Part 0: Foundations — Analytics for Decision Making

The big idea

This course is not mainly about formulas, software, or memorizing procedures. It is about learning how to use data to make better business decisions.

A useful model is valuable only if you can: - connect it to a real business question, - interpret what it means, - understand its limitations, and - communicate the result clearly enough for action.

0.1 What This Course Is About

In many analytics classes, students spend a great deal of time learning how to calculate things. That matters, but in business settings the more important question is often:

What does this result mean, and what should we do with it?

That is the focus of this course.

You will learn how to use ideas such as correlation, simple linear regression, multiple linear regression, and variable selection to support decision-making. You will also learn when these tools are helpful, when they are misleading, and how to avoid overclaiming what the data actually shows.

By the end of the course, you should be able to:

translate a business problem into an analytical question,
use output from JMP to identify important patterns,
explain model results in plain business language,
recognize common errors in interpretation,
and make recommendations with appropriate confidence and caution.

A practical mindset

In this class, success means more than getting the “right” number. It means being able to say:

what the analysis found,
how strong the evidence is,
what assumptions are being made,
and what risks remain before acting.

0.2 A Simple Framework for Analytics in Decision Making

Most of the work in this course can be organized into the same basic sequence.

flowchart LR
    A[Business Question] --> B[Data]
    B --> C[Model]
    C --> D[Insight]
    D --> E[Decision]
    E --> F[Risk and Limitations]

Step 1: Start with the business question

A model should begin with a real decision problem.

Examples: - Should we increase advertising spend? - Which customer traits are associated with higher spending? - Which factors best explain employee performance? - How strongly does price affect demand?

A model without a business question is usually just an exercise.

Step 2: Understand the data

Before modeling, ask: - What variables do we have? - How were they measured? - Are important variables missing? - Is the data representative of the decision we want to make?

For example, if we want to study sales, we may have: - advertising spend, - price, - seasonality, - region, - promotion indicators, - and maybe competitor behavior.

Step 3: Use a model

The model is a tool for summarizing patterns in the data.

In this course, the main tools are: - correlation, - simple linear regression, - multiple linear regression, - and stepwise variable selection.

Step 4: Extract insight

The model may suggest things like: - sales tend to rise when advertising rises, - price appears negatively related to demand, - or one variable matters much less after controlling for others.

Step 5: Make a decision

This is where analytics becomes business.

A decision might be: - increase ad spending, - change pricing, - prioritize certain customer segments, - or investigate a relationship further before acting.

Step 6: Evaluate risk and limitations

This final step is where many mistakes happen.

Ask: - Are we confusing association with causation? - Are we missing important variables? - Is the pattern strong enough to matter? - Would this result likely hold in new data? - What could go wrong if we act on this?

Important

A statistically clean model can still lead to a poor business decision if it is interpreted badly.

0.3 What Students in This Course Should Learn to Do

This course is designed for students who have already seen some basic statistics and are ready to move from calculation toward interpretation.

You are not expected to become a programmer in this class. You will mainly use JMP for analysis. However, you should become comfortable reading outputs, understanding what the software is doing conceptually, and explaining results clearly.

By the end of the course, you should be able to answer questions such as:

Is there evidence of a meaningful relationship between two variables?
How should I interpret the slope in a regression model?
What changes when I add more variables to the model?
Why can a variable look important by itself but not in a multiple regression?
Why is a “significant” relationship not automatically causal?
How do I compare models in a way that supports a real decision?

These are the kinds of questions managers, analysts, consultants, and decision-makers face all the time.

0.4 Correlation, Regression, and Decision Making

At a high level, this course moves through three connected ideas:

Concept	Main Purpose	Typical Business Question
Correlation	Measure strength and direction of a relationship	Do these two variables move together?
Simple Linear Regression	Model one predictor and one outcome	How much does sales change as advertising changes?
Multiple Linear Regression	Model several predictors at once	Which factors matter after controlling for others?
Variable Selection	Build a more useful model	Which predictors should we keep in the model?

The progression matters.

Students often want to jump directly to a full regression model, but that can create confusion. A better learning path is:

See the relationship
Measure the relationship
Model the relationship
Interpret the relationship
Question the relationship

That final step—questioning the relationship—is one of the most important skills in this course.

0.5 Correlation Is Useful, but It Is Not Enough

One of the biggest themes in analytics is that variables can move together for many reasons. Some of those reasons are meaningful. Some are accidental. Some are driven by a third factor that we do not see clearly.

Example: Ice cream sales and drowning incidents

These two variables may be positively correlated.

Does ice cream cause drowning? No.

A more likely explanation is that both increase in warmer weather. The hidden driver is temperature or season.

This example may sound simple, but the same mistake appears in business constantly.

