Interpretation and Communication | Malone Math

The Central Idea

A numerical answer is not the same as an interpreted result.

Suppose a model gives

\[ q=625. \]

By itself, $625$ tells us very little.

A complete interpretation might say:

Under the stated cost and price assumptions, the model predicts a break-even production quantity of approximately $625$ units.

The mathematics produces a quantity. Interpretation explains what that quantity means in context.

Return to the Original Question

Every model begins with a real question.

After solving or simulating the model, return to that question.

Ask:

What quantity did we actually calculate?
What units does it have?
What time period or location does it refer to?
What assumptions were required?
Is the result inside the intended domain?
How uncertain is it?
What decision or conclusion does it support?

Always Include Units

Units are part of the meaning of a result.

If a calculation gives

\[ v=60, \]

we still need to know whether this means

$60$ miles per hour;
$60$ kilometers per hour;
$60$ meters per second;
something else entirely.

A unitless answer can be mathematically correct and practically unusable.

Interpret Parameters Too

Model parameters often have practical meanings.

\[ C(q)=500+8q, \]

the intercept $500$ can be interpreted as fixed cost, while the slope $8$ represents an additional $\$8$ of cost per unit produced.

Parameter interpretation makes the structure of the model understandable.

Interpret Slopes as Rates

In a linear model

\[ y=mx+b, \]

the slope

\[ m=\frac{\Delta y}{\Delta x} \]

is a rate of change.

If $y$ is dollars and $x$ is units produced, then $m$ has units

\[ \frac{\text{dollars}}{\text{unit}}. \]

The slope should be explained using both magnitude and units.

Interpret Intercepts Carefully

An intercept often represents the predicted output when the input equals zero.

\[ y=mx+b, \]

we have

\[ y=b \]

when

\[ x=0. \]

But this interpretation is meaningful only if $x=0$ belongs to the modeling domain.

Example: An Intercept Outside the Domain

Suppose a linear model relates adult height to age using data only for adults aged $20$ to $60$.

The algebraic intercept at age $0$ may exist, but interpreting it as the predicted height of a newborn would be inappropriate.

The intercept is mathematically defined but outside the intended modeling domain.

Interpret Signs

The sign of a quantity often carries meaning.

\[ \frac{dP}{dt}>0, \]

then $P$ is increasing at that moment.

\[ \frac{dP}{dt}<0, \]

then $P$ is decreasing.

In applied work, signs should be translated into the language of the modeled system.

Interpret Magnitude

Mathematical magnitude and practical magnitude are not always the same.

A change of

\[ 0.001 \]

may be negligible in one system and critical in another.

Interpretation requires scale.

Compare a result with:

typical values;
historical ranges;
decision thresholds;
measurement precision;
operational tolerances.

Statistical Significance versus Practical Significance

A statistically detectable effect is not automatically important in practice.

With a large enough sample, a very small difference may be statistically significant.

Model communication should therefore distinguish:

whether evidence suggests an effect exists;
whether the effect is large enough to matter.

Interpret Results Within the Domain

Suppose a model was developed for

\[ 0\leq x\leq 50. \]

A prediction at

\[ x=25 \]

is an interpolation within the stated domain.

A prediction at

\[ x=500 \]

is far outside it.

Communication should make this distinction explicit.

Interpolation and Extrapolation

Interpolation estimates within the range of observed or supported conditions.

Extrapolation extends beyond that range.

Extrapolation depends more heavily on assumptions because there is less direct evidence supporting the model there.

Extrapolated results should usually be communicated with greater caution.

State the Assumptions That Matter

A result is conditional on the assumptions used to produce it.

Suppose a forecast assumes:

constant unit cost;
constant demand growth;
no capacity expansion;
no major policy change.

Those assumptions are part of the interpretation.

A forecast should not be presented as though it were independent of them.

Use Conditional Language

Model conclusions are often best expressed conditionally.

Instead of saying:

Sales will reach 12,000 units.

a more accurate statement may be:

Under the model's current growth assumptions, projected sales are about 12,000 units.

Conditional language preserves the distinction between a model projection and a guaranteed outcome.

Prediction Is Not Certainty

A model prediction is an inference from assumptions, data, and structure.

It is not a promise that the future will match the model.

This matters especially for:

economic forecasts;
weather forecasts;
population projections;
risk estimates;
financial models;
demand forecasts.

Communicate Uncertainty

If important inputs or parameters are uncertain, the output should not be presented with false precision.

Instead of reporting only

\[ \hat y=742.6381, \]

it may be more useful to report

\[ \hat y\approx 743 \]

together with an uncertainty range.

Point Estimates and Intervals

A point estimate gives one central value.

