LEGO Dataset Analysis¶
In 1998 LEGO posted its first annual loss. By 2003 the company was close to insolvency — overextended across too many product lines, with inventory it couldn't sell and costs it couldn't cover. It was reportedly days from being acquired by Mattel. The turnaround that followed is one of the most studied in business school curricula. A central part of it was licensing: Star Wars launched in 1999, Harry Potter in 2001, and the number of sets and themes in LEGO's catalogue explodes from that point forward.
This notebook uses 70 years of product data from Rebrickable — 15,710 sets, 596 themes in a self-referential hierarchy, 135 colours with RGB values, and 25,000 minifigure inventory records — to quantify what that transformation looked like. The central question is not just how much licensed IP LEGO produced, but whether it changed the structure of what LEGO makes: the complexity of sets, the breadth of the colour palette, and the role of collectible minifigures as a product feature.
The notebook is in two parts. Part 1 establishes the dataset: the colour catalogue, the set inventory, volume and theme diversity over time, complexity growth, and theme rankings. These are the baselines against which the licensed IP story is measured. Part 2 contains six analyses that answer the business question directly — licensed vs original IP classification, K-Means complexity clustering, colour palette evolution traced through inventory data, recursive theme hierarchy resolution, a decade-by-decade summary table, and a minifigure density analysis that closes the argument.
Setup¶
Part 0 — Dataset Overview¶
Six CSV files from Rebrickable make up the full dataset. This section loads all six, inspects shape, columns, types, and key distributions for each, and maps the join relationships between them — establishing exactly what's available before any analysis begins.
Dataset Rows Cols Columns ---------------------------------------------------------------------- colors 135 4 id, name, rgb, is_trans sets 15,710 5 set_num, name, year, theme_id, num_parts themes 596 3 id, name, parent_id inventories 45,443 3 id, version, set_num inv_minifigs 25,103 3 inventory_id, fig_num, quantity inv_parts 1,496,680 6 inventory_id, part_num, color_id, quantity, is_spare, img_url
colors.csv — Colour definitions¶
Every LEGO colour ever used in production, with RGB hex value and transparency flag.
id int64 name object rgb object is_trans object dtype: object
| id | name | rgb | is_trans | |
|---|---|---|---|---|
| 0 | -1 | Unknown | 0033B2 | f |
| 1 | 0 | Black | 05131D | f |
| 2 | 1 | Blue | 0055BF | f |
| 3 | 2 | Green | 237841 | f |
| 4 | 3 | Dark Turquoise | 008F9B | f |
| 5 | 4 | Red | C91A09 | f |
| 6 | 5 | Dark Pink | C870A0 | f |
| 7 | 6 | Brown | 583927 | f |
Transparency split: is_trans f 107 t 28 Name: count, dtype: int64 Null values per column: id 0 name 0 rgb 0 is_trans 0 dtype: int64
sets.csv — Product catalogue¶
One row per retail set: the primary table for every analysis in this notebook.
set_num object name object year int64 theme_id int64 num_parts int64 dtype: object
| set_num | name | year | theme_id | num_parts | |
|---|---|---|---|---|---|
| 0 | 001-1 | Gears | 1965 | 1 | 43 |
| 1 | 0011-2 | Town Mini-Figures | 1978 | 84 | 12 |
| 2 | 0011-3 | Castle 2 for 1 Bonus Offer | 1987 | 199 | 0 |
| 3 | 0012-1 | Space Mini-Figures | 1979 | 143 | 12 |
| 4 | 0013-1 | Space Mini-Figures | 1979 | 143 | 12 |
Year range: 1949 – 2021 num_parts range: 0 – 9,987 Sets with 0 parts: 1796 (11.4%) Unique themes used: 571 Null values per column: set_num 0 name 0 year 0 theme_id 0 num_parts 0 dtype: int64 Sets per decade (rough count): year 1940 5 1950 135 1960 376 1970 620 1980 1212 1990 2041 2000 3831 2010 6813 2020 677 Name: set_count, dtype: int64
themes.csv — Product lines and sub-lines¶
596 theme entries in a self-referential hierarchy: parent_id points back to themes.id.
id int64 name object parent_id float64 dtype: object
| id | name | parent_id | |
|---|---|---|---|
| 0 | 1 | Technic | NaN |
| 1 | 2 | Arctic Technic | 1.0 |
| 2 | 3 | Competition | 1.0 |
| 3 | 4 | Expert Builder | 1.0 |
| 4 | 5 | Model | 1.0 |
| 5 | 6 | Airport | 5.0 |
| 6 | 7 | Construction | 5.0 |
| 7 | 8 | Farm | 5.0 |
Root themes (no parent): 132 Child themes (has parent): 464 Sample root themes (top 10 by name): ['4 Juniors', 'Adventurers', 'Agents', 'Alpha Team', 'Angry Birds', 'Aquazone', 'Architecture', 'Atlantis', 'Avatar', 'Belville'] Sample child/grandchild themes: id name parent_id 2 Arctic Technic 1.0 3 Competition 1.0 4 Expert Builder 1.0 5 Model 1.0 6 Airport 5.0 7 Construction 5.0 8 Farm 5.0 9 Fire 5.0
inventories.csv — Set-to-inventory bridge¶
Links set numbers to inventory IDs used in the parts and minifigures tables. Some sets have multiple inventory versions.
id int64 version int64 set_num object dtype: object id version set_num 1 1 7922-1 3 1 3931-1 4 1 6942-1 15 1 5158-1 16 1 903-1 17 1 850950-1 19 1 4444-1 21 1 3474-1 Total inventory records: 45,443 Unique sets in inventories: 43,537 Sets with >1 inventory: 1255 (2.9%) Max inventories for one set: 12
inventory_minifigs.csv — Minifigure records¶
One row per unique minifigure type in an inventory, with quantity. Joins to inventories on inventory_id.
inventory_id int64
fig_num object
quantity int64
dtype: object
inventory_id fig_num quantity
3 fig-001549 1
4 fig-000764 1
19 fig-000555 1
25 fig-000574 1
26 fig-000842 1
26 fig-008641 1
26 fig-008642 1
26 fig-008643 1
Total fig records: 25,103
Unique fig types (fig_num): 16,390
Sets containing >= 1 minifig: 9,173 (58.4% of all sets)
Quantity range per record: 1 – 100
inventory_parts.csv.gz — Part colour records¶
The largest file: one row per unique colour in each inventory, excluding or including spare parts.
