Analytical Capability Academy · Module 04

The Rigor
of Analytical Work

What separates work that survives challenge from work that only looks like it would.

J
Fast-Track Validation
K
Model Choice & Assumptions
L
Learning by Disproving
Corporate Training - 2026 · Cards J · K · L
Module 04 · The Rigor

The Discipline of Analytical Rigor

Three cards on what separates work that survives challenge from work that only looks like it would.

Sophistication on a flawed premise produces confident wrong answers;

Simplicity grounded in sound reasoning produces honest provisional ones.
Card J Module 04

Validation vs Verification

THE ORDER THAT PROTECTS THE WORK STEP 01 Validation Does the logic make sense? Can this hypothesis even produce the observed effect? ~15 minutes passes? STEP 02 Verification Does the model implement the analysis correctly? Is the math right? days to weeks Skipping step 01 means verifying a model that should not have been built.
Validation is cheap and slow; verification is expensive and fast-to-look-rigorous.
Card J Module 04

The Plausibility Check

15 minutes of arithmetic that can prevent weeks of confidently wrong analysis.

An analyst at a telecom operator is asked to investigate a 6% decline in quarterly service revenue. A stakeholder proposes: "It's the new prepaid bundle redesign that launched in week three — subscribers are confused and not finding the right plan." Before building any model, the disciplined analyst runs a quick check:

BACK-OF-THE-ENVELOPE PLAUSIBILITY CHECK
1 · Bundle launched in week 3 of a 13-week quarter → active ~10 weeks (~77% of the quarter)
2 · The affected bundle accounts for ~40% of total revenue
3 · Even if every subscriber was affected → max impact = 40% × 77% × effect size
4 · To produce a 6% revenue decline → bundle would have to depress this product ~19.5%
5 · Historical benchmark on bundle changes: 1–3% typical effect
Conclusion: The hypothesis cannot explain the observed magnitude.
Card J Module 04

The Plausibility Check

15 minutes of arithmetic that can prevent weeks of confidently wrong analysis.

An analyst at a telecom operator is asked to investigate a 6% decline in quarterly service revenue. A stakeholder proposes: "It's the new prepaid bundle redesign that launched in week three — subscribers are confused and not finding the right plan." Before building any model, the disciplined analyst runs a quick check:

What the check accomplished — and didn't

The check did not prove the redesign had zero effect; it almost certainly had some. What it established is that the redesign cannot be the dominant cause of the 6% decline.
It does not produce a precise answer; it produces a precise rejection — in 15 minutes, before any model.

Card J Practice · Interactive

Practice · Plausibility Checker

Pick the value the math actually allows at each step.

Pick a value for each step in the BOTE check, then choose your verdict.
Card J Module 04

Simple Tools, Used Well

Willingness to use the simplest tool that answer questions is analytical maturity.

ToolWhat it doesWhen to reach for it
Back-of-the-envelope calculationsQuick arithmetic that establishes whether a proposed effect could plausibly produce the observed outcome. Order-of-magnitude reasoning, not precision.Always, before any formal model. The default first move on any new hypothesis.
Rough scenario tablesSimple matrices examining outcomes under best-, worst-, and base-case assumptions for the main parameters.Testing whether a hypothesis is consistent with observed magnitudes under different assumptions.
Simple ratios & ratesPer-subscriber, per-period, per-unit comparisons that ground the analysis in interpretable benchmarks.Sanity-checking whether the implied behaviour is reasonable at the unit level.
Spreadsheet logic checksLight models that test whether the proposed mechanism produces the observed effect when reasonable inputs are applied.When a single calculation is too coarse but a full model is still premature.
Card K Module 04

Models Don't Reveal Truth

They formalise assumptions then estimate parameters in the framework they create.

Every model is a mathematical statement of belief about the system being studied.

In practice, models do not reveal reality.
They encode assumptions about how reality works — then estimate parameters within the framework those assumptions create.
Linear regression assumes effects add up in straight lines.
Tree methods assume the data carves into rectangular regions.
None of these are facts about the world; they are choices about how to represent it.

Card K Module 04

Complex Models Hide Assumptions

More sophisticated models do not eliminate assumptions — they merely make them harder to see.

⚠ More complex models do not eliminate assumptions

A random forest assumes the data can be partitioned by axis-aligned splits. A gradient-boosted model assumes errors can be successively corrected by additive weak learners. A neural network assumes the function lies within the class its architecture can represent.

