Visualization Design Review

World Cup 2026 EPL Post — Against VISUAL_STANDARDS.md, Tufte, Wickham

Thinkers Referenced

Tufte — data-ink ratio, chartjunk, proportional ink (The Visual Display of Quantitative Information, 1983)
Cleveland — perceptual hierarchy: position > length > angle > area > color (Cleveland & McGill, 1984 JASA)
Wickham / Wilkinson — grammar of graphics: each aesthetic maps a variable through a declared scale
Wilke — Fundamentals of Data Visualization (2019) — most directly actionable modern reference
Knaflic — Storytelling with Data — titles state the finding, not the variable

Chart-by-Chart Assessment

1. Club Minutes Bar (chart-epl-club-minutes)

What works - Horizontal bars sorted descending — correct for long labels (Cleveland) - End labels substitute for gridlines (Tufte: reduce non-data ink) - Interactive tooltips avoid labeling every bar (Knaflic: Goldilocks annotation)

Issues - Title “2026 World Cup: Minutes by EPL Club” describes the data; Knaflic/Wilke standard is the finding: e.g. “Manchester City and Arsenal Sent the Most World Cup Minutes” - Logo inside bar is redundant with bar color for most clubs (Tufte would call it chartjunk for Arsenal, Chelsea, etc. — only earns its keep for disambiguating red clubs) - Club colors include 5 distinct reds (Liverpool, Arsenal, Nottm Forest, Southampton, Man United); colorblind simulation not run (VISUAL_STANDARDS §8.2 required)

2. Nation Stacked Bar (chart-nation-stacked)

What works - Club ordering (sorted by total minutes) is intentional, not alphabetical (Cleveland) - Standalone legend as a separate vconcat strip is an honest workaround for Vega-Lite’s emoji limitations

Issues — most serious of all charts - Stacked bar with 43 nations violates perceptual hierarchy. Only the first segment (England) can be compared across clubs with position encoding. All middle and end segments require comparing lengths from non-zero baselines — the least accurate perceptual task (Cleveland §1.4, VISUAL_STANDARDS §5) - VISUAL_STANDARDS §5 explicitly says: stacked bar appropriate for ≤5 categories; use treemap/sunburst for more - Tufte’s preferred alternative: small multiples — one bar per club showing only top-5 nations, faceted. Each sub-chart has a zero baseline and position encoding - Color scheme (tableau20) has no perceptual ordering; 43 flag emojis in the legend are difficult to scan

Recommended rework: normalize to % of each club’s total and show only top-5 nations + “Other” (6 stacked segments max). OR switch to a dot plot: clubs on x, one row per nation, dot size = minutes.

3. EPL Players Bar (chart-players, 60 rows)

What works - Player+flag on y-axis elegantly encodes nationality without a separate legend - Club logo inside bar provides club identity without needing a second legend

Issues - Wilke’s practical guideline: stop at ~20 rows for horizontal bar readability. 60 rows requires a very tall chart and relies on the user scrolling — legibility degrades toward the bottom - Player_flag (name + emoji concatenated string) is a grammar hack; the flag is data that should get its own aesthetic channel (Wickham) - Colorblind issue: multiple red-bared clubs (Liverpool, Arsenal, etc.) are indistinguishable on the bars; logos partially mitigate but don’t eliminate

4. EPL Position vs WC Minutes Scatter (chart-scatter-position-minutes)

Strongest chart in the post - Position encodes both variables on common scales — highest perceptual accuracy (Cleveland) - OLS trendline with R²/slope/p in subtitle is the correct place for statistical annotation (reduces chartjunk) - x-domain extended to [0,21] prevents logo clipping

Issues - Logos as point marks instead of standard shapes violates grammar: mark types should come from a finite, perceivable, mappable set (Wickham). In practice logos work here because they directly label club identity — a legitimate trade-off - “EPL Position vs. World Cup Minutes (2025/26)” as title is still description, not finding. Consider: “Better EPL Clubs Tended to Send More World Cup Minutes”

5. G+A per 90 by Club (chart-epl-club-ga-overview)

What works - Single question: which clubs were most efficient? - Sorted correctly