inventory_id int64
part_num object
color_id int64
quantity int64
is_spare bool
img_url object
dtype: object
inventory_id part_num color_id quantity is_spare img_url
1 48379c04 72 1 False https://cdn.rebrickable.com/media/parts/photos/1/48379c01-1-839cbcec-62de-4733-ba23-20f35f4dd5d5.jpg
1 48391 0 1 False https://cdn.rebrickable.com/media/parts/photos/135/48391-135-cde3e1b7-1f79-40bf-9b79-46fcf5dbae96.jpg
1 48395 7 1 False https://cdn.rebrickable.com/media/parts/photos/7/48395-7-b9152acf-2fa5-4836-a04d-5b7fd39c2406.jpg
1 48864c01 25 1 False NaN
1 stickerupn0077 9999 1 False NaN
3 2343 47 1 False https://cdn.rebrickable.com/media/parts/elements/3000240.jpg
3 3003 29 1 False https://cdn.rebrickable.com/media/parts/elements/4550359.jpg
3 30176 2 1 False https://cdn.rebrickable.com/media/parts/elements/4114348.jpg
Total records: 1,496,680
Unique color_ids referenced: 273
Unique inventory_ids: 37,997
is_spare distribution:
is_spare
False 1390741
True 105939
Name: count, dtype: int64
From table column To table column ----------------------------------------------------------------- sets theme_id themes id inventories set_num sets set_num inv_parts inventory_id inventories id inv_parts color_id colors id inv_minifigs inventory_id inventories id themes parent_id themes id (self-join) Part-level join chain: inv_parts → inventories → sets → themes Minifigure join chain: inv_minifigs → inventories → sets → themes Colour evolution chain: inv_parts → inventories → sets (for year)
The central table is inventories — it is the bridge between the product catalogue (sets) and the item-level records (inv_parts, inv_minifigs). Every multi-table analysis routes through it. colors is a lookup table referenced only by inv_parts.color_id. themes is a lookup table with a self-join on parent_id for hierarchy resolution.
Part 1 — The Catalogue¶
The Colour Range¶
The colors.csv dataset lists every colour in LEGO's production range, with a hex RGB value and an is_trans flag marking transparent entries. The cells below establish how many distinct colours exist and how the palette is composed before the evolution analysis in Part 2.
| id | name | rgb | is_trans | |
|---|---|---|---|---|
| 0 | -1 | Unknown | 0033B2 | f |
| 1 | 0 | Black | 05131D | f |
| 2 | 1 | Blue | 0055BF | f |
| 3 | 2 | Green | 237841 | f |
| 4 | 3 | Dark Turquoise | 008F9B | f |
135
135 is a large palette for a construction toy range. The dataset records each colour with a unique ID, a name, an RGB hex value, and a transparency flag. The is_trans column is a string — 't' for transparent, 'f' for opaque — rather than a boolean, which affects the groupby output in the next cell.
Grouping by is_trans shows the transparent/opaque composition of the palette:
| id | name | rgb | |
|---|---|---|---|
| is_trans | |||
| f | 107 | 107 | 107 |
| t | 28 | 28 | 28 |
is_trans f 107 t 28 Name: count, dtype: int64
Opaque: 107 (79.3%) Transparent: 28 (20.7%)
107 opaque, 28 transparent — a roughly 79/21 split. Transparent colours are less common but not rare: more than one in five colours in LEGO's range is translucent. That proportion matters for the colour evolution analysis: transparent colours tend to cluster around specific eras and themes (Space, Castle, and later licensed lines requiring character-specific visors and window effects).
Themes and Sets¶
LEGO organises its product range into themes — Star Wars, Harry Potter, Technic, City, and hundreds more. A theme is a brand or concept that groups related sets together under a shared visual identity and storyline. A single theme can run for decades, accumulate hundreds of products, and operate entirely under licence from a third-party IP holder.
Each individual product on a shelf is a set: one box, one build, one unique identifier. The Millennium Falcon (set 75192, 7,541 parts, released 2017) is a single set within the Star Wars theme. Part count is the primary complexity signal throughout this analysis — it is a reliable proxy for build time, engineering sophistication, and retail price point.
The Set Inventory¶
The sets.csv file is the backbone of the analysis. Each row is one product: a unique set number, a name, a release year, a theme ID (foreign key to themes.csv), and a part count. The .head() and .tail() calls below show the shape of the data and the date range it covers.
| set_num | name | year | theme_id | num_parts | |
|---|---|---|---|---|---|
| 0 | 001-1 | Gears | 1965 | 1 | 43 |
| 1 | 0011-2 | Town Mini-Figures | 1978 | 84 | 12 |
| 2 | 0011-3 | Castle 2 for 1 Bonus Offer | 1987 | 199 | 0 |
| 3 | 0012-1 | Space Mini-Figures | 1979 | 143 | 12 |
| 4 | 0013-1 | Space Mini-Figures | 1979 | 143 | 12 |
| set_num | name | year | theme_id | num_parts | |
|---|---|---|---|---|---|
| 15705 | wwgp1-1 | Wild West Limited Edition Gift Pack | 1996 | 476 | 0 |
| 15706 | XMASTREE-1 | Christmas Tree | 2019 | 410 | 26 |
| 15707 | XWING-1 | Mini X-Wing Fighter | 2019 | 158 | 60 |
| 15708 | XWING-2 | X-Wing Trench Run | 2019 | 158 | 52 |
| 15709 | YODACHRON-1 | Yoda Chronicles Promotional Set | 2013 | 158 | 413 |
LEGO's Debut Year¶
Sorting by year surfaces the earliest entries — the products LEGO sold when it first entered the construction toy market. These sets predate standardised part systems, the modern stud-and-tube connection, and the minifigure (introduced in 1978). They're a useful anchor before looking at what the catalogue became: a 70-year product line with over 600 new sets per year at its peak.