An interval communicates a range of plausible values.

For example,

\[ \hat y=100 \]

may be accompanied by

\[ 90\leq y\leq 115 \]

under a particular uncertainty analysis.

The interval often communicates more than the point estimate alone.

Be Precise About What an Interval Means

Not all intervals have the same interpretation.

A range from scenario analysis is not automatically a confidence interval.

A confidence interval is not the same as a prediction interval.

A minimum and maximum from a simulation are not automatically hard physical bounds.

State how the interval was constructed.

Precision Should Match Evidence

Reporting many decimal places can imply more certainty than the data or model support.

If input measurements are only accurate to the nearest unit, reporting an output to six decimal places may be misleading.

Numerical precision should reflect:

measurement precision;
parameter uncertainty;
model accuracy;
decision needs.

Round at the End

During calculation, retain enough digits to avoid unnecessary rounding error.

Round the final reported result to a level appropriate for the application.

Premature rounding can distort a final answer, while excessive displayed precision can distort its interpretation.

Use Percentages Carefully

Percentage change and percentage-point change are different.

If a rate increases from $20\%$ to $25\%$, the increase is

\[ 5 \]

percentage points.

The relative percentage increase is

\[ \frac{25-20}{20}\times 100\%=25\%. \]

Communication should distinguish these quantities.

Absolute and Relative Change

Suppose revenue rises from $\$1{,}000{,}000$ to $\$1{,}100{,}000$.

The absolute increase is

\[ \$100{,}000. \]

The relative increase is

\[ \frac{100{,}000}{1{,}000{,}000}=0.10=10\%. \]

Both may be useful, depending on the audience and decision.

Communicate Comparisons with a Baseline

Model outputs are often easier to understand when compared with a baseline.

Instead of saying

\[ C=5.4\text{ million dollars}, \]

we might say:

The modeled cost is $\$5.4$ million, approximately $8\%$ above the baseline scenario.

Baseline comparisons give scale and context.

Communicate Sensitivity

If a conclusion changes substantially when a parameter changes slightly, that fragility should be visible.

For example:

The project is profitable in the baseline case, but becomes unprofitable if unit demand falls by more than about 6%.

This communicates more decision-relevant information than the baseline profit alone.

Communicate Robustness

If a conclusion remains stable across plausible scenarios, say so.

For example:

Strategy A remains the lowest-cost option across all tested demand and fuel price scenarios.

Robustness is an important property of a decision recommendation.

Distinguish Description from Recommendation

A model may describe what is likely to happen without deciding what should be done.

For example:

The model predicts that inventory will fall below $500$ units in three weeks.

That is descriptive.

Therefore, increase the reorder quantity next week.

That is a recommendation.

Recommendations require objectives, constraints, and value judgments in addition to predictions.

Distinguish Association from Causation

If a model identifies a relationship between $x$ and $y$, that does not automatically mean changing $x$ will cause $y$ to change.

A predictive regression such as

\[ \hat y=\beta_0+\beta_1x \]

may capture association without establishing a causal mechanism.

Causal language should be used only when the model and evidence justify it.

Explain the Model Before the Output

A useful communication sequence is:

state the question;
identify the principal quantities;
briefly explain the model structure;
state the major assumptions;
present the result;
interpret its practical meaning;
state uncertainty and limitations.

This gives the audience enough context to understand what the result actually represents.

Use Equations Where They Add Clarity

Equations are powerful when the audience needs to see the mechanism.

For example,

\[ P=(p-c)q-F \]

makes the profit mechanism compact and explicit.

But equations should support explanation, not replace it.

Translate Equations into Words

After presenting

\[ P=(p-c)q-F, \]

explain:

Profit equals the contribution earned on each unit, multiplied by the number of units sold, minus fixed cost.

This translation connects mathematical structure to mechanism.

Tables

Tables are useful when exact values matter or when several scenarios must be compared.

A table might organize:

baseline, optimistic, and pessimistic scenarios;
parameter values;
predicted outcomes;
errors;
decision thresholds.

Tables should be labeled clearly and should not contain unnecessary digits.

Graphs

Graphs are useful for showing:

trends;
nonlinear relationships;
uncertainty bands;
thresholds;
comparisons;
time evolution;
sensitivity.

A good graph communicates structure that may be difficult to see in a table of numbers.

Graph Axes Need Meaning

Every graph should make clear:

what each axis represents;
the units;
the relevant scale;
whether the axis begins at zero;
whether any transformation has been applied.

Axis choices can change how dramatic a pattern appears.

Do Not Use Visual Scale to Mislead

Suppose values rise from $98$ to $100$.

A vertical axis from $97.5$ to $100.5$ may make the increase appear visually enormous.