Issues - Same title format issue (description vs. finding) - Small sample size problem not addressed in the chart: Newcastle at #1 with very few total minutes looks dominant but is noise. Consider a minimum-minutes filter or an annotation flagging small-sample clubs

Systemic Issues Against VISUAL_STANDARDS Checklist (§12-13)

Check	Status
Chart answers one question (§1.5, §12)	✓ mostly — stacked bar tries to answer two
Encoding matches task (§5)	⚠ stacked bar is wrong chart type for 43 categories
Color palette colorblind-safe (§8.2)	✗ not verified; 5 red clubs fail deuteranopia
Bar charts include zero baseline (§10.1)	✓
Axis labels include units (§9.2)	✓ (“World Cup minutes”, “G+A per 90 min”)
Title states finding not description (§11.2)	✗ all titles are descriptive
Annotation of significant features only (§11.1)	✓
Caption includes data source (§12)	✓ footnote
Categorical variables sorted intentionally (§13)	✓ sorted by value throughout
Direct labels instead of legend for ≤5 series (§11.2)	⚠ stacked bar legend has 43 entries

Priority Fixes (Highest ROI)

Nation stacked bar: Limit to top-5 nations + Other (6 segments). Transforms a perceptually broken chart into a readable one. (High impact, medium effort)
Colorblind check: Run Coblis or colorspace::cvd_emulator() on the multi-red club bars. Add logo disambiguation note in methodology. (Low effort, compliance)
Chart titles: Change from description to finding. One-line edits. (Low effort)
Small-sample caveat: Add annotation on G+A chart flagging clubs with <200 WC minutes as noisy. (Low effort)

Written 2026-07-24. Source: VISUAL_STANDARDS.md at /Users/kivan/repos/incubator/mgmt/VISUAL_STANDARDS.md.