| set_num | name | year | theme_id | num_parts | |
|---|---|---|---|---|---|
| 9521 | 700.1-1 | Extra-Large Gift Set (ABB) | 1949 | 365 | 142 |
| 9534 | 700.2-1 | Large Gift Set (ABB) | 1949 | 365 | 178 |
| 9539 | 700.3-1 | Medium Gift Set (ABB) | 1949 | 365 | 142 |
| 9544 | 700.A-1 | Small Brick Set (ABB) | 1949 | 371 | 24 |
| 9545 | 700.B-1 | Small Doors and Windows Set (ABB) | 1949 | 371 | 12 |
| 9526 | 700.1.2-1 | Single 2 x 2 Brick (ABB) | 1950 | 371 | 1 |
| 9546 | 700.B.1-1 | Single 1 x 4 x 2 Window without Glass (ABB) | 1950 | 371 | 1 |
| 9523 | 700.1.1-1 | Single 2 x 4 Brick (ABB) | 1950 | 371 | 1 |
| 9548 | 700.B.3-1 | Single 1 x 2 x 2 Window without Glass (ABB) | 1950 | 371 | 1 |
| 9549 | 700.B.4-1 | Single 1 x 2 x 4 Door without Glass (ABB) | 1950 | 371 | 1 |
All five of LEGO's debut sets were released in 1949 under the ABB brand (LEGO's early Scandinavian distributor). They were simple wooden and plastic gift sets — no themes, no minifigures, no standard part system. The set_num format (700.1-1, 700.2-1) reflects the catalogue numbering of the era. Everything in this dataset that follows builds on products that didn't exist when these five sets were sold.
The minimum year in the dataset establishes the start of the timeline. The count of sets in that debut year is the starting point for a growth curve that ends at over 600 releases per year in the 2010s — an increase of two orders of magnitude over seven decades.
1949
5
5 sets in 1949. The growth rate and the peak catalogue volume are both visible in the charts that follow — the tail of sets_by_year and the line chart together tell that story.
The Scale Ceiling¶
The sets with the highest part counts mark the upper boundary of LEGO's product range in any given era. These are collector-grade flagship releases — the Millennium Falcon, the Eiffel Tower, the Colosseum. Their part counts are relevant to the K-Means clustering in Part 2: they define what the Expert tier looks like, and their presence in the catalogue is almost exclusively a post-licensed-IP phenomenon.
| set_num | name | year | theme_id | num_parts | |
|---|---|---|---|---|---|
| 15004 | BIGBOX-1 | The Ultimate Battle for Chima | 2015 | 571 | 9987 |
| 11183 | 75192-1 | UCS Millennium Falcon | 2017 | 171 | 7541 |
| 10551 | 71043-1 | Hogwarts Castle | 2018 | 246 | 6020 |
| 295 | 10256-1 | Taj Mahal | 2017 | 673 | 5923 |
| 221 | 10189-1 | Taj Mahal | 2008 | 673 | 5922 |
The largest set in the dataset is not a Star Wars flagship — it is "The Ultimate Battle for Chima" (2015), a bulk battle-pack set with 9,987 parts. The UCS Millennium Falcon (2017, 7,541 parts) is second. Hogwarts Castle (2018, 6,020 parts) is third. The Chima outlier is worth noting: it is a quantity-of-figures set rather than a complex engineering build, which is why it tops the absolute part count despite not being a collector-tier product. The K-Means clustering in Part 2 handles this correctly by using era-relative complexity as a second feature alongside raw part count.
Volume Over Time¶
Grouping by year and counting sets maps LEGO's rate of catalogue expansion. One immediate data quality issue: the CSV was compiled mid-year, so the final two entries (2020, 2021) are incomplete and produce a misleading drop at the right edge of any trend chart. Both are excluded from all visualisations using a [:-2] slice.
One year to pay attention to: 1999. That is when Star Wars launched. The inflection visible in this chart is not a coincidence — it is the moment licensed IP began driving volume.
year 1949 5 1950 6 1953 4 1954 14 1955 28 Name: set_num, dtype: int64
year 2017 786 2018 816 2019 840 2020 674 2021 3 Name: set_num, dtype: int64
The tail confirms the data quality issue: 2020 shows 674 sets (plausible but partial) and 2021 shows just 3 (clearly incomplete). Both years are excluded from all trend charts with a [:-2] slice. The peak complete year is 2019 at 840 sets.
The clean trend line reveals four phases: near-flat through the 1970s (< 100 sets/year), steady growth through the 1980s, acceleration through the 1990s, and a steep climb from 1999 onward that roughly triples annual output by the 2010s.
Theme Diversity Over Time¶
Volume growth tells us how much LEGO was making. Theme diversity tells us how broadly it was making it. A company releasing 600 sets across 5 themes has a concentrated strategy; one releasing the same 600 sets across 200 themes has a fragmented one. Counting the number of distinct theme IDs active per year — using .agg() with pd.Series.nunique — reveals which kind of growth LEGO was pursuing at each stage.
The .agg() function is doing something that a simple .groupby().count() cannot: aggregating with a custom function (unique count) on a specific column while grouping on another. This is the general-purpose pattern for multi-metric aggregations throughout the analyses that follow.
year 1949 2 1950 1 1953 2 1954 2 1955 4 Name: theme_id, dtype: int64
| theme_id | |
|---|---|
| year | |
| 1949 | 2 |
| 1950 | 1 |
| 1953 | 2 |
| 1954 | 2 |
| 1955 | 4 |
The earliest years show minimal theme diversity: 2 themes in 1949, 1 in 1950. Theme count grew slowly through the 1960s and 1970s, then accelerated through the 1980s and 1990s as LEGO launched Castle, Space, Pirates, Technic, and dozens of sub-lines. The plateau comes later — visible in the chart that follows.
Theme diversity grew roughly in parallel with set volume through the 1980s and 1990s. Then, around 2003–2005, it plateaued: the number of active themes stopped growing even as annual set releases continued to climb sharply. LEGO had stopped launching new themes and started concentrating volume within established franchises. The two-metric divergence after that point is the key visual insight in this baseline section — it sets up the licensed IP analysis directly.
