question 4
CloudLedger, a B2B accounting platform, wants to predict renewal from the number of active users per account. A histogram shows most accounts have 1 to 25 users, while a few enterprise accounts have over 3,000. The data scientist proposes deleting the enterprise accounts so the predictor looks normally distributed.
What should the analyst recommend?
- A. Delete the enterprise accounts, since normality matters most for the model
- B. Replace the large values with the median user count for all accounts
- C. Keep the accounts and discuss transforming the skewed predictor instead
- D. Switch to a pie chart so stakeholders can see the share of each segment
Question 5
The warehouse manager at Nordhaven Logistics claims a new pick-routing tool has cut the average pick time below 90 seconds. The analyst plans to test the claim using a sample of timed picks.
Which formulation is correct?
- A. H0: mean pick time is below 90 seconds; H1: it is 90 seconds or more; upper-tailed test
- B. H0: mean pick time is 90 seconds or more; H1: it is below 90 seconds; lower-tailed test
- C. H0: mean pick time equals 90 seconds; H1: it differs from 90 seconds; two-tailed test
- D. H0: mean pick time is 90 seconds or less; H1: it is above 90 seconds; upper-tailed test
Question 6
Meridian Health Insurance tests whether a redesigned claims form reduces errors. Before data collection, the compliance lead and the analyst agree on a significance level of 0.01 because a wrong rollout is costly. The test returns a p-value of 0.03, and the product owner asks to report the result as significant at 0.05.
What should the analyst do?
- A. Report that H0 is not rejected at the agreed 0.01 level, citing the p-value
- B. Change the significance level to 0.05 so the redesign can go live sooner
- C. Report that the redesign cuts errors by 97%, since the p-value is 0.03
- D. Discard the test and run the experiment again until the p-value drops
Want more than 12? Create a free account on certificate.tips to unlock more CBDA practice questions, including the full 75-question CBDA exam simulator with exam and practice modes. certificate.tips is our product.
Question 7
Fjordline Travel ran an A/B test on a new booking page with H0: the new page does not change conversion. The team rejected H0, concluded the page lifted conversion, and rolled it out. A much larger holdout study six months later found no difference between the old and new pages.
Which error did the original test most likely make?
- A. A Type II error, because it failed to detect a real effect
- B. A sampling bias error, because the holdout study was larger
- C. A confidence interval error, because the interval was too wide
- D. A Type I error, because it rejected a null hypothesis that was true
Question 8
Pellucid Software surveyed 100 new customers on onboarding satisfaction (0 to 100 scale). The sample mean is 32 and the sample standard deviation is 20. The customer success lead wants to announce that average satisfaction now exceeds the target of 30.
Using a 95% confidence interval (mean ± 1.96 standard errors), what should the analyst say?
- A. The claim holds, because the sample mean of 32 is above the target of 30
- B. The claim holds, because 95% of customers scored between 28 and 36
- C. The claim is not supported, because the interval includes values below 30
- D. The claim is not supported, because a sample of 100 can never be used
Question 9
Calderon Retail ran a store-experience survey of 400 shoppers, and the data team reported a margin of error of ±4 points. The regional director wants the next survey to have a margin of ±2 points at the same confidence level.
About how many responses will the next survey need?
- A. About 1,600 responses
- B. About 800 responses
- C. About 600 responses
- D. About 3,200 responses
Question 10
At Tamsin Contact Centres, the analyst finds a correlation of r = 0.6 between monthly training hours and agents' quality scores. In the steering meeting, the operations head says training explains 60% of the differences in quality scores and asks for the training budget to double.
How should the analyst correct the statement?
- A. Confirm it, because a correlation of 0.6 means 60% of variation is explained
- B. Say the correlation is too weak to matter, because it is below 0.8
- C. Say the figure should be 40%, because the remaining share is unexplained
- D. Explain that r² is 0.36, so about 36%, and it shows association only
Question 11
Halvard Beverages is building a weekly sales model. The correlation map of candidate predictors shows:
| Pair | r |
|---|---|
| Ad spend and number of retail partners | 0.91 |
| Ad spend and average price | -0.12 |
| Number of retail partners and average price | -0.08 |
| Temperature and average price | 0.05 |
What should the analyst raise with the data scientist?
- A. Average price should be removed, because its correlations are negative
- B. Ad spend and partner count overlap, so their effect may be double counted
- C. Temperature should be removed, because it correlates weakly with price
- D. All four predictors are independent, so the model can use all of them
Question 12
Rosewell Energy uses a linear regression to forecast daily gas demand from outdoor temperature. The residual plot of errors against predicted values forms a clear inverted U: residuals are negative at both ends and positive in the middle. The finance team wants to use the forecast in next quarter's budget.