A truncated axis is not automatically wrong, but it should be chosen and labeled so that the display supports accurate interpretation rather than exaggeration.

Show Uncertainty Visually When Useful

A forecast line can be accompanied by a shaded or bounded uncertainty region.

This helps communicate that the central forecast is not the only plausible outcome.

The legend or accompanying text should explain what the band represents.

Match the Communication to the Audience

Different audiences need different levels of detail.

A technical audience may need:

equations;
parameter estimates;
error metrics;
validation design;
uncertainty assumptions.

A decision-maker may need:

the main result;
important assumptions;
decision thresholds;
risks;
recommended action.

Do Not Hide Important Detail

Simplifying communication does not mean removing material facts.

If a conclusion depends critically on one uncertain assumption, that dependency belongs in the summary even for a nontechnical audience.

Good communication is concise without being misleading.

Lead with the Decision-Relevant Result

In applied settings, audiences often need the principal conclusion first.

For example:

Under the baseline assumptions, the project breaks even at approximately 625 units. The result is most sensitive to unit margin and expected demand.

Supporting derivations can follow.

Separate Facts, Model Outputs, and Judgments

These categories should not be blended.

For example:

Observed fact: last year's demand was 9,800 units.
Model output: next year's modeled demand is 10,600 units.
Judgment: capacity should be increased before next year.

Clear separation improves accountability and interpretation.

Communicate Model Error

If validation found an RMSE of

\[ 12 \]

units, that information may be important when presenting forecasts.

A prediction of $100$ should not be communicated as though historical errors were negligible.

Model performance should accompany model output when it affects the decision.

Communicate Model Comparison

When several models are plausible, explain why one was selected.

Useful comparison criteria include:

validation performance;
simplicity;
interpretability;
mechanistic plausibility;
computational cost;
robustness;
data requirements.

Example: Cost Forecast

Suppose the calibrated model is

\[ C(q)=800+6q. \]

\[ q=500, \]

predicted cost is

\[ C(500)=800+6(500)=3800. \]

A complete interpretation is:

At a production level of $500$ units, the model predicts total cost of approximately $\$3{,}800$, assuming the calibrated fixed and per-unit costs remain applicable at that production level.

Example: Growth Forecast

Suppose

\[ P(t)=1000e^{0.03t}. \]

At $t=10$,

\[ P(10)=1000e^{0.3}\approx 1350. \]

An appropriate interpretation is:

Under a constant continuous growth-rate assumption of $3\%$ per year, the model projects a population of about $1{,}350$ after ten years.

The constant-growth assumption should remain visible in the conclusion.

Example: Sensitivity Communication

Suppose baseline profit is

\[ \$80{,}000. \]

If a $10\%$ decrease in demand reduces modeled profit to

\[ \$20{,}000, \]

then the conclusion should communicate that sensitivity.

Saying only “expected profit is $\$80{,}000$” hides a material risk.

Example: Uncertainty Communication

Suppose simulation gives median profit

\[ \$75{,}000 \]

with $90\%$ of simulated outcomes between

\[ -\$10{,}000 \]

and

\[ \$140{,}000. \]

The range is essential to interpretation because the central estimate alone hides the possibility of loss.

Common Error: Reporting a Bare Number

An answer such as

\[ 42.7 \]

is incomplete unless the reader knows:

what the quantity represents;
its units;
the relevant time or location;
the assumptions behind it.

Common Error: False Precision

Suppose uncertain inputs produce a forecast around

\[ 1.2\text{ million}. \]

Reporting

\[ 1{,}203{,}847.291 \]

may imply a level of precision the model cannot support.

Precision should reflect evidence, not calculator output.

Common Error: Hiding Assumptions

A forecast based on constant growth should not be presented without mentioning that assumption when it materially affects the result.

Hidden assumptions make model outputs appear more universal than they are.

Common Error: Treating a Scenario as a Probability

An optimistic scenario and a pessimistic scenario do not automatically have known probabilities.

Unless probabilities were justified and assigned, do not say that one scenario has a particular chance of occurring.

Common Error: Confusing Correlation with Causation

If two quantities move together, that alone does not establish that one causes the other.

Model language should not become stronger than the study design or evidence supports.

Common Error: Ignoring Negative Results

If validation shows poor model performance, that belongs in the communication.

Reporting only favorable cases can create a distorted picture of model reliability.

Common Error: Overselling the Model

A model is a simplified representation, not reality itself.

Avoid claims such as:

The model proves this will happen.

unless the context genuinely supports a mathematical proof about the model rather than an uncertain claim about the real world.