Volume and Theme Diversity on a Common Axis¶
The divergence is easier to read when both series share a time axis. Because sets and unique themes operate at different magnitudes — you cannot overlay them on a single y-axis without one flattening the other — ax.twinx() creates a shared x-axis with independent y-scales. Green (left axis) tracks set volume; blue (right axis) tracks theme count. The post-2005 pattern becomes unmistakable: sets climbing, themes flat. LEGO had found the franchises that worked and was scaling within them rather than diversifying further.
Set Complexity Over Time¶
Part count is simultaneously a measure of engineering ambition, build time, and price point. A 9,000-part Technic supercar occupies a completely different market position from a 50-part City starter set — different buyer, different retail channel, different margin. Averaging part count by year shows whether LEGO's product range has moved upmarket, stayed consistent, or bifurcated.
The 1954 versus 2017 comparison is the sharpest single number in this section: it captures 63 years of product evolution in one comparison, before the full time-series makes the trajectory visible. The K-Means analysis in Part 2 extends this into four labelled complexity tiers across the full catalogue.
| num_parts | |
|---|---|
| year | |
| 1954 | 12.357143 |
| 2017 | 221.840967 |
Average parts 1954: 12.4 Average parts 2017: 221.8 Complexity ratio 1954 → 2017: 18×
1954: 12.4 average parts per set. 2017: 221.8 average parts per set. That is an 18× increase over 63 years. 1954 sits at the low end of the 1950s average — a cluster of simple brick sets with barely a dozen parts each. The decade summary in Analysis 5 shows how this growth looks across every decade from the 1940s to the 2020s.
The trend is clear but noisy in early decades. Annual averages fluctuated between 80 and 140 parts through the 1970s and 1980s — the small number of sets per year means a single large flagship can swing the mean by 40 points. The upward direction becomes consistent from the late 1990s: average complexity crossed 150 in the mid-2000s as licensed-era sets scaled, and is approaching 260 in the 2020s. The sharpest sustained increase coincides precisely with the licensed IP era: flagship Star Wars and Technic sets were pulling the annual average up while the volume of simpler City and Friends sets was also increasing at the bottom of the range. K-Means separates these two populations into distinct tiers and names them.
Theme Rankings¶
Which themes have accumulated the most sets? The answer comes from .value_counts() on theme_id — but the result is a table of numbers, not names. Resolving those IDs to readable theme names requires joining sets.csv against themes.csv on the theme ID. That join also introduces the relational structure that underlies all of Part 2.
The themes dataset is not a flat lookup table. It has 596 entries in a self-referential parent/child hierarchy: each theme can have a parent_id pointing back to another row in the same table. Star Wars, for example, is registered under multiple IDs — one for the root franchise, others for sub-lines like Clone Wars, The Clone Wars, and The Old Republic. A direct join returns the immediate theme name; the hierarchy analysis in Part 2 resolves everything to root parents before counting, giving a truer picture of which umbrella franchises have accumulated the most sets across all their sub-lines.
theme_id 158 753 501 656 494 398 435 356 503 329 Name: count, dtype: int64
The entity-relationship diagram above shows the full Rebrickable structure. sets.theme_id → themes.id is the join used below. themes.parent_id is the self-referential key resolved recursively in Part 2. The right side of the diagram — inventories, inventory_parts, inventory_minifigs — is the join path used in the colour evolution and minifigure density analyses.
Querying themes.csv for 'Star Wars' returns multiple rows — each sub-line registered as a separate theme entry with its own ID. Filtering sets.csv to a single Star Wars theme ID therefore captures only a subset of the full franchise catalogue. This is why the ranked bar chart below undercounts Star Wars relative to its true total: the direct join does not aggregate sub-themes. The hierarchy analysis corrects for this.
| id | name | parent_id | |
|---|---|---|---|
| 0 | 1 | Technic | NaN |
| 1 | 2 | Arctic Technic | 1.0 |
| 2 | 3 | Competition | 1.0 |
| 3 | 4 | Expert Builder | 1.0 |
| 4 | 5 | Model | 1.0 |
| id | name | parent_id | |
|---|---|---|---|
| 17 | 18 | Star Wars | 1.0 |
| 150 | 158 | Star Wars | NaN |
| 174 | 209 | Star Wars | 207.0 |
| 211 | 261 | Star Wars | 258.0 |
Star Wars appears four times in themes.csv, each with a different ID: 18 (a child of Technic — an early Technic-style Star Wars sub-line), 158 (the root Star Wars theme), 209 (child of Seasonal — the Advent calendar line), and 261 (child of Mindstorms — a Mindstorms Star Wars kit). A simple join on theme_id therefore captures only a subset of Star Wars sets. The total across all four IDs is 776 sets. The hierarchy analysis in Part 2 recovers a slightly higher total of 791 by also aggregating the Ultimate Collector Series (theme 171), which is a registered child of theme 158 rather than a separate root-level Star Wars entry.