What should the analyst do NEXT?
- A. Approve the forecast, because residuals average close to zero overall
- B. Present the residual plot to finance as the main budget visual
- C. Ask the data scientist to revisit the model form for a non-linear pattern
- D. Remove the days with the largest residuals and refit the same model
Ready for the full exam experience? Sign up and create a free account to keep practising. The CBDA exam simulator has a full 75-question set you can take in timed exam mode or review step by step in practice mode.
Answer Key and Explanations
1. Correct answer: B. One 31-day order pulls the mean up to 7 days; the other eight orders average 4 days, the same as the median. The Guide notes that the mean is sensitive to outliers and that the right measure depends on the research problem. The 31-day order still needs a check for a lost parcel or an entry error. Source: Guide to Business Data Analytics, 3.9 Descriptive and Inferential Statistics Why the others are wrong: A overstates a typical delivery by 75%; C fails because 3, 4 and 5 each appear twice; D describes spread, not a typical value.
2. Correct answer: D. IQR = 78 − 42 = 36, so the fences are −€12 and 78 + 1.5 × 36 = €132; only €138 and €210 lie beyond them. The Guide names boxplots as the outlier visual, and its Explore Data example shows an analyst checking collection rules and finding that extreme prices were placeholders. NIST also advises investigating outliers before deleting them. Source: Guide to Business Data Analytics, 3.11 Exploratory Data Analysis; 2.3.3 Explore Data Why the others are wrong: A uses a 3 × IQR fence (€186) and deletes unchecked; B uses Q3 + 1 × IQR (€114); C ignores two values above the fence.
3. Correct answer: A. Both means are 10 days, but the sample standard deviation is about 0.63 days for X and 4.24 days for Y. The Guide makes the same point with two stocks that share a mean return, where the one with wild daily swings is riskier. A just-in-time line needs predictability. Source: Guide to Business Data Analytics, 3.9 Descriptive and Inferential Statistics Why the others are wrong: B ignores dispersion; C rewards one fast delivery and ignores the 16-day one; D delays a decision descriptive statistics already support.
4. Correct answer: C. The Guide explains that many models assume normally distributed predictors and that heavily skewed variables may be rescaled, which a Box-Cox plot helps assess. It cautions that removing extreme values can conflict with the business context; large enterprise accounts are real and valuable. Source: Guide to Business Data Analytics, 3.18 Technical Visualizations Why the others are wrong: A discards valid, high-value customers; B distorts the predictor; D changes the visual without fixing the model.
5. Correct answer: B. The Guide defines the null hypothesis as the default, no-change position and the alternate as its negation. The manager's claim needs evidence, so it becomes H1, and only the lower tail matters, as in the Guide's lower-tailed bank-visits example. Source: Guide to Business Data Analytics, 3.12 Hypothesis Formulation and Testing Why the others are wrong: A puts the claim in H0; C tests for any difference; D tests for an increase.
6. Correct answer: A. A p-value of 0.03 is above the agreed alpha of 0.01, so H0 is not rejected. In the Guide's seven-step process, the significance level and decision rule are fixed before samples are collected. Our guide to MVP experiment design covers setting success criteria before a test starts. Source: Guide to Business Data Analytics, 3.12 Hypothesis Formulation and Testing Why the others are wrong: B changes the rule after seeing the data; C misreads the p-value as an effect size; D retests until a desired result appears.
7. Correct answer: D. Rejecting a true null hypothesis is a Type I error, or false positive. The Guide lists Type I and II errors as background analysts need and warns that test results are probabilistic. NIST adds that alpha is the risk of this error, and lowering alpha raises the Type II risk. Source: Guide to Business Data Analytics, 3.12 Hypothesis Formulation and Testing Why the others are wrong: A describes missing a real effect; B confuses sample size with bias; C names no recognised error type.
8. Correct answer: C. The standard error is 20 / √100 = 2, so the 95% interval is 32 ± 1.96 × 2, about 28.1 to 35.9. It includes values below 30, so the claim is not supported at 95% confidence. The Guide uses a customer-score survey to show why tests attach a confidence interval. Source: Guide to Business Data Analytics, 3.12 Hypothesis Formulation and Testing Why the others are wrong: A ignores sampling uncertainty; B treats an interval for the mean as a range of individual scores; D overstates the problem.
9. Correct answer: A. The Guide describes the standard error as the standard deviation divided by the square root of the number of observations. Halving the margin requires √n to double, so n grows fourfold: 400 × 4 = 1,600. Sample size is set when determining data sets. Source: Guide to Business Data Analytics, 3.12 Hypothesis Formulation and Testing; 2.2.2 Determine the Data Sets Why the others are wrong: B assumes precision grows in step with n; C is too few; D assumes eightfold growth.
