Lecture 2

Data Analytics Thinking in Sports and Business

Byeong-Hak Choe

bchoe@geneseo.edu

SUNY Geneseo

August 26, 2026

🧠 Data Analytics Thinking

☕ Where should the company open its next café?

A small café company can afford one new location and must choose between two available leases.

Think, pair, share

Which evidence would you request first—and how could it change the choice?

• foot traffic by hour and day—and how many passersby might buy;
• rent, labor costs, and lease terms;
• nearby competitors and businesses that generate demand; or
• sales, margins, and seasonal patterns at comparable cafés.

A busier location may generate more sales—but not more profit.

Pacing: 3 minutes (0:00–0:03).

Give students 30 seconds to choose one piece of evidence, 60 seconds to compare with a neighbor, and take two responses. Connect each fact to the lease decision. Foot traffic suggests potential demand, but conversion, rent, labor, competition, and purchasing patterns determine whether that demand can become profit. This is a hypothetical classroom scenario.

🎯 Data analytics thinking starts with a decision

Evidence-based decision making is the goal; data analytics thinking is the method.

Weak starting point

“What patterns are in the location data?”

Missing: a decision and a goal.

Decision-ready starting point

“Which site is more likely to break even within 12 months, given demand and costs?”

Decision maker: café owner
Action: sign one lease
Outcome: break even within 12 months

Is an accurate sales forecast enough to choose a lease? Why or why not?

Pacing: 4 minutes (0:03–0:07).

The purpose is to distinguish prediction accuracy from decision usefulness. Students should recognize that a sales forecast addresses only part of the lease decision; the goal is break-even, which also depends on costs, constraints, and uncertainty. Page 6 then turns that insight into a more answerable analytical question.

🧭 Four kinds of analytical questions

Type	What it asks	Café-location example
Descriptive	What happened?	How busy were similar cafés?
Diagnostic	Why did it happen?	Why were some cafés more profitable?
Predictive	What may happen?	What sales and costs might each site have?
Prescriptive	What should we do?	Which lease should the company choose?

Quick classification: “Comparable cafés near transit had high sales but modest profits.” What does this describe, and what would you compare to investigate why?

Pacing: 4 minutes (0:07–0:11).

“High sales” and “modest profits” describe observed outcomes. Rent, labor, product margins, hours, and customer mix suggest diagnostic comparisons, but the sentence does not establish a cause. Prediction estimates what may happen at each site; prescription weighs the forecasts, costs, risks, and constraints to recommend an action.

🔍 A useful question tells us what to look for

Too broad: “Which location looks best?”

Answerable: “Which site is more likely to break even within 12 months after accounting for customer demand, rent, labor, and seasonal variation?”

• Specific: identifies the sites, outcome, and time horizon.
• Comparable: evaluates the same outcome for both sites.
• Actionable: the answer can change the lease decision.

Turn and talk

“Break even” means total revenue covers total costs. To estimate whether each site will break even, what data would you need?

Pacing: 4 minutes (0:11–0:15).

Use the hints to elicit expected revenue—customers and average purchases—plus fixed and variable costs such as rent, labor, supplies, and start-up expenses. Keep the discussion conceptual; no calculation is required. Draw out the two sites, the 12-month horizon, the common outcome, and the lease decision.

🗺️ Which location should the café choose?

Hypothetical scenario: The company can sign only one lease.

Factor	Site A: Downtown offices	Site B: Transit neighborhood
Average weekday passersby	8,000	4,800
Monthly rent	$18,000	$10,000
Nearby cafés	6	2
Traffic pattern	Strong lunch; quiet evenings and weekends	Strong mornings and evenings; steadier weekends

Decide with a partner

1. Choose Site A, Site B, or ask for more information.
2. What missing fact is most likely to reverse your choice?
3. What should decide the lease: sales, profit, or time to break even?

Pacing: 5 minutes (0:15–0:20).

Give students 30 seconds for an individual choice, two and a half minutes with a partner, and two minutes for responses. Useful missing facts include the share of passersby likely to buy, average purchase, product margin, labor and renovation costs, lease terms, and seasonality. Foot traffic is potential exposure—not customers. There is no predetermined correct site; reward explanations of what evidence could change the decision. All figures are illustrative.

