Introduction to Sports Analytics - A Beginner’s Guide

Clip a stopwatch to your clipboard, open a Google Sheet, and log every pass, turnover, and shot location for one half of any match. After 45 minutes you will have 120-150 rows of raw data; run =COUNTIF(C:C,"shot")/COUNTIF(C:C,"pass") and you already have a shot-to-pass ratio that separates 65 % possession teams from 45 % possession teams in the English Championship. No subscriptions, no Python-just a ratio above 0.35 usually flags a top-six side.

Move the sheet into BigQuery (free tier: 1 TB queries per month) and join it to the open FIFA 23 player attributes CSV. A single SQL line-SELECT AVG(pace), AVG(passing) FROM squad WHERE minutes > 700 GROUP BY position-shows centre-backs average 48 pace and 66 passing; if your logged centre-back scores below 60 passing after 10 games, replace him: historical data say teams concede 0.18 goals per match more for every point lost in that metric.

Build a scatter plot in Tableau Public: x-axis = minutes played, y-axis = progressive passes per 90. Set filters to isolate full-backs; a vertical reference line at 2700 minutes splits starters from backups. Every dot above 8.5 progressive passes and right of the line correlates with clubs that gain 0.41 extra points per fixture-sell any full-back below that threshold while market value peaks at age 24.

Export Wyscout’s free 900-row sample, run a logistic regression in R with glm(goal ~ xG + distance + angle, family=binomial); the coefficient on xG is 2.34. Multiply any future chance’s xG by that factor, add the intercept, and you predict conversion within ±3 % accuracy-good enough to decide in real time whether to substitute a striker or switch wingers.

Pick One Metric and Track It for 30 Days

Choose effective Field-Goal percentage (eFG%) and log every shot your rec-league squad takes for the next month; it needs only three columns-player, attempt location, make/miss-and a free calculator (basketball-reference.com/about/faq.html) spits out the number in two clicks.

Shooting 28-for-63 inside the arc but 11-for-27 beyond it yields eFG% = 100 + (11 × 0.5) = 105.5 points on 90 tries → 0.586; a mid-range-heavy rival who hits 35-for-70 posts 0.500. That 8.6 % gap equals roughly +5 points per 100 possessions, the margin between 12th and 4th in most adult leagues.

Week	Attempts	Makes	3PA	eFG%
1	88	37	21	46.6
2	92	42	24	50.0
3	90	45	27	53.3
4	94	48	30	55.3

Ignore everything else until the calendar flips; coaches who chase ten numbers at once abandon the log by day nine.

Google Sheets template: column A = date, B = half (1 or 2), C = shooter jersey, D = x-coordinate from left hash, E = y-coordinate from baseline, F = 3? (1/0), G = result (1/0). Conditional format column G green/red; your eyes spot cold stretches instantly.

After 30 days, run a pivot: filter sub-25 % shooters on >15 tries, move them to cutter duty, run extra 25 spot-up reps each practice; squads copying that tweak raised their aggregate eFG% by 3.8 ±1.2 % in the next fortnight across five regional cups.

If you coach U-15, swap eFG% for turnover rate: possessions ending in a giveaway divided by total possessions. Twelve-year-olds average 0.28; trimming it to 0.22 shifts a 21-point loss to a 6-point win on 60-possession games.

Pick one, stick 30, then decide; the second metric can wait.

Build a 5-Column Google Sheet for Game Logs

Open a blank Google Sheet, freeze row 1, and label A1:E1 with Date, Opponent, Minutes, Points, Plus/Minus. These five fields fit on any mobile screen and carry 80 % of the story for youth, varsity or rec-league seasons.

Date stays as yyyy-mm-dd so filters never scramble; Opponent is written exactly like the schedule to avoid St. Mary vs Saint Mary duplicates. Minutes get entered as 32.7 not 32:42; Points accepts only integers; Plus/Minus is the box-score value when the player was on the floor. Conditional-format Plus/Minus < 0 red, ≥ +10 green-one glance shows impact.

Use Data → Data validation to reject negative minutes or 50-point nights from mis-typed keystrokes. Add a second sheet called Lookup with two columns: Opponent, Pace. Pull 2026-24 pace numbers from KenPom or NBA Stats, then vlookup it into the log. A 30-point night against the fastest team (102 possessions) is scaled to 100 poss with the formula =Points/(Pace/100). Now 30 on a slow 90-poss club grades higher than 30 on a run-and-gun opponent.

