On this page
Your assignment
Role: Analyst supporting Juniper Wholesale's controller and finance committee
Deliverable: A reproducible calculation appendix, a two-page accounting data-control note, and a three-page finance-committee memo that state the target and source contract, compare both sample distributions, construct and challenge the separate mean intervals, disposition the 78-day record, separate supported and unsupported claims, and prioritize next evidence.
Evidence basis: fictional
Evaluation criteria
- data identity and source control (30%): Defines and controls unit, variables, scales, population, frame, sample design, source, cutoff, uniqueness, completeness, eligibility, missingness, and the retained tail record.
- quantitative accuracy and reconciliation (30%): Reproduces ordered values, center, quartiles, fences, sample variance and standard deviation, standard error, margin, endpoints, differences, units, and full-precision reconciliations.
- inference and uncertainty boundary (25%): Distinguishes statistic from parameter, observation spread from estimate precision, sampling from nonsampling error, and description or estimation from test, cause, forecast, risk, valuation, and decision.
- audience specific communication (15%): Gives accounting readers executable controls and finance readers a concise bounded conclusion with specific competing explanations, stop conditions, and prioritized evidence requests.
Assignment context
Juniper Wholesale's controller receives a dashboard with one sentence:
Average settlement fell from 31.88 to 29.38 days, proving that the Year 2 credit-policy change improved collections and will reduce future credit risk.
The dashboard omits the population definition, sampling frames, medians, spread, 78-day invoice, interval method, and source controls. A second analyst objects that the two 95% intervals overlap and therefore the years are exactly the same.
Neither conclusion is defensible from the packet. Replace them.
Evidence packet
Use the Juniper invoice-settlement samples. Every entity, record, date, and value is fictional. The packet supplies the data dictionary, target population, source and cutoff, frame sizes, selection seeds and probabilities, 32 selected records, interval convention, quality controls, and limitations.
Do not assume facts outside it. In particular, no credit-policy change, customer or product mix, open-invoice analysis, dispute cause, customer-level independence, loss experience, future outcome, or decision objective is given.
Calculation appendix
- State the unit, response variable, target populations, frames, sample sizes, selection design, source, cutoff, missing policy, outlier policy, units, and measurement scales.
- Reproduce both selected ID lists and test uniqueness, completeness, valid values, sample counts, and selection probabilities.
- Present the ordered settlement values for each year.
- Compute count, sum, mean, median, minimum, maximum, range, Q1, Q3, IQR, 1.5-IQR inner fences, potential-outlier count, sum of squared deviations, sample variance, and sample standard deviation.
- Compute Year 2 minus Year 1 differences in mean, median, and sample standard deviation. Label the standard-deviation subtraction as a descriptive arithmetic comparison, not a statistic with its own sampling distribution.
- Compute each standard error, supplied-t margin, and endpoint at full precision. Reconcile interval midpoints and half-widths, then show a display-rounded version.
Accounting data-control note
Design executable tests for row grain, unique invoice ID, issue and final-application dates, eligibility, classification, frame completeness, excluded and open invoices, late postings, joins, record and amount controls, selection reproduction, and the 78-day source trace. Separate the packet's stipulations from evidence you would require in an actual engagement.
Finance-committee memo
- State the sample-level center, spread, and tail comparison.
- Explain statistic, parameter, sampling variability, standard deviation, standard error, margin of error, and nonsampling error in this fact pattern.
- Interpret the 95% method without assigning its coverage probability directly to one fixed realized interval.
- Explain why separate interval overlap is not a two-sample test.
- Give at least four competing explanations for the observed distributions.
- Name unsupported causal, forecast, risk, valuation, audit, and policy claims.
- Prioritize at least ten specific evidence requests and state which question each would answer.
Constraints
Retain the 78-day invoice in the primary calculation. You may propose a separately labeled sensitivity or robust view, but may not replace the original population or result silently. Preserve full precision until final display. Do not reproduce credential questions or claim CPA, CFA, FINRA, audit, or regulatory coverage.