A Reliable Interpretation Procedure

Identify the calculated or simulated quantity.
Attach the correct units.
Translate the result into the language of the original problem.
Check that the result lies inside the intended domain.
Interpret important parameter values and signs.
Compare the magnitude with meaningful baselines or thresholds.
State the assumptions needed for the interpretation.
Include uncertainty or sensitivity information when material.
Distinguish model output from recommendation.
Avoid causal claims unless the model and evidence justify them.

A Reliable Communication Procedure

Lead with the question and decision-relevant result.
Explain the model structure briefly.
State the most important assumptions.
Report outputs with appropriate units and precision.
Include ranges, scenarios, or probabilities when uncertainty matters.
Show sensitivity or thresholds that could change the conclusion.
Use tables or graphs when they improve clarity.
Separate observed facts, model outputs, and judgments.
State model limitations and validation evidence.
Match technical detail to the audience without hiding material facts.

Practice 1: Add Meaning

A model returns

\[ t=4.6. \]

What additional information is needed before this can be interpreted properly?

Practice 2: Interpret a Slope

A cost model is

\[ C(q)=300+12q. \]

Interpret the slope.

Practice 3: Interpret an Intercept

Using

\[ C(q)=300+12q, \]

interpret the intercept.

Practice 4: Percentage Points

A rate rises from $30\%$ to $36\%$.

Find the increase in percentage points and the relative percentage increase.

Practice 5: Conditional Language

Rewrite the claim

Revenue will be $2 million next year.

so that it is clearly presented as a model projection.

Practice 6: Extrapolation

A model was validated for values from $10$ to $50$.

What should be communicated about a prediction at $x=400$?

Practice 7: Point Estimate versus Range

A forecast has a central estimate of $500$ units but plausible outcomes range from $350$ to $700$.

Why is reporting only $500$ incomplete?

Practice 8: Association or Causation?

A regression finds that higher advertising spending is associated with higher sales.

Can the regression result alone establish that increasing advertising will cause sales to rise?

Practice 9: Practical Significance

A very large study detects an average improvement of $0.01$ units.

What additional question should be asked before calling the result important?

Practice 10: Model versus Recommendation

A model predicts that warehouse inventory will fall below the target level next month.

Is the prediction itself the same thing as a recommendation to place an order?

Practice Answers

Practice 1

We need to know what $t$ represents, its units, the relevant context, and the assumptions or domain under which the value was calculated.

Practice 2

The slope $12$ means modeled total cost increases by $\$12$ for each additional unit produced.

Practice 3

The intercept $300$ represents modeled cost when

\[ q=0. \]

In this context, it can be interpreted as fixed cost if zero production is inside the intended domain.

Practice 4

The increase is

\[ 36\%-30\%=6 \]

percentage points.

The relative percentage increase is

\[ \frac{36-30}{30}\times100\%=20\%. \]

Practice 5

One appropriate version is:

Under the model's current assumptions, projected revenue for next year is approximately $\$2$ million.

Practice 6

The prediction is a large extrapolation beyond the validated range. Reliability at $x=400$ has not been established by the available validation evidence.

Practice 7

Reporting only $500$ hides the substantial uncertainty around the forecast. The range from $350$ to $700$ may materially affect planning and should be communicated.

Practice 8

No. The regression establishes an association under the model but does not, by itself, establish that increasing advertising causes higher sales.

Practice 9

Ask whether an improvement of $0.01$ units is practically meaningful relative to the scale, cost, measurement precision, and decision context.

Practice 10

No. The prediction describes a modeled future condition. A recommendation to order inventory also requires objectives, costs, constraints, and decision criteria.

What to Remember

A model result must be translated back into the language of the original problem.
Units are part of the answer.
Slopes, intercepts, signs, and magnitudes should be interpreted in context.
Interpretations must respect the model's domain.
Extrapolation should be communicated more cautiously than interpolation.
Model conclusions are conditional on assumptions.
Predictions are not guarantees.
Precision should match the quality of the data and model.
Uncertainty, sensitivity, and validation results belong in model communication when they affect decisions.
Point estimates often need ranges or scenarios for proper interpretation.
Observed facts, model outputs, and recommendations should be kept distinct.
Association should not be presented as causation without appropriate justification.
Graphs and tables should clarify rather than exaggerate.
Communication should be adapted to the audience without hiding material limitations.

The central habit is: say what the model result means, under what conditions it means it, and how much confidence the evidence justifies.

Where We Go Next

Interpretation and communication make model results usable, but every mathematical model remains a simplified representation of a larger reality.

The final lesson in this section, Model Limitations and Revision, examines where models fail, how those failures are diagnosed, and how model structure, assumptions, parameters, domains, and data can be revised as new evidence becomes available.