🏟️ Sports Analytics Applications

⚾ Moneyball made measurement part of strategy

• The book appeared in 2003; the film adaptation appeared in 2011.
• Data can complement expertise by questioning familiar measures and testing assumptions.

Discuss

What can a scout capture that a table may miss—and what can the table reveal that memory may miss?

Pacing: 3 minutes (0:20–0:23).

Use responses to establish complementarity between domain expertise and systematic evidence.

[Sources] - Local course assets: ../lec_figs/moneyball-book.png and ../lec_figs/moneyball.png; original external image sources are not recorded in the repository. [/Sources]

🏢 Sports analytics supports team-wide decisions

Decision area	Decision-ready example question
Player performance	Which skills predict success in this role next season?
Team tactics	Which option performs best against this opponent and game state?
Health and workload	When should workload change training or recovery?
Recruitment	Who best fits the role, roster, and salary limit?
Fan and business	Who may not renew—and which outreach changes behavior?

To support a decision, clarify who will act, what action is possible, what outcome matters, what is being compared, and when the answer is needed. Also consider which mistake would be more costly.

Pacing: 3 minutes (0:23–0:26).

Briefly point out that useful questions connect evidence to an actor, action, outcome, comparison, and deadline. Error costs may differ—for example, signing a poor fit versus passing on a valuable player. Do not ask students to operationalize every row.

🗣️ Stated intent predicts behavior—but imperfectly

A major-league baseball team surveyed season-ticket holders before renewal and later observed whether they actually renewed.

Tier = a seat-location segment. Columns: Highly Likely = strong yes; Likely = probable yes; Maybe = uncertain; Probably Not = probable no; Certainly Not = strong no. Cells = the percentage who actually renewed.

Tier	Highly Likely	Likely	Maybe	Probably Not	Certainly Not
1	92	88	75	67	45
2	88	81	70	65	38
3	80	76	66	55	36
4	77	72	65	45	25
5	75	70	60	35	25

Use the table: What pattern shows that customers’ stated intentions are useful? What pattern shows that they are not guarantees?

Pacing: 2 minutes (0:26–0:28).

Give students time to scan both across survey responses and down seat-location tiers. Across a row, renewal generally falls as stated intent weakens. Down a column, actual renewal differs by seat segment even when customers report the same intent.

[Sources] - Ramesh Sharda, Dursun Delen, and Efraim Turban, Business Intelligence, Analytics, Data Science, and AI, 5th ed. (Pearson, 2024), Example 1 and Figure 1.2, p. 5. Pearson catalog: https://www.pearson.com/en-gb/subject-catalog/p/business-intelligence-analytics-data-science-and-ai-global-edition/P200000010649?view=educator. - Local photographed reference: ../refs/IMG_1228.PNG. The classroom table adapts two cells as specified for this lecture: Tier 1/“Probably Not” = 67 and Tier 3/“Maybe” = 66; the photographed source displays 69 and 68. [/Sources]

📣 Prediction is not persuasion

67%

Among Tier 1 respondents who said “Probably Not,” 67% ultimately renewed.

Prediction: Who will renew?
Persuasion: Whose outcome changes because of outreach?

A brief targeting decision

The team can contact only 1,000 accounts.

1. Which segment would you prioritize, and why?
2. What missing fact could reverse your choice?

Pacing: 4 minutes (0:28–0:32).

The table shows that stated intention is informative but imperfect. There is no uniquely correct targeting choice without account value, outreach cost, reachability, capacity, and evidence about which customers can be influenced. Keep the discussion at the decision level rather than requiring students to design an experiment.

[Sources] - Sharda, Delen, and Turban (2024), Example 1 and Figure 1.2, p. 5; local reference ../refs/IMG_1228.PNG. The 67% value is the classroom adaptation used on the preceding slide. The questions about outreach costs and persuasion evidence are course interpretation rather than the textbook’s stated recommendation. [/Sources]

🌳 A decision tree follows game conditions

• First split — Off_Pers (offensive personnel): “11” means 1 running back + 1 tight end + 3 wide receivers. The tree first groups plays by who is on the field.

Pacing: 2 minutes (0:32–0:34).

Begin with the first split. Off_Pers is short for offensive personnel—the group of skill players on the field. In this common two-digit shorthand, the first digit counts running backs and the second counts tight ends; the remaining eligible skill players are wide receivers. For example, “11” means one running back, one tight end, and three wide receivers.