| set_num | name | year | theme_id | num_parts | |
|---|---|---|---|---|---|
| 8786 | 65081-1 | R2-D2 / C-3PO Droid Collectors Set | 2002 | 18 | 1 |
| 12051 | 8000-1 | Pit Droid | 2000 | 18 | 223 |
| 12058 | 8001-1 | Battle Droid | 2000 | 18 | 336 |
| 12066 | 8002-1 | Destroyer Droid | 2000 | 18 | 567 |
| 12071 | 8007-1 | C-3PO | 2001 | 18 | 339 |
| 12072 | 8008-1 | Stormtrooper | 2001 | 18 | 360 |
| 12073 | 8009-1 | R2-D2 | 2002 | 18 | 239 |
| 12074 | 8010-1 | Darth Vader | 2002 | 18 | 388 |
| 12081 | 8011-1 | Jango Fett | 2002 | 18 | 425 |
| 12083 | 8012-1 | Super Battle Droid | 2002 | 18 | 378 |
| 15273 | K8008-1 | Darth Vader / Stormtrooper Kit | 2002 | 18 | 0 |
| set_num | name | year | theme_id | num_parts | |
|---|---|---|---|---|---|
| 850 | 11912-1 | Star Wars: Build Your Own Adventure | 2016 | 158 | 73 |
| 855 | 11920-1 | Parts for Star Wars Build Your Own Adventure: ... | 2019 | 158 | 70 |
| 1717 | 20006-1 | Clone Turbo Tank - Mini | 2008 | 158 | 64 |
| 1728 | 20007-1 | Republic Attack Cruiser - Mini | 2009 | 158 | 84 |
| 1738 | 20009-1 | AT-TE Walker - Mini | 2009 | 158 | 94 |
| ... | ... | ... | ... | ... | ... |
| 15686 | VP-4 | Star Wars Co-Pack of 7101 7111 and 7171 | 2000 | 158 | 0 |
| 15689 | VP-8 | Star Wars Co-Pack of 7130 and 7150 | 2000 | 158 | 0 |
| 15707 | XWING-1 | Mini X-Wing Fighter | 2019 | 158 | 60 |
| 15708 | XWING-2 | X-Wing Trench Run | 2019 | 158 | 52 |
| 15709 | YODACHRON-1 | Yoda Chronicles Promotional Set | 2013 | 158 | 413 |
753 rows × 5 columns
Theme 18: 11 sets Theme 158: 753 sets Theme 209: 10 sets Theme 261: 2 sets Total across all four Star Wars theme IDs: 776 sets
Theme ID 158 (the root Star Wars entry) contains the main franchise sets — the numbered UCS and system sets from the core product line. Theme ID 18 contains a smaller cluster of early droid-and-vehicle sets from around 2000. The other two IDs contain promotional and supplemental sets. Combined, the four IDs account for 776 sets released between 1999 and 2021 (753 under the root theme ID 158, plus 23 across three registered sub-lines). The hierarchy analysis in Part 2 resolves a slightly higher total of 791 by also catching sub-themes of theme 158 that are not themselves named "Star Wars".
All four theme IDs — 18, 158, 209, and 261 — correspond to Star Wars content: different registered sub-lines in the Rebrickable taxonomy. Adding their set counts gives the true licensed total (776); the recursive root resolution in Part 2 adds the UCS sub-theme to reach 791.
Merging the value count series with themes.csv on id replaces numeric IDs with readable names. The result — sorted descending — is a ranked table of LEGO themes by direct set count, and a natural entry point into the licensed IP question: how many of the top themes are licensed, and what share of the total catalogue do they represent?
| id | set_count | |
|---|---|---|
| 0 | 158 | 753 |
| 1 | 501 | 656 |
| 2 | 494 | 398 |
| 3 | 435 | 356 |
| 4 | 503 | 329 |
| id | set_count | name | parent_id | |
|---|---|---|---|---|
| 0 | 158 | 753 | Star Wars | NaN |
| 1 | 501 | 656 | Gear | NaN |
| 2 | 494 | 398 | Friends | NaN |
| 3 | 435 | 356 | Ninjago | NaN |
| 4 | 503 | 329 | Key Chain | 501.0 |
The merge replaces numeric IDs with readable names. The top five entries — Star Wars (753 direct-ID sets), Gear (656), Friends (398), Ninjago (356), Key Chain (329) — reveal an important subtlety: "Gear" is not a construction theme but a catch-all category for non-brick merchandise. Key Chain is a sub-line of Gear. The bar chart below displays the top 10 by direct theme ID; the hierarchy analysis in Part 2 corrects for the Star Wars multi-ID split and removes the Gear distortion by aggregating to root parents.
Part 2 — Six Analyses¶
Analysis 1 — Licensed vs Original IP¶
Skills demonstrated: feature engineering · business framing
In 1998 LEGO posted its first annual loss. By 2003 the company was close to insolvency — overextended across too many product lines, haemorrhaging cash, reportedly close to being acquired by Mattel. The turnaround is a business school case study. A major part of it was licensing: Star Wars launched in 1999, Harry Potter in 2001, and the number of sets and themes in this dataset explodes from that point. The inflection is already visible in the dual-axis chart above — this analysis names it and quantifies it.
The method is a manually reviewed ground-truth classification: each of the 596 themes in themes.csv is labelled Licensed or Original in data/theme_licensing.csv. String matching was ruled out because licensed themes are registered under non-licensed parent roots — Star Wars sets under Technic, licensed Advent calendars under Seasonal, licensed Juniors sets under the Juniors root — making keyword matching on theme names unreliable. The CSV is the authoritative source; every licensed/original comparison in this notebook reads from it.
Business question: How much of LEGO's post-1999 growth was licensed IP, which franchises carried the most volume, and how has the licensed share of annual releases shifted over the two decades since the Star Wars deal?
Licensed themes: 75 Original themes: 521
Each of the 596 themes in themes.csv is labelled Licensed or Original in data/theme_licensing.csv — a manually reviewed ground-truth file. The classification covers sub-themes registered under non-licensed parent roots: Star Wars sets catalogued under Technic (theme 18), the Seasonal Advent line (theme 209), and Mindstorms (theme 261) are all correctly labelled Licensed, as are licensed Duplo and Juniors sub-lines (Disney, Marvel, Jurassic World), licensed Collectible Minifigure series (Disney, Harry Potter, The LEGO Batman Movie, The LEGO Movie), and all franchise-specific DC and Marvel sub-themes. The result is 75 licensed themes across 15,710 sets — fully auditable by opening the CSV.
| set_num | name | year | theme_id | num_parts | theme_name | ip_type | |
|---|---|---|---|---|---|---|---|
| 0 | 001-1 | Gears | 1965 | 1 | 43 | Technic | Original |
| 1 | 0011-2 | Town Mini-Figures | 1978 | 84 | 12 | Supplemental | Original |
| 2 | 0011-3 | Castle 2 for 1 Bonus Offer | 1987 | 199 | 0 | Lion Knights | Original |
1990: 0.0% licensed (108 total sets) 2000: 9.3% licensed (367 total sets) 2010: 16.1% licensed (435 total sets) 2020: 30.6% licensed (674 total sets)
1990: 0.0% licensed — licensed IP did not exist in LEGO's product line before the Star Wars deal. 2000: 9.3% — Star Wars and Harry Potter had both launched but were still a small share of a 367-set catalogue. 2010: 16.1% — a decade of franchise expansion (Marvel, DC, Pirates of the Caribbean, Indiana Jones, Cars) has roughly doubled the share. 2020: 30.6% — nearly one in three sets released in 2020 carries a licensed brand. The growth is steady rather than explosive, which reflects LEGO's strategy of expanding licensed lines gradually without crowding out the original portfolio.