These assumptions are less visible than those of linear regression — which makes them more dangerous, not less. The disciplined analyst asks the same question regardless of sophistication: what is this method assuming, and would those assumptions hold for my problem?

Telecom Business Example
The pricing model that averaged across reality

An analyst models how subscriber demand responds to tariff changes using linear regression. The model estimates a $1 increase reduces demand by 4 units. The coefficient is significant, the R² acceptable. The output looks clean.

In reality demand is highly nonlinear: at low prices subscribers are insensitive; at moderate prices sensitivity grows; at high prices demand collapses as they switch to competitors. The linear model averaged across these regimes — "we can raise prices by $3 with limited impact" may be dangerously incorrect. The math was right; the choice of model was wrong.

Card K Module 04

What Common Models Assume

Match the model to the question — not the question to the model.

If the question is about…Outcome typeAppropriate model familyKey assumptions
How much (continuous magnitude)Continuous variable (revenue, cost, time)Linear regression, generalised linear modelsLinearity of relationships, additivity, stability across the predictor range
Whether (binary classification)Yes/no, will/won't, has/hasn'tLogistic regression, classification treesProbabilities are the appropriate target; predictors influence log-odds linearly
When (timing of event)Time-to-event with possible censoringSurvival analysis (Cox models, Kaplan-Meier)Hazard rates can be modelled; censoring is non-informative
Which group (multi-class)Categorical with multiple classesMultinomial regression, classification methodsCategories are mutually exclusive; a probabilistic structure is appropriate
The danger signal: reshaping the question to fit the tool

When the analyst finds themselves saying "let's just dichotomise the outcome" or "let's just assume linearity for now" — that is a signal to step back. The model is selected because it matches the question, not because it is convenient.

Card K Practice · Interactive

Practice · Model Match

For each business question, pick the model family, the most consequential assumption it embeds, and what would invalidate that assumption.

Pick a model family, the key assumption it embeds, and what would invalidate that assumption.
Card K Module 04

Reading Outputs Conditionally

A coefficient is not a finding by itself — it is a finding under a model's assumptions.

A coefficient of 0.34 with p<0.001 is not a finding by itself. It is a finding under the model's assumptions. If they hold, the coefficient is meaningful. If they do not, it is an artefact of the model rather than a fact about the world.

Unconditional languageConditional language
"The model proves that pricing drives churn at 4.2% per dollar increase.""Under the assumptions of the linear model, the data is consistent with a churn effect of 4.2% per dollar, within the observed price range. If the relationship saturates above $X, the estimate would shift downward."
"The data shows that account-care calls drive renewal.""Holding the observable controls fixed, accounts receiving more calls renewed at a higher rate. Whether this is causal depends on whether call selection is independent of unobserved retention factors."
"The algorithm found three segments.""Under the clustering algorithm's distance metric and chosen cluster count, the data partitions into three groups. A different metric or count would produce a different partition."
Why this honesty preserves credibility

When stakeholders later discover that conclusions depended on assumptions the analyst did not realise they were making, trust in the analytical function erodes. Communicating findings as conditional is harder to deliver and easier to defend.

Key Takeaway
Models do not discover truth; they formalise assumptions.
Card L Module 04

The First Result Is Not the Conclusion

Strong analysis treats initial results not as findings, but as claims that must survive challenge.

This card addresses a common failure: stopping analysis as soon as a model produces a "good" result. Once a model appears significant or performs well on a chosen metric, the work is considered complete. This feels efficient and produces results that look rigorous — but it creates a dangerous illusion of certainty.

The work is not done when a result emerges; the work is done when the result has been tested and survived.

The trap is structural. Modelling tools produce outputs whether or not the underlying analysis is sound. A regression with significant coefficients looks identical to one whose coefficients would vanish under a slightly different specification. The visible output is the same; the underlying robustness is not.

Card L Module 04

Confirmation · Robustness · Sensitivity

Three concepts that look similar — conflating them is why fragile findings get presented with unwarranted confidence.