Business examples of the same problem

Stores with higher staffing may also have higher sales.
But is staffing causing sales, or do busy stores simply need more staff?
Employees who attend more training may perform better.
But is training causing better performance, or are stronger employees more likely to be selected for training?
Customers receiving more promotional emails may purchase more.
But is email driving purchases, or are the most engaged customers being targeted more heavily?

Core principle

A relationship in the data is a starting point for investigation, not automatic proof of a causal effect.

0.6 Why Causality Requires Caution

Many business decisions depend on causal language:

Will increasing price reduce demand?
Will more advertising increase sales?
Will training improve productivity?
Will a change in service quality reduce churn?

These are not merely questions about association. They are questions about cause and effect.

Regression can help us study these relationships, but it does not automatically prove causality.

Why not?

Because an observed relationship may be influenced by:

omitted variables,
reverse causality,
selection effects,
measurement problems,
or simple coincidence.

A practical way to speak more carefully

Instead of saying:

“Advertising caused sales to increase.”

say:

“Advertising is positively associated with sales in this dataset, though other explanations may also be possible.”

That kind of wording is more disciplined, more honest, and usually more credible.

0.7 What Makes a Strong Analyst

A strong analyst does more than run software. A strong analyst asks better questions.

Weak analyst mindset

“The software says it is significant, so it must matter.”
“The coefficient is positive, so we should invest more.”
“The model fit is high, so the conclusion is proven.”

Strong analyst mindset

“What exactly does this coefficient mean?”
“How large is the effect in practical terms?”
“Could another variable explain this?”
“Would this relationship likely hold outside this sample?”
“What decision would I actually make based on this result?”

A useful habit

Whenever you see a result, ask both of these questions:

What does this suggest?
What else could explain it?

0.8 How JMP Fits into the Course

JMP will be your primary tool for building models and exploring data.

You will use it to: - create visual displays, - estimate regressions, - compare models, - and interpret output.

The goal is not to memorize every menu path. The goal is to understand what the output is telling you.

That means you should be able to interpret things such as: - direction of a relationship, - strength of a relationship, - slope coefficients, - model fit measures, - statistical significance, - and warning signs that a model may be misleading.

When visuals are especially helpful, we will use them heavily. A strong graph often teaches a concept more clearly than a page of text.

0.9 Mini Case: Should the Company Increase Advertising?

Scenario

A company notices that months with higher advertising spend also tend to have higher sales. The estimated correlation between advertising and sales is 0.75, which suggests a strong positive relationship.

At first glance, this sounds like a clear argument for increasing advertising. But before acting, a good analyst should slow down and ask whether the relationship is as simple as it appears.

Questions to consider

Are high-sales months also holiday months?
Are promotions happening at the same time as ad increases?
Are some regions receiving more advertising because they are already stronger markets?
Could sales expectations be driving ad budgets rather than the other way around?

Why this case matters

This example captures the main theme of the course:

A pattern can be real and still be misunderstood.

Correlation gives us a reason to look closer. Regression helps us look more carefully. Good judgment helps us decide what to do.

0.10 How This Course Will Build Your Understanding

Because this is a 5-week online course with weekly meetings, the material should be clear enough for self-study but still deep enough to support real learning.

The chapters will generally follow this rhythm:

Business motivation
Why this topic matters in practice
Conceptual explanation
What the idea means in plain language
Visual intuition
Graphs or diagrams to make the idea easier to see
JMP interpretation
What to look for in the software output
Common mistakes
Where students often get confused
Exercises with answers
So students can check understanding
End-of-section case
A short applied business scenario

This structure is meant to support both exam preparation and case-based decision-making.

0.11 Check Your Understanding

Questions

What is the difference between a model and a decision?
Why is correlation not enough, by itself, to justify action?
Why can a strong relationship still lead to a bad decision?
In the advertising case, what are two plausible missing variables that could influence both advertising and sales?
Why is it useful to speak carefully about causality when presenting results?

0.12 Common Pitfalls We Will Revisit Repeatedly

The same interpretation mistakes appear again and again in analytics. This course will return to these throughout the semester.

Pitfall	Why It Matters
Correlation is mistaken for causation	Can lead to false business conclusions
Important variables are omitted	Can distort estimated relationships
Coefficients are interpreted too literally	Can create bad recommendations
Statistical significance is overemphasized	Small but significant effects may have little business value
Models are trusted too quickly	Good fit does not guarantee good decisions

A major goal of this course is to help you slow down enough to avoid these traps.

0.13 Key Takeaways

What to remember from Part 0

Analytics should begin with a business question
Models are tools for understanding patterns, not automatic proof of truth
Decision-making requires both analysis and judgment
Correlation can be informative, but it does not prove causality
Good analysts challenge their own conclusions before recommending action

0.14 Looking Ahead

In the next section, we begin with correlation, the simplest formal tool for studying whether two variables move together.

That topic may sound basic, but it lays the groundwork for everything that follows: - regression, - interpretation, - model building, - and disciplined thinking about causality.