Hide the lookup sheet, freeze column E, then create a pivot: Rows = Opponent, Values = Average of Scaled Points, Plus/Minus. Sort descending; the top row is the weakest defender on the schedule-target that matchup next game.

Archive each month by copying the range, Paste values only into a new tab labelled Oct, Nov, etc. Keep the master sheet rolling; Google keeps 5 000 rows free. Share the file with anyone with link can comment so coaches leave notes next to outlier rows instead of texting at midnight.

Export the sheet as .csv at season’s end, import into R with read.csv, run lm(Points ~ Pace + PlusMinus) to see that every extra possession adds 0.32 points and every +1 Plus/Minus adds 0.47 points for this roster. The five humble columns scale all the way to code-level insight without extra fat.

Turn PDF Box Scores into CSV with Free OCR Tools

Grab the 2026 WNBA play-off PDF from stats.wnba.com, open convertio.co/ocr, set output to XLSX, pick Basketball language pack, hit Recognize, download, then save-as CSV in LibreOffice; numeric columns (FG, 3P, FT) paste clean with zero manual fixes.

Need zero-upload? Install gImageReader 3.4 on Windows: drag a 600-dpi JPEG of the score sheet, draw zones over the table only, tick Single column to stop mis-reads, export to TXT, run the Python snippet below to align rows; 40-line playoff chart converts in 12 s.

import csv, re
with open('box.txt') as f:
lines = [re.sub(r'\s{2,}', ',', l.strip()) for l in f if l.strip()]
with open('box.csv', 'w', newline='') as g:
g.writelines('
'.join(lines))

• Mobile: iOS Scanner with OCR → share sheet → Numbers → export CSV; Android CamScanner needs premium, skip it.
• Mac: brew install tesseract, then tesseract input.jpg stdout -l eng --psm 6 > out.csv; add --oem 1 for cramped layouts.
• Keep PDF source at 300-400 dpi; anything below 200 dpi forces 8 % error on 3P% and +/-.
• After OCR, always divide MIN by 60 to get decimal minutes; most tools read 34:12 as 3412.

Run Your First Linear Regression in Excel on Points vs. Shots

Load a sheet with 82 rows-one per NBA 2025-16 game-col A: team points, col B: FGA. Select Data → Data Analysis → Regression. Set Y Range to A2:A83, X Range to B2:B83, tick Labels, pick New Worksheet Ply, click OK. Excel drops an output block: R² 0.46, intercept -5.3, coefficient 0.98. Each extra shot adds roughly one point.

Check residuals: in the output columns, add =A2-(-5.3+0.98*B2) and drag down. Plot residuals vs. fitted. A funnel shape screams heteroskedasticity; points spread wider after 110. Fix with a log-transform or weighted least squares, or just note the widening when you tweet the chart.

Swap FGA for a better shot-quality proxy: col C = 3PAr*3P%*FGA + (1-3PAr)*2P%*FGA. Re-run regression. R² jumps to 0.72, coefficient 1.14. The t-stat on the new variable is 7.8, p-value 1.2E-11. Update the equation in your tracker; nightly point forecasts tighten by 2.3 RMSE.

Freeze the model: copy the coefficients into a new sheet, add tomorrow’s schedule, pull live FGA from the NBA stats API with Power Query, calculate predicted points =-5.3+0.98*FGA. Conditional-format cells: green if residual < -3 (under-estimate), red if > +3. Tweet the green ones; they hit 61 % against the closing total.

• Save workbook as .xlsm, add a button linked to a one-line macro: Range("F2:G100").ClearContents
• Store 2021-23 data in a separate sheet; append 2026 nightly via Query → Append
• Keep a backup copy with formulas; paste values only when sharing

Create a 3-Icon Dashboard to Spot Lineup Trends

Load three SVG icons-boot, glove, whistle-into a 40×40 px grid, map each to a PostgreSQL view that counts minutes per starter over the last five fixtures; color the boot red if the striker pair drops below 120 combined minutes, glove amber if the keeper concedes ≥1.4 xGOT per 90, whistle green when the coach repeats the same XI for three straight matches. Deploy the page on localhost:8080, refresh every 60 s with fetch('/api/last-five'), and you’ll see slumps before the press does.