ConceptWhat it asksWhat it can't tell you
ConfirmationDoes the data support this hypothesis under one specific setup — one model spec, one time window, one sample, one set of assumptions?Whether the support would persist if the setup changed. It establishes a result is possible, not robust.
RobustnessDoes the support persist when reasonable alternatives are considered? Do different specifications, samples, or windows produce broadly similar conclusions?Whether the result is causally correct. It is about stability across analytical choices, not ruling out confounders.
Sensitivity analysisHow much do conclusions change when key assumptions, parameters, or inputs vary? Which inputs are the conclusions most dependent on?Whether the variations tested cover the full space of plausible alternatives. It only tests what the analyst chooses to vary.
Reframing humility as strength

The willingness to attack one's own results feels like betraying the work already done. Card L reframes it: a result that has survived challenge is far more defensible than one that has not. The analyst who openly stress-tests their findings demonstrates more rigor, not less. The three concepts together form a layered defence against fragility — skipping any layer leaves the analysis exposed.

Card L Module 04

Model Dependence

When one specification carries the result, the conclusion is about the modelling decision — not about the world.

When conclusions rely heavily on one specific choice — a particular regression specification, time window, variable definition, or handling of missing data — the analysis becomes vulnerable in a specific way. The result holds, but only because of the choice. A reasonable alternative would produce a different result.

⚠ When a single choice carries the result

An analyst builds a regression finding a strong relationship between marketing spend and sales. The coefficient is precise, the p-value small. Before recommending a 30% spend increase, the analyst tries an alternative: defining spend by month rather than week, and adding a quarterly seasonality control. The relationship vanishes. The original finding was an artefact of the time aggregation and missing control.

Model dependence is often invisible to the analyst who created the model — the result feels stable because it has not been challenged. Challenge is exactly what reveals it. The remedy is not more sophisticated models; it is multiple specifications, checked for agreement.

Card L Module 04

The Stress-Testing Toolkit

The goal is not to optimise results — it is to find the specifications under which the answer changes.

ToolWhat it does
Alternative model specificationsRe-run the analysis with different functional forms, control sets, or model families. If the conclusion is robust, it survives reasonable alternatives.
Variable transformationsTest whether log-transforming variables, using percentages instead of levels, or different aggregations changes the conclusion. Many findings appear or disappear depending on these.
Different time windows or samplesExamine whether the result holds when restricted to a subperiod, a different subset of subscribers, or a different region. A finding that holds only in one slice is by definition fragile.
Sensitivity tables & scenario checksExamine how outputs change as key parameters vary across plausible ranges. The output is not a single number but a sensitivity surface.
Regression diagnostics & assumption togglesFormal checks of model assumptions, with deliberate exploration of what changes when assumptions are relaxed or modified.
Stress-test, don't optimise

The analyst is not looking for the specification that produces the most favourable answer — they are looking for the specifications under which the answer changes. A result that survives many such tests is genuinely strong. A result that breaks under several is fragile and should be communicated as such.

Card L Practice · Interactive

Practice · Robustness Stress-Test

An initial result is shown under five alternative specifications. Judge whether the conclusion is robust, fragile, or model-dependent.

Read the five alternative specifications, then pick your verdict.
Card L Module 04

Honesty About Fragility

Selective reporting is not a robustness check — it is a confirmation exercise dressed up as one.

⚠ What selective reporting looks like

An analyst runs ten specifications. Eight produce findings consistent with the preferred conclusion; two contradict it. The analyst presents the eight as a robustness check, omitting the two that disagreed. The analysis looks rigorous — ten specifications, eight in agreement — but the work is dishonest.

The two contradicting specifications were the most analytically informative results — they revealed the boundary of robustness. Hiding them produces a false sense of strength and exposes the organisation to risk when the conclusion later fails to hold.

Many analysts believe admitting uncertainty weakens their position. The opposite is true. The disciplined alternative: lead with the strength of the result, then disclose the boundaries — name the conditions under which it would weaken or vanish — and recommend with calibrated confidence, tying the recommendation to the strength of the evidence.

Card L Module 04

Closing the Legitimacy Ladder

Module 04 closes the tool legitimacy progression that has built across the whole programme.

THE COMPLETE TOOL LEGITIMACY LADDER CARDS A–B Tools forbidden — bias examination and problem framing only. CARD C Descriptive tools only — trends, comparisons, time-series. CARDS D–F Inferential family opens — conditional on structure and hypotheses. CARDS G–I Narrowed to the hypotheses and drivers worth modelling. CARDS J–K Modelling itself disciplined — plausibility, assumptions, conditionality. CARD L Every result stress-tested against its own fragility before it leaves the desk.
Judgment → structure → focus → rigor

Cards A–B forbade tools. Card C permitted only descriptive ones. Cards D–F opened the inferential family conditional on disciplined structure. Cards G–I narrowed the set worth modelling. Cards J–K disciplined the modelling itself. Card L closes the loop — this is what rigor looks like in practice.