Theme Set Count
Star Wars 776
Harry Potter 100
The LEGO Batman Movie 89
Disney Princess 80
Batman 74
Star Wars is the largest licensed franchise — 776 sets across all four Star Wars theme entries. Harry Potter follows at 100 sets. The LEGO Batman Movie (89 sets) and Disney Princess (80 sets) complete the top five along with Batman (74 sets). Star Wars has produced more sets than the next six licensed franchises combined, compressed into just 22 years (1999–2021).
Licensed sets never exceeded 50% of annual releases in this dataset Peak licensed share: 31.5% in 2019
Licensed sets never exceeded 50% of LEGO's annual releases in this dataset. The peak was 31.5% in 2019. This is the single most important number in the licensed IP analysis — it means the transformation was structural and economic (licensed sets commanding premium prices, driving complexity, introducing new colours) rather than numerical dominance. LEGO retained a majority original-IP product line throughout the licensed era. The strategy was augmentation, not replacement.
Finding: Licensed IP grew from 0% of LEGO's annual releases before 1999 to a peak of 31.5% in 2019 — significant, but never the majority of the portfolio. Star Wars is the single largest licensed franchise with 776 sets (across all theme registrations), more than double the next largest (Harry Potter, 100 sets). The licensed share at benchmark years — 9.3% in 2000, 16.1% in 2010, 30.6% in 2020 — shows steady growth, not a sudden shift.
That the figure never exceeded 50% is itself an important finding. It means licensed IP was a catalyst and a premium tier, not a replacement for LEGO's original product lines. City, Technic, and Creator continued to grow alongside Star Wars and Harry Potter. The strategy was portfolio expansion, not substitution — and the complexity and minifigure data in the analyses that follow explain why licensed sets punched above their weight despite representing less than a third of annual releases.
Analysis 2 — Complexity Clustering (K-Means)¶
Skills demonstrated: unsupervised ML · StandardScaler · KMeans
The average parts-per-set trend established in Part 1 shows a single upward curve. But an average obscures distribution. Is the catalogue genuinely shifting upmarket, or is a long tail of simple sets growing alongside an increasingly complex premium tier — making the average meaningless?
K-Means on two features answers this: num_parts (absolute complexity) and relative_complexity (parts divided by the annual mean — how complex is this set relative to its era). Using the era-relative feature is important: a 200-part set in 1985 was a large, complex product; the same set in 2015 is standard. Without normalising for era, the clustering simply separates old sets from new ones rather than finding genuine complexity tiers.
Four clusters are fitted and labelled by ascending mean part count: Starter, Standard, Advanced, Expert. The scatter and distribution charts show where those tiers sit across time and which themes belong to each.
Business question: Does LEGO's range span clearly distinct complexity tiers, or is part count a continuous distribution? Which themes own the Expert tier?
mean median count complexity Advanced 1247.232179 1153.0 491 Expert 3532.894737 3172.5 76 Standard 467.508756 425.0 2170 Starter 63.795831 39.0 11177
Total sets in catalogue: 15,710 Sets excluded (0 parts): 1,796 Sets clustered: 13,914 Starter % of all sets: 71.1% Starter % of clustered sets: 80.3%
K-Means has found four genuinely distinct tiers — the cluster sizes confirm this is not an arbitrary partition of a continuous distribution:
| Tier | Mean parts | Median parts | Set count |
|---|---|---|---|
| Starter | 64 | 39 | 11,177 |
| Standard | 468 | 425 | 2,170 |
| Advanced | 1,247 | 1,153 | 491 |
| Expert | 3,533 | 3,173 | 76 |
The Starter tier contains 71% of all 15,710 sets (80% of the 13,914 clustered sets). 1,796 sets with zero parts — service packs, spare-part bags, and gear items — are excluded from the clustering, as they carry no complexity signal. Within the clustered population, the distribution is similarly skewed: 80% Starter, 16% Standard, 4% Advanced, under 1% Expert.
The Expert tier has only 76 sets, but those sets define LEGO's premium collector positioning. The jump from Standard to Advanced (468 → 1,247 parts) and from Advanced to Expert (1,247 → 3,533) are both roughly 3×, suggesting the tiers reflect genuine price-point architecture rather than arbitrary statistical cuts.
Themes skewing Expert: theme_name Construction 1.960784 Town 0.289017 Star Wars 0.284900 Technic 0.226757 Basic Set 0.000000 Themes skewing Starter: theme_name Service Packs 99.227799 Duplo and Explore 96.875000 Town 93.930636 Supplemental 90.583804 Technic 89.115646
Expert-tier sets by theme: theme_name Creator Expert 12 Ultimate Collector Series 11 Modular Buildings 4 Construction 4 LEGO Art 4 Mosaic 3 Traffic 3 Educational and Dacta 3 Star Wars 2 Sculptures 2
Among the top 10 themes by set count, Construction has the highest proportion of Expert-tier sets (1.96%), followed by Town (0.29%), Star Wars (0.28%), and Technic (0.23%). These percentages are low because the top-10 themes include many Starter-heavy lines (City, Friends, Seasonal). The more telling figures are at the other end: Service Packs are 99.2% Starter, Duplo and Explore are 96.9% Starter, and Supplemental sets are almost entirely Starter-tier — bulk pieces, accessories, and educational materials that bring the average part count down without adding product complexity. Star Wars spans all four tiers, reflecting its deliberately broad price architecture from small battle packs (Starter) to flagship collector builds (Expert).
Finding: The four clusters are genuinely distinct — K-Means finds real structure in the data rather than arbitrary cuts through a continuum. The Expert tier (76 sets) is led by Creator Expert (12 sets) and the Ultimate Collector Series (11 sets) — LEGO's dedicated collector-grade lines — followed by Modular Buildings, LEGO Art, and Construction. The Starter tier is dominated by City, Friends, and seasonal sets.
The cluster scatter makes a second finding visible: the Expert tier has grown upward over time, with part counts increasing in absolute terms even after controlling for era via relative complexity. The gap between tiers has widened — LEGO's premium tier is genuinely more premium than it was 20 years ago, not just more expensive.