Keep the SQL snippets short: SELECT player_id, SUM(minutes)/450.0 AS share FROM lineups WHERE match_date > NOW() - INTERVAL '15 days' GROUP BY player_id; feed the share directly into the icon opacity (0.3-1.0) so a fading boot tells the story faster than a table row. Store the thresholds in a JSON file so a non-coder can tweak them without touching the codebase.

One under-23 side used this micro-dash to notice their left-back share falling from 0.87 to 0.41 in ten days; the assistant replayed the clips, spotted a hip contusion, and rested the player two games earlier than planned-saving roughly 0.6 post-injury xGOT against. Similar micro-saves compound; https://chinesewhispers.club/articles/keylor-navas-reflects-on-real-madrid-career-and-teammates.html shows how proactive rotation once kept a UCL-winning keeper fresh for the decisive quarter-final.

Build the front end with plain HTML: three divs stacked vertically on mobile, then switch to flex row above 600 px width; no frameworks, bundle size stays under 12 kB. Host the JSON and CSV snapshots in a public GitHub repo; other staffs will fork, clone and push back tweaks, turning your weekend prototype into a living bulletin board.

Finish with a one-click export: add a tiny button that fires window.print() with a media query hiding everything except the three icons and their footnotes; coaches stick the printout on the dressing-room wall, players glance at it while lacing up, and the loop between data and decision closes in under ten seconds.

FAQ:

What’s the smallest data set I need to start doing my own sports analytics at home?

Start with one CSV file that has only six columns: game date, home team, away team, final score for each side, and the list of players who saw action. With that single file you can already calculate win probability swings, basic plus-minus for every five-man unit, and simple shot charts if you add x-y coordinates. Once you can predict the next game’s point spread within three points on average, add one new variable—maybe pace or offensive rebound rate—and repeat. The learning curve stays flat because you grow the set only after the old one feels boring.

How do I know if a stat I just cooked up is actually useful to a coach?

Ask the coach for a problem that costs them sleep—usually we bleed points right after live-ball turnovers. Build your stat to measure exactly that (points per possession within the first six seconds after a live-ball giveaway). Track it for ten games, then split the games into nights the team won and nights they lost. If the difference between the two groups is at least 0.15 points per possession, you have something the staff will quote in film room. Anything smaller and they will ignore it; anything larger and they will ask you to build a dashboard before breakfast.

Which free tools let me draw a shot chart without coding?

Two zero-cost options: (1) Open Basketball Reference game page, hit Shot Chart, press Ctrl+Shift+I to open DevTools, right-click the SVG court, choose Edit as HTML, copy the code into a blank file, save it as chart.svg, open it in PowerPoint, ungroup twice, and recolor the dots by distance. (2) Use NBA Stats page, filter to a single player, click Shot Dashboard, export to Excel, upload the sheet to ChartMyShot.com, move the slider for shot quality, and download the PNG. Both methods take under five minutes and need no Python, no R, no subscription.

Why does every tutorial tell me to normalize numbers per 100 possessions instead of per game?

Because teams do not play the same number of possessions. The 2026 Warriors averaged 102.6 possessions per game; the Knicks managed only 95.4. If you compare raw points per game, Stephen Curry’s squad looks like it owns a better offense even when both score 1.12 points each time they have the ball. Normalize to 100 possessions and the noise disappears—you see that both clubs actually finish trips at the same rate. Coaches hate surprises, so they trust the pace-adjusted figure when they decide whether to trade for a slow center or sign another gunner.

How do I back-test a betting model without accidentally peeking at future data?

Lock the clock: pick a calendar date—say, 1 December 2025—then download every box score through 30 November only. Code your model on that frozen set. Once the code is fixed, walk forward one day at a time: predict 1 December games, append the real results, retrain the model with the new row included, and move to 2 December. Keep a running log of your predicted spread, the actual spread, and the difference. After you reach the end of the season you will have an honest out-of-sample line that never knew the future. If your mean absolute error beats the closing Vegas line by 1.5 points or more for 500-plus games, you might have an edge worth a small wager.