Key Takeaway
Strong analysis survives challenge; weak analysis collapses under it.
Module 04 · Assessment

Knowledge Check · Part 1

Three questions across Cards J and K. Select the best answer — feedback appears immediately.

0 / 5 answered
Q1 — A stakeholder hypothesises that a bundle redesign launched in week 3 of a 13-week quarter explains a 6% quarterly revenue decline, in a product accounting for 40% of revenue. What's the single most useful first step?
A
Run a regression of revenue against bundle-related variables
B
Build a forecasting model with the bundle as a key feature
C
Run a 15-minute back-of-the-envelope check to see whether the math can even support the hypothesis
D
Begin a subscriber survey to capture bundle perception
Validation precedes verification. Before building any model, the disciplined analyst checks whether the hypothesis is mathematically capable of producing the observed effect. Here: 40% × ~77% (active weeks) × maximum-effect = at most ~30% of the decline even under generous assumptions. The hypothesis fails plausibility. A model would dress the implausibility in statistical machinery.
Q2 — Which of these statements is most accurate about analytical models?
A
Sophisticated models (random forests, neural networks) don't require assumptions — they let the data speak for itself
B
All models encode assumptions about how reality works — more complex models hide those assumptions rather than eliminating them
C
Only regression-style models have assumptions; tree and ML methods are assumption-free
D
Models reveal truth — that's their purpose
Every model is a mathematical statement of belief about the system. Random forests assume axis-aligned splits. Gradient boosting assumes errors can be corrected by additive weak learners. Neural networks assume the function lies within the class their architecture can represent. Hidden assumptions are still assumptions — just harder to audit, which makes them more dangerous, not less.
Q3 — An analyst uses linear regression to model how subscriber demand responds to tariff. The model estimates −4 units per $1 increase, with significant p-value and good fit. Then demand drops catastrophically at a $3 increase the model predicted as "limited impact." What likely went wrong?
A
The p-value was wrong — the relationship wasn't statistically significant
B
The linear assumption averaged across distinct demand regimes — fine on average, badly wrong at any specific operating point
C
Sample size was too small
D
Subscribers behaved irrationally
Linear regression assumes effects add up in straight lines and stay stable across the predictor range. Real demand is often non-linear: insensitive at low prices, sensitive at moderate prices, collapsing at high prices where substitutes appear. The model's math is correct; the model choice was wrong. Recommendations based on averaged-across-regimes coefficients can be dangerously misleading at any actual operating point.
Module 04 · Assessment

Knowledge Check · Part 2

Two questions on Card L — completing the module assessment.

0 / 5 answered
Q4 — A regression shows a strong, significant relationship between marketing spend and sales. The relationship disappears when the time aggregation is changed from weekly to monthly and a quarterly seasonality control is added. What's the right conclusion?
A
The first model is correct — the weekly aggregation captures the true relationship
B
The second model is correct — monthly aggregation is always better
C
The result is model-dependent — the conclusion was an artefact of the original specification, not a feature of the world
D
Both models are correct in different contexts; report only the favourable one
A classic case of model dependence. The original finding existed only because of the specific time aggregation and missing seasonal control. A reasonable alternative specification dissolves it. The disciplined response is to report the fragility honestly — not pick the specification that produced the preferred answer (selective reporting), and not assume one specification is "right" without further investigation.
Q5 — An analyst ran 10 specifications; 8 supported the preferred conclusion, 2 contradicted it. They present only the 8 as a "robustness check." What's wrong?
A
Nothing — 8 out of 10 is a strong robustness result
B
They should have run more specifications
C
Selective reporting — the 2 contradicting specifications were the most informative because they revealed the boundary of robustness. Hiding them is dishonest
D
The p-values from the 2 contradicting specifications must have been wrong
The contradicting specifications are the analytically informative results — they reveal the conditions under which the conclusion fails to hold. Suppressing them produces a false sense of strength and creates risk for the organisation. The disciplined alternative is full transparency: "the result holds under specifications A–H but not under I–J, meaning the conclusion is conditional on the assumption that…" This is harder to deliver and easier to defend.
About Us

Module 04 · The Rigor