Then trace the right-side branch: 540 plays → personnel 10, 11, 20, 22, or missing → third or fourth down → more than five yards. Each condition makes the situation more specific while the number of observed plays falls.

[Sources] - Sharda, Delen, and Turban (2024), Example 2 and Figure 1.4, pp. 6–7; local references ../refs/IMG_1229.PNG and ../refs/IMG_1230.PNG. Local crop: ../lec_figs/decision-tree-football.png. [/Sources]

🎲 A 95% pattern is still not a certainty

63 of 66

plays in this leaf were passes: 95.45%.

Discuss

Should the defense act as if a pass is guaranteed? What is the cost of preparing for the wrong play?

Pacing: 3 minutes (0:34–0:37).

The pattern is strong, but three plays were not passes. Ask what could change: opponent, season, game situation, strategic adaptation, and sampling variation. Discuss the costs of preparing for the wrong play.

[Sources] - Sharda, Delen, and Turban (2024), Figure 1.4, p. 7; local reference ../refs/IMG_1230.PNG. The count 63 of 66 is inferred from the displayed 95.45% pass rate. [/Sources]

🏒 Plus-minus is common—but context-sensitive

Plus-minus (PM): +1 for an on-ice even-strength or short-handed goal for; −1 for one against.

Official since 1959–60, PM is a long-standing, routinely reported NHL statistic—not a stand-alone measure of player ability.

2025–26 regular season	Team	Points	PM
Nathan MacKinnon	COL	127	+57
Nikita Kucherov	TBL	130	+43
Connor McDavid	EDM	138	+17

Discuss

The players rank differently by points and PM. Using only this table, what can you reasonably say about individual performance? What additional evidence would you request before choosing a player?

Pacing: 5 minutes (0:37–0:42).

Let students identify what the table does and does not establish before offering explanations. Useful follow-up evidence includes teammates and opponents, time on ice, deployment, game state, goaltending, shot quality, and repeated performance over time. Connect this to context and confounding without requiring formal terminology.

[Sources] - NHL.com, “Hughes’ strong play for Canucks at both ends of ice reminiscent of Orr,” noting that plus-minus became an official statistic in 1959–60: https://www.nhl.com/news/quinn-hughes-strong-play-at-both-ends-of-ice-for-vancouver-canucks. - NHL.com, “Evolution of Evans: Arrives and Thrives,” illustrating both use of and debate about plus-minus: https://www.nhl.com/kraken/news/evolution-of-evans-arrives-and-thrives-ft. - Robert B. Gramacy, Shane T. Jensen, and Matt Taddy, “Estimating Player Contribution in Hockey with Regularized Logistic Regression,” Journal of Quantitative Analysis in Sports 9(1), 2013, on the historical popularity and limitations of traditional plus-minus: https://faculty.wharton.upenn.edu/estimating-player-contribution-in-hockey-with-regularized-logistic-regression-publishedpaper-file/. - Official NHL player endpoints for 2025–26 points and plus-minus: https://api-web.nhle.com/v1/player/8477492/landing (MacKinnon), https://api-web.nhle.com/v1/player/8476453/landing (Kucherov), and https://api-web.nhle.com/v1/player/8478402/landing (McDavid). NHL season summary: https://www.nhl.com/news/connor-mcdavid-art-ross-nathan-mackinnon-rocket-richard-avalanche-jennings-trophy. [/Sources]

📈 Business Intelligence

⏱️ Business intelligence enables recurring decisions

Business intelligence (BI) combines regularly updated data, consistent definitions, analysis, and reporting so an organization can monitor performance and make recurring decisions.

Operations data → Defined metrics → MTA dashboard → Decision → Review

One-time analysis

Why did one subway line fall last month?

BI system

Each month, monitor journey time, waiting, service, and equipment by line.

Dashboard: a shared visual display of selected, regularly refreshed metrics for comparison and action—MTA Performance Metrics.

Pacing: 4 minutes (0:42–0:46).

Use the MTA example to distinguish a one-time explanation from a recurring system. The dashboard standardizes definitions and update cycles so service planners can notice exceptions, drill down, decide where to investigate, and review whether action changed performance.