# Visualization Design Review ## World Cup 2026 EPL Post — Against VISUAL_STANDARDS.md, Tufte, Wickham --- ## Thinkers Referenced - **Tufte** — data-ink ratio, chartjunk, proportional ink (*The Visual Display of Quantitative Information*, 1983) - **Cleveland** — perceptual hierarchy: position > length > angle > area > color (Cleveland & McGill, 1984 JASA) - **Wickham / Wilkinson** — grammar of graphics: each aesthetic maps a variable through a declared scale - **Wilke** — *Fundamentals of Data Visualization* (2019) — most directly actionable modern reference - **Knaflic** — *Storytelling with Data* — titles state the finding, not the variable --- ## Chart-by-Chart Assessment ### 1. Club Minutes Bar (chart-epl-club-minutes) **What works** - Horizontal bars sorted descending — correct for long labels (Cleveland) - End labels substitute for gridlines (Tufte: reduce non-data ink) - Interactive tooltips avoid labeling every bar (Knaflic: Goldilocks annotation) **Issues** - Title "2026 World Cup: Minutes by EPL Club" describes the data; Knaflic/Wilke standard is the finding: e.g. "Manchester City and Arsenal Sent the Most World Cup Minutes" - Logo inside bar is redundant with bar color for most clubs (Tufte would call it chartjunk for Arsenal, Chelsea, etc. — only earns its keep for disambiguating red clubs) - Club colors include 5 distinct reds (Liverpool, Arsenal, Nottm Forest, Southampton, Man United); colorblind simulation not run (VISUAL_STANDARDS §8.2 required) --- ### 2. Nation Stacked Bar (chart-nation-stacked) **What works** - Club ordering (sorted by total minutes) is intentional, not alphabetical (Cleveland) - Standalone legend as a separate vconcat strip is an honest workaround for Vega-Lite's emoji limitations **Issues — most serious of all charts** - **Stacked bar with 43 nations violates perceptual hierarchy.** Only the first segment (England) can be compared across clubs with position encoding. All middle and end segments require comparing lengths from non-zero baselines — the least accurate perceptual task (Cleveland §1.4, VISUAL_STANDARDS §5) - VISUAL_STANDARDS §5 explicitly says: stacked bar appropriate for ≤5 categories; use treemap/sunburst for more - Tufte's preferred alternative: **small multiples** — one bar per club showing only top-5 nations, faceted. Each sub-chart has a zero baseline and position encoding - Color scheme (tableau20) has no perceptual ordering; 43 flag emojis in the legend are difficult to scan **Recommended rework**: normalize to % of each club's total and show only top-5 nations + "Other" (6 stacked segments max). OR switch to a dot plot: clubs on x, one row per nation, dot size = minutes. --- ### 3. EPL Players Bar (chart-players, 60 rows) **What works** - Player+flag on y-axis elegantly encodes nationality without a separate legend - Club logo inside bar provides club identity without needing a second legend **Issues** - Wilke's practical guideline: stop at ~20 rows for horizontal bar readability. 60 rows requires a very tall chart and relies on the user scrolling — legibility degrades toward the bottom - `Player_flag` (name + emoji concatenated string) is a grammar hack; the flag is data that should get its own aesthetic channel (Wickham) - Colorblind issue: multiple red-bared clubs (Liverpool, Arsenal, etc.) are indistinguishable on the bars; logos partially mitigate but don't eliminate --- ### 4. EPL Position vs WC Minutes Scatter (chart-scatter-position-minutes) **Strongest chart in the post** - Position encodes both variables on common scales — highest perceptual accuracy (Cleveland) - OLS trendline with R²/slope/p in subtitle is the correct place for statistical annotation (reduces chartjunk) - x-domain extended to [0,21] prevents logo clipping **Issues** - Logos as point marks instead of standard shapes violates grammar: mark types should come from a finite, perceivable, mappable set (Wickham). In practice logos work here because they directly label club identity — a legitimate trade-off - "EPL Position vs. World Cup Minutes (2025/26)" as title is still description, not finding. Consider: "Better EPL Clubs Tended to Send More World Cup Minutes" --- ### 5. G+A per 90 by Club (chart-epl-club-ga-overview) **What works** - Single question: which clubs were most efficient? - Sorted correctly **Issues** - Same title format issue (description vs. finding) - Small sample size problem not addressed in the chart: Newcastle at #1 with very few total minutes looks dominant but is noise. Consider a minimum-minutes filter or an annotation flagging small-sample clubs --- ### 6. G+A per 90 by Player Dropdown (chart-epl-club-ga) **What works** - Dropdown is the right interaction here: the question is per-club, not cross-club **Issues** - Healy: interactive filtering that requires user action hides data by default — the user must know to check each club. Consider starting with all clubs visible and using the dropdown as a highlight rather than a filter - Duplicate `add_params(club_sel)` on layered charts caused Altair warning (fixed) --- ## Systemic Issues Against VISUAL_STANDARDS Checklist (§12-13) | Check | Status | |---|---| | Chart answers one question (§1.5, §12) | ✓ mostly — stacked bar tries to answer two | | Encoding matches task (§5) | ⚠ stacked bar is wrong chart type for 43 categories | | Color palette colorblind-safe (§8.2) | ✗ not verified; 5 red clubs fail deuteranopia | | Bar charts include zero baseline (§10.1) | ✓ | | Axis labels include units (§9.2) | ✓ ("World Cup minutes", "G+A per 90 min") | | Title states finding not description (§11.2) | ✗ all titles are descriptive | | Annotation of significant features only (§11.1) | ✓ | | Caption includes data source (§12) | ✓ footnote | | Categorical variables sorted intentionally (§13) | ✓ sorted by value throughout | | Direct labels instead of legend for ≤5 series (§11.2) | ⚠ stacked bar legend has 43 entries | --- ## Priority Fixes (Highest ROI) 1. **Nation stacked bar**: Limit to top-5 nations + Other (6 segments). Transforms a perceptually broken chart into a readable one. (High impact, medium effort) 2. **Colorblind check**: Run Coblis or `colorspace::cvd_emulator()` on the multi-red club bars. Add logo disambiguation note in methodology. (Low effort, compliance) 3. **Chart titles**: Change from description to finding. One-line edits. (Low effort) 4. **Small-sample caveat**: Add annotation on G+A chart flagging clubs with <200 WC minutes as noisy. (Low effort) --- *Written 2026-07-24. Source: VISUAL_STANDARDS.md at `/Users/kivan/repos/incubator/mgmt/VISUAL_STANDARDS.md`.*