The theme distribution chart makes the structure clear. City and Duplo skew overwhelmingly Starter. Star Wars spans all four tiers — its range from small polybag sets to flagship UCS builds covers the full complexity spectrum. Technic appears Starter-dominated by raw set count (89% Starter) because LEGO registers hundreds of small Technic accessory bags, gear packs, and service items as individual sets — the handful of large flagship Technic builds are Expert-sized but too few to move the overall percentage.
Analysis 3 — Colour Palette Evolution¶
Skills demonstrated: multi-table joins · RGB manipulation · creative data viz
The RGB values in colors.csv are unused in the baseline analysis. But they contain a temporal story: each colour has a debut year, derivable by finding the earliest set that used it via the inventory join chain. That chain runs: inventory_parts → inventories → sets → year.
The connection to the licensed IP thesis is direct. Franchises imposed colour requirements that LEGO's original palette did not include: flesh tones for human minifigures (introduced when licensed themes moved away from the yellow-faced generic figure), specific metallics for vehicles, transparent colours for visors and special effects. Tracking when colours were introduced should reveal a concentrated expansion period aligned with the licensed IP era.
inventory_parts.csv.gz is decompressed on first run — it is stored compressed in the repository (13 MB) to avoid committing the 122 MB uncompressed file, which exceeds GitHub's per-file limit.
Business question: How did LEGO's colour vocabulary grow across seven decades, and does the expansion pattern align with the licensed IP timeline?
inventory_parts.csv already present
Colour-year records: 812,469 year color_id name is_trans 0 2004 72 Dark Bluish Gray f 1 2004 0 Black f 2 2004 7 Light Gray f 3 2004 25 Orange f 4 2004 9999 [No Color] f
812,469 colour-year records — each row represents one colour appearing in one inventory. This is the raw material for identifying when each colour was first used in a production set. The join chain correctly excludes spare parts (is_spare == False) to count only intentional colour usage rather than the supplemental pieces LEGO includes as replacements.
Transparent colours first appeared in: 1954 Total unique colours used in sets: 132
Transparent colours first appeared in 1954 — earlier than the widely cited association with the Space theme (1978). The 1954 entry is a small cluster of clear bricks in early gift sets, predating the modern stud-and-tube connection standard. The Space theme's 1978 launch popularised transparent colours for visors and cockpits, which is when the variety of transparent colours expanded — but the category itself is older.
132 of 135 catalogue colours have been used in at least one production set. Three colours exist in the catalogue but have no confirmed inventory appearances in this dataset snapshot — likely reserved, prototype, or regional-only entries.
New colours introduced by decade: decade new_colours 1940 6 1950 4 1960 4 1970 7 1980 5 1990 38 2000 53 2010 14 2020 1
New colours in 1990s: 38 New colours in 2000s: 53 1990s + 2000s combined: 91 Share of 132 production colours: 68.9%
The decade table tells a striking story:
| Decade | New colours introduced |
|---|---|
| 1940s | 6 |
| 1950s | 4 |
| 1960s | 4 |
| 1970s | 7 |
| 1980s | 5 |
| 1990s | 38 |
| 2000s | 53 |
| 2010s | 14 |
| 2020s | 1 |
Five decades of near-stasis, then two decades of explosion. The 1990s–2000s expansion introduced 91 new colours — 69% of the entire current palette — in just 20 years. The 2010s added only 14 more, suggesting the palette had reached practical saturation. One colour introduced so far in the 2020s confirms the trend has plateaued.
Finding: LEGO's colour vocabulary grew from 6 colours in the 1940s to 132 used in production sets by 2020. Transparent colours first appeared in 1954. The palette remained almost static through the 1960s–1980s (fewer than 10 new colours per decade), then expanded dramatically: 38 new colours in the 1990s and 53 in the 2000s — more introductions in that single decade than in the preceding five combined.
The 2000s expansion tracks the licensed IP timeline directly. Harry Potter required new flesh tones for human minifigures; Star Wars required specific metallic finishes for droids and spacecraft; Pirates of the Caribbean required skin-tone and aged-wood colours that had no place in the original palette. The palette did not expand for its own sake — it expanded because franchises demanded colours that the original brick range had never needed.
Analysis 4 — Theme Hierarchy Resolution¶
Skills demonstrated: recursive joins · relational data · graph traversal
themes.csv contains a parent_id column that is a foreign key pointing back to the same table — a self-referential structure that encodes a parent/child theme hierarchy. When you query for "Star Wars" sets using a direct theme_id join, you only get sets tagged with the root Star Wars theme ID. Sub-lines like Clone Wars, The Clone Wars, and The Old Republic have their own IDs and their own sets, and those are missed.
A recursive function that walks up the parent_id chain until it reaches a root (a theme with no parent) resolves every theme to its ultimate ancestor. Aggregating set counts at that root level gives a truer picture of which umbrella franchises have accumulated the most product across their full sub-theme ecosystems.
This is a graph traversal problem dressed as a pandas operation. The parent_id structure defines a forest of trees (multiple root nodes, each with a subtree of child and grandchild themes). The recursive resolver walks each tree upward.
Business question: Which top-level product lines have built the largest sub-theme ecosystems, and how deep does LEGO's theme taxonomy actually go?