[Sources] - Metropolitan Transportation Authority, “MTA Performance Metrics”: https://metrics.mta.info/ (accessed 2026-08-26). - New York City Transit, Key Performance Metrics, July 2025, pp. 8–9: https://www.mta.info/document/179606. [/Sources]

🛠️ Popular dashboard tools take different paths

Tool	Often chosen for	Example
Tableau	Interactive public data stories	Zillow Market Heat Index
Power BI	Reporting in Microsoft-centered organizations	SUNY Geneseo IR&E dashboards
Looker	Shared metrics built from governed data models	Sales and Customer Insights sample
R / Python	Code-based, reproducible dashboards	Quarto dashboard gallery

Choose for the data ecosystem, audience, update cycle, and who will maintain the dashboard—not for the logo.

Pacing: 2 minutes (0:46–0:48).

Treat these as examples of different publishing paths, not a ranking. Zillow embeds its Market Heat Index from Tableau Public. SUNY Geneseo points users to a Power BI menu of institutional dashboards. Google’s Looker quickstart builds a sample Sales and Customer Insights dashboard. Quarto’s gallery includes dashboards created with both R and Python.

[Sources] - Zillow Research, “Market Heat Index” (embedded from Tableau Public): https://www.zillow.com/research/market-heat-index-34054/. - SUNY Geneseo, “Institutional Research & Effectiveness Dashboards” (links to Power BI): https://www.geneseo.edu/ire/institutional-research-effectiveness-dashboards/. - Google Cloud, “Build a dashboard with sample data” (Looker): https://docs.cloud.google.com/looker/docs/looker-core-create-dashboard-quickstart. - Quarto, “Gallery—Dashboards”: https://quarto.org/docs/gallery/#dashboards. [/Sources]

🔗 Key performance indicator (KPI): goal to measure

Key performance indicator (KPI): a metric deliberately selected to judge progress toward a goal and guide action.

Goal

Reliable subway travel.

Metric

Customer journey time performance: the share of trips completed within five minutes of schedule.

KPI in use

Compare with a target or baseline; investigate lines that fall short.

A decision-ready KPI needs: a definition, desired direction or target, time window, comparison, owner, and linked response.

Pacing: 2 minutes (0:48–0:50).

Build the sequence aloud: goal → metric definition → target or comparison → owner → response. A measured quantity becomes a KPI only when the organization deliberately uses it to evaluate progress and guide a recurring decision.

[Sources] - Metropolitan Transportation Authority, “MTA Performance Metrics”: https://metrics.mta.info/. - New York City Transit, Key Performance Metrics, July 2025, p. 9, defining customer journey time performance as the share of trips within five minutes of schedule: https://www.mta.info/document/179606. [/Sources]

🚦 Key performance indicator (KPI) guides action

The primary KPI represents the goal; supporting metrics help explain why it moved.

Dashboard metric	What it measures	Decision use
Customer journey time performance	% of trips within 5 minutes of schedule	Track the rider outcome
Additional platform time	Average extra wait beyond schedule	Diagnose waiting problems
Service delivered	% of scheduled service operated	Diagnose cancellations or capacity
Mean distance between failures	Miles between equipment failures	Target maintenance

Discuss: Service delivered rises, but journey performance falls and extra waiting rises. Has service improved? Choose the primary KPI, one diagnostic metric, and a comparison before acting.

Pacing: 2 minutes (0:50–0:52).

There is no defensible answer without a stated goal and comparison. If the goal is rider experience, customer journey time performance is the primary outcome; service delivered and extra platform time help diagnose the pattern. Ask for a target, recent trend, line-level drill-down, incident context, and an owner who can respond.

[Sources] - Metropolitan Transportation Authority, “MTA Performance Metrics”: https://metrics.mta.info/. - New York City Transit, Key Performance Metrics, July 2025, pp. 8–9, for definitions of service delivered, additional platform time, customer journey time performance, and equipment reliability: https://www.mta.info/document/179606. [/Sources]

🧪 Classwork

Try it out → Classwork 1: NYC 311 Dashboard I

🧰 DANL Tools

🧭 Each tool has a different job

Tool	Category	Best fit
Excel	Spreadsheet	Quick calculations, charts, and sharing
SQL	Query language	Retrieve, join, and summarize database tables
R / Python	Programming languages	Analysis, visualization, modeling, and automation
RStudio / Jupyter	Working environments	Write, run, and explain code
Git / GitHub	Version control / hosting	Record history; collaborate and publish

Use a stack: choose tools for the task, team, and output.