10. Correct answer: D. In simple linear regression, the share of variation explained is r², and 0.6² = 0.36. The Guide describes correlation as a scaled covariance between −1 and 1 that suggests a relationship only in the right context; tenure, for example, may drive both training and quality. Source: Guide to Business Data Analytics, 3.9 Descriptive and Inferential Statistics Why the others are wrong: A confuses r with r²; B applies an arbitrary cut-off; C subtracts instead of squaring.
11. Correct answer: B. An r of 0.91 means the two inputs carry much of the same information. The Guide gives a near-identical example: correlated ad spend and distribution channels risk double counting sales. Correlation maps help remove or combine such variables. Source: Guide to Business Data Analytics, 3.9 Descriptive and Inferential Statistics; 3.18 Technical Visualizations Why the others are wrong: A treats a negative sign as a problem; C judges a predictor by its link to another input; D ignores the 0.91 pair.
12. Correct answer: C. Random residual scatter suggests a good linear fit; a clear pattern means missing structure. The Guide's residual plot example uses the same inverted U to signal a non-linear relationship, so the model form should be revisited before finance relies on it. Source: Guide to Business Data Analytics, 3.18 Technical Visualizations Why the others are wrong: A ignores systematic errors; B uses a technical visual for a business audience; D hides the pattern.
How to score this set
- 11–12 correct: Your statistical reasoning is exam-ready. Move to timed full-length practice.
- 8–10 correct: Solid base. Review the questions you missed, redo the arithmetic by hand, and reread 3.12 on how hypotheses and significance levels are set.
- Below 8: Rebuild the basics first: dispersion and outliers (3.9, 3.11), then confidence intervals and Type I and II errors (3.12), then regression diagnostics (3.18).
Statistics questions also show up inside reporting scenarios, so follow up with our Interpret and Report Results quiz. When you want a full timed run, the CBDA exam simulator at /exams/cbda has a 75-question set, and you can create a free account to save your progress.
What candidates say about statistics on the CBDA
The certificate.tips author who passed flagged about 20 of 75 questions for review and found the exam significantly harder than question banks, with more weight on statistical reasoning and analytical decisions than on definitions. Other candidates stress practice volume. Mufaddal Haidermota, who earned the CBDA in December 2022, recommends per-domain mock questions, full-length tests, mind maps and cheat sheets, and notes his course videos were thin in places. Trainer LN Mishra of Adaptive US advises daily practice questions, timed mocks and reviewing every wrong answer. A Techcanvass candidate story describes a developer who used weekend trainer-led sessions and two to three hours of daily study. Practitioners agree on one habit: work the numbers yourself rather than memorising answers.
FAQ
How much statistics is on the CBDA exam?
IIBA does not publish a count of statistics questions. The techniques behind them, such as descriptive statistics, hypothesis testing and EDA, support Domains 3 and 4, which together carry 36% of the exam. Candidates who have passed report that statistical reasoning appears often in scenario questions.
Do I need to calculate p-values or test statistics by hand on the CBDA?
The Guide says the analyst's main skill is formulating the hypothesis and interpreting the result, while a data team runs the test. Expect to compare a p-value with alpha, read a confidence interval, or do simple arithmetic such as a standard error. Heavy formula work is unlikely.
What is the difference between r and R² for the exam?
r is the correlation coefficient, from −1 to 1, and shows the direction and strength of a linear relationship. In simple linear regression, R² equals r squared and gives the share of variation explained. An r of 0.6 therefore explains about 36% of the variation, not 60%.
Which charts should I know for CBDA statistics questions?
Know histograms and probability distribution plots for shape, boxplots for outliers, scatter plots and correlation maps for relationships, and residual plots for model fit. The Guide also covers Box-Cox plots for skewed data. Our CBDA Analyze Data questions cover more of these visuals.
Sources
- Business Data Analytics Certification (CBDA) — IIBA, checked October 2026
- Certification Fees — IIBA, checked October 2026
- CBDA Certification Handbook — IIBA, checked October 2026
- What are statistical tests? — NIST/SEMATECH e-Handbook of Statistical Methods, checked October 2026
- Critical values and p values — NIST/SEMATECH e-Handbook of Statistical Methods, checked October 2026
- What are confidence intervals? — NIST/SEMATECH e-Handbook of Statistical Methods, checked October 2026
- What are outliers in the data? — NIST/SEMATECH e-Handbook of Statistical Methods, checked October 2026
- Business Data Analytics Certification — Mufaddal Haidermota, Medium, January 2023
- CBDA Certification Journey: Ivan — Techcanvass, 2024
- Mastering the CBDA Exam — LN Mishra, Adaptive US, October 2024