Maximum hierarchy depth: 2 id name depth 6 Airport 2 7 Construction 2 8 Farm 2 9 Fire 2 10 Harbor 2 11 Off-Road 2 12 Race 2 13 Riding Cycle 2 14 Robot 2 15 Traffic 2 24 Airport 2 25 Castle 2 26 Construction 2 27 Race 2 28 Harbor 2 29 Train 2 30 Traffic 2 31 Creature 2 32 Robot 2 33 Food & Drink 2 34 Building 2 35 Cargo 2 39 Traffic 2 40 Creature 2 42 Airport 2 43 Building 2 44 Recreation 2 45 Cargo 2 46 Harbor 2 53 Airport 2 54 Cargo 2 55 Coast Guard 2 56 Construction 2 57 Farm 2 58 Fire 2 59 Harbor 2 60 Hospital 2 61 Police 2 62 Supplemental 2 63 Traffic 2 64 Off-Road 2 65 Arctic 2 66 Trains 2 68 Airport 2 69 Building 2 70 Cargo 2 71 Coast Guard 2 72 Construction 2 73 Farm 2 74 Fire 2 75 Food & Drink 2 76 Station 2 77 Harbor 2 78 Hospital 2 79 Off-Road 2 80 Police 2 81 Post Office 2 82 Race 2 83 Recreation 2 84 Supplemental 2 85 Traffic 2 95 Cargo 2 96 Coast Guard 2 97 Construction 2 98 Fire 2 99 Gas Station 2 100 Police 2 101 Race 2 102 Supplemental 2 103 Traffic 2 106 Airport 2 107 Coast Guard 2 108 Fire 2 111 Police 2 149 Imperial Armada 2 151 Imperial Soldiers 2 152 Islanders 2 208 City 2 209 Star Wars 2 210 Belville 2 211 Castle 2 212 Classic Basic 2 213 Clikits 2 214 Creator 2 215 Pirates 2 216 Friends 2 228 Creator 2 237 My Own Creation 2 238 My Own Train 2 239 World City 2 242 12V 2 243 4.5V 2 244 9V 2 245 RC Train 2 374 Airport 2 375 Farm 2 376 Fire 2 377 Harbor 2 378 Traffic 2 379 Supplemental 2 380 Train 2 381 Construction 2 382 Cargo 2 455 Classic Town 2 519 NXT 2 520 RCX 2 521 WeDo 2 525 Technic 2 530 Control Lab 2 531 eLAB 2 532 Supplemental 2 614 Jungle 2 636 My Town 2 650 Legoville 2 679 Mars Exploration 2 698 UCS 2 708 The LEGO Batman Movie 2
Maximum hierarchy depth: 2. The deepest themes — Airport, Construction, Farm, Fire, Harbor, Off-Road, Race, and others — are all grandchildren: sub-themes of sub-themes. There are no great-grandchild themes. LEGO's taxonomy is broad, not deep: 596 themes arranged in a shallow two-level structure, with 132 top-level root themes and 464 child or grandchild themes beneath them. This is a product-catalogue classification system, not a deep semantic hierarchy.
Top 15 parent themes by total set count:
theme total_sets
Town 1304
Duplo 1268
Gear 1049
Star Wars 791
Collectible Minifigures 709
Educational and Dacta 660
Technic 568
Creator 509
Service Packs 471
Universal Building Set 448
System 437
Bionicle 428
Friends 398
Ninjago 386
Space 332
When sets are aggregated to root parents:
| Rank | Theme | Total sets |
|---|---|---|
| 1 | Town | 1,304 |
| 2 | Duplo | 1,268 |
| 3 | Gear | 1,049 |
| 4 | Star Wars | 791 |
| 5 | Collectible Minifigures | 709 |
Town and Duplo outrank Star Wars in total set count — a finding that surprises most people. Town has been in continuous production since 1978; its 61 sub-themes accumulated sets over four decades. Duplo has benefited from LEGO's near-monopoly on the toddler construction segment. Gear (1,049 sets) is not a construction line — it is a catch-all for non-brick merchandise and should be excluded from product-line comparisons. Star Wars at 791 sets, compressed into just 22 years (1999–2021), has a rate of set production that no other franchise matches.
Top 10 parent themes by sub-theme count:
theme sub_theme_count
Town 61
Bionicle 38
Collectible Minifigures 33
Duplo 33
Creator 27
Educational and Dacta 25
Technic 20
Space 20
System 18
Seasonal 16
Town leads on sub-theme count with 61 sub-themes — more than any other parent line. Bionicle follows with 38, then Collectible Minifigures (33) and Duplo (33). The high Bionicle count reflects LEGO's strategy of launching new character sub-lines within the Bionicle universe each year rather than a single continuous line. Creator (27 sub-themes) reflects the breadth of its "build anything" positioning. Star Wars does not appear in the top 10 for sub-theme count because LEGO registered its Star Wars lines as separate root themes rather than nested sub-themes of a single Star Wars root — which is why the set-count analysis required aggregating four distinct Star Wars theme IDs.
Finding: The hierarchy reaches a maximum depth of 2 levels — LEGO's taxonomy is broad rather than deep. There are no great-grandchild themes. When sets are aggregated to root parents, Town leads with 1,304 sets, followed by Duplo (1,268) and Gear (1,049). Note that Gear is a catch-all category for non-brick merchandise — clothing, accessories, storage — rather than a construction product line. Excluding it, Star Wars ranks third with 791 sets despite spanning only 22 years of production.
The sub-theme count table tells a complementary story: Technic leads with the most sub-categories, reflecting decades of product line expansion into specialised engineering niches. Star Wars has accumulated sub-lines faster than any other franchise — a direct consequence of the breadth of the IP and LEGO's strategy of creating sub-themes around individual film releases, TV series, and character collections. The hierarchy is LEGO's organisational chart for its product portfolio, and it shows which properties were managed as long-term platforms versus short-run licensed deals.
Analysis 5 — Decade Summary¶
Skills demonstrated: aggregation · pandas Styler · consulting-style output
The previous four analyses each answer a single-dimension question. This one synthesises across all of them: for each decade from the 1940s to the 2020s, what were the five key metrics simultaneously? Total set count, unique themes active, average parts per set, most popular theme by set count, and percentage of sets that were licensed IP.
Reading the table decade by decade tells the three-era story of LEGO's history more clearly than any individual chart. It also makes the inflection decade visible as a single row.
Business question: Which decade was the turning point, and what do the numbers show happened to all five metrics at once?
Decade Total Sets Unique Themes Avg Parts Most Popular Theme % Licensed 1940s 5 2 99.6 System 0.0 1950s 135 6 32.3 Town Plan 0.0 1960s 376 37 73.8 System 0.0 1970s 620 62 107.8 Construction 0.0 1980s 1212 91 90.4 Technic 0.0 1990s 2041 170 124.7 Technic 1.0 2000s 3831 279 166.4 Star Wars 9.7 2010s 6813 254 195.6 Star Wars 22.2 2020s 677 82 258.6 Friends 30.9
The decade table puts all five metrics side by side:
| Decade | Total Sets | Unique Themes | Avg Parts | Most Popular | % Licensed |
|---|---|---|---|---|---|
| 1940s | 5 | 2 | 99.6 | System | 0% |
| 1