Start by separating categories. Excel is a spreadsheet; SQL is a language for working with relational databases; R and Python are programming languages; RStudio and Jupyter are interfaces for working with code; Git records versions; and GitHub hosts repositories and supports collaboration.

[Sources] - Microsoft Support, “Overview of formulas in Excel”: https://support.microsoft.com/en-us/excel/get-started/overview-of-formulas-in-excel. - PostgreSQL Documentation, “The SQL Language”: https://www.postgresql.org/docs/current/tutorial-sql.html. - R Project, “What is R?”: https://www.r-project.org/about.html. - Python Software Foundation, “About Python”: https://www.python.org/about/. - Posit, “RStudio IDE”: https://posit.co/downloads/. - Project Jupyter: https://jupyter.org/. - GitHub Docs, “What is GitHub?”: https://docs.github.com/en/get-started/start-your-journey/what-is-github. [/Sources]

⚖️ Excel and programming support different workflows

Dimension	Excel	Programming with R / Python
Work is recorded as	Cells, formulas, and workbook steps	Code and scripts
Especially convenient for	Quick viewing, editing, calculations, and charts	Repeatable, multi-step analysis
Collaboration advantage	Familiar files for spreadsheet users	Exact reruns, reviewable logic, and version history
Move this direction when	The task is focused and hands-on	Updates, many files, automation, or extension matter

Practical rule: use Excel for quick viewing and exchange; use programming when the analysis must be rerun, checked, automated, or extended.

Avoid presenting Excel as “bad” or code as automatically superior. The key distinction is the work required: visible, hands-on interaction versus a recorded workflow that can be rerun. Excel and code often exchange the same .xlsx or .csv files.

[Sources] - Microsoft Support, “Overview of formulas in Excel”: https://support.microsoft.com/en-us/excel/get-started/overview-of-formulas-in-excel. - The workflow comparison and recommendation are course guidance. [/Sources]

🗄️ SQL retrieves the rows an analysis needs

SQL (Structured Query Language) queries tables stored in relational databases.

Ask the database

• SELECT chooses the output.
• FROM identifies the source table.
• WHERE filters rows.
• GROUP BY creates groups.
• JOIN connects tables.

Example: requests by borough

SELECT borough,
       COUNT(*) AS requests
FROM nyc311
WHERE complaint_type = 'Noise'
GROUP BY borough;

Result: one row per borough with a request count.

Where it fits: SQL retrieves and summarizes database data; Excel inspects it; R or Python analyze, visualize, and automate.

SQL is a language, not a database product. PostgreSQL, MySQL, SQLite, and SQL Server are database systems that support SQL dialects. Walk through the example in its logical pieces: choose output, choose the table, filter input rows, then aggregate by borough. The database returns a smaller result that can feed Excel, R, Python, or a dashboard.

[Sources] - PostgreSQL Documentation, “Querying a Table,” for SELECT, FROM, and WHERE: https://www.postgresql.org/docs/current/tutorial-select.html. - PostgreSQL Documentation, “Aggregate Functions,” for COUNT with GROUP BY: https://www.postgresql.org/docs/current/tutorial-agg.html. - PostgreSQL Documentation, “Joins Between Tables”: https://www.postgresql.org/docs/current/tutorial-join.html. [/Sources]

💻 R and Python have different strengths

Dimension	R	Python
Design center	Statistical computing and graphics	General-purpose programming
Especially natural for	Statistical modeling, visualization, and analytical reports	Automation, machine learning, and software or data products
Common data tools	tidyverse, ggplot2, Quarto	pandas, NumPy, scikit-learn, Jupyter
Package ecosystem	CRAN	PyPI

Think emphasis—not limits: both languages can collect, clean, model, visualize, and support machine-learning workflows.

The IBM video contrasts R’s statistical and reporting orientation with Python’s general-purpose and production orientation, but its final recommendation is that many organizations use both. Treat the comparison as a difference in emphasis, not a boundary between capabilities.

[Sources] - IBM Technology, “R vs Python,” YouTube: https://www.youtube.com/watch?v=4lcwTGA7MZw. - Local auto-generated transcript: ../refs/[English (auto-generated)] IBM Youtube - R vs Python.txt. - R Project, “What is R?”: https://www.r-project.org/about.html. - Python Software Foundation, “About Python”: https://www.python.org/about/. [/Sources]

▶️ R vs. Python: IBM overview

Ask students to listen for the distinction between a language’s design emphasis and the many tasks it is still capable of doing. The durable takeaway is to choose based on use case and team context—not to treat either language as universally better.

[Sources] - IBM Technology, “R vs Python,” YouTube: https://www.youtube.com/watch?v=4lcwTGA7MZw. - Local auto-generated transcript: ../refs/[English (auto-generated)] IBM Youtube - R vs Python.txt. [/Sources]

🎯 Choose for the task, team, and destination

• Task: Is the center of the work statistical analysis and reporting, or automation and application development?
• Team: Which language, packages, and conventions do collaborators already use?
• Destination: Will the result become a report, notebook, dashboard, model, or production system?

Our course path

DANL 101 begins with R because it keeps statistical reasoning, visualization, and reporting close together. The underlying analytical concepts transfer to Python.

There is no universally best language. The IBM video’s most durable rule is to use the language that fits the use case and the surrounding team. Many analysts eventually use both.

[Sources] - IBM Technology, “R vs Python,” YouTube: https://www.youtube.com/watch?v=4lcwTGA7MZw. - Local auto-generated transcript: ../refs/[English (auto-generated)] IBM Youtube - R vs Python.txt. [/Sources]

🧩 Code needs a working environment

RStudio + Posit Cloud

• RStudio is an integrated development environment (IDE) with an editor, console, plots, files, help, and debugging tools.
• Posit Cloud runs RStudio projects in a browser, so our class can use R without local setup.

Jupyter + Google Colab

• Jupyter notebooks combine executable code, narrative text, equations, and rich output in one document.
• Jupyter supports many languages, including Python and R.
• Google Colab is a hosted Jupyter Notebook service that requires no setup.

Language ≠ working environment: R can run outside RStudio, and Python can run outside Jupyter.

Correct a common category mistake: Jupyter is a notebook ecosystem and interface, not a programming language. JupyterLab can function as a web-based IDE; the notebook document is especially useful for combining code with explanation and output.

[Sources] - Posit, “RStudio IDE”: https://posit.co/downloads/. - Posit Docs, “Cloud products”: https://docs.posit.co/other/rstudio-cloud/. - Project Jupyter: https://jupyter.org/. - Google Colab FAQ: https://research.google.com/colaboratory/faq.html. [/Sources]

🗂️ Git records history; GitHub shares repositories

Working files → Git commit → Local repository → Push → GitHub repository

Git

• A version control system that tracks changes to files.
• Lets you compare versions, recover earlier work, and combine changes safely.
• Works locally on your computer.

GitHub

• An online platform that hosts Git repositories.
• Adds sharing, collaboration, review, issue tracking, and website publishing.
• Connects a local repository to a remote copy.

Keep the distinction clear: Git tracks versions; GitHub hosts and connects repositories.

Git is not a programming language, and GitHub is not the same thing as Git. A Git repository can exist entirely on one computer; GitHub becomes useful when the project needs a remote home for sharing, collaboration, or publishing.

[Sources] - Git, “About Version Control”: https://git-scm.com/book/en/v2/Getting-Started-About-Version-Control. - GitHub Docs, “What is GitHub?”: https://docs.github.com/en/get-started/start-your-journey/what-is-github. [/Sources]

▶️ What is GitHub?

Use the video to reinforce the Git-versus-GitHub distinction from the previous slide.

[Sources] - GitHub, “What is GitHub?”, YouTube: https://www.youtube.com/watch?v=pBy1zgt0XPc. [/Sources]

🔁 A reproducible workflow connects the tools

1. Retrieve and reshape database tables with SQL.
2. Inspect or exchange tabular data in Excel.
3. Analyze and communicate with R or Python in RStudio or Jupyter.
4. Save and commit meaningful project versions with Git.
5. Share or publish the repository through GitHub.

The goal

You do not need every tool for every task. Use enough structure to make the analysis understandable, repeatable, and shareable.

Close by connecting the tools into one workflow. The course goal is not tool collecting; it is producing analysis whose reasoning and steps can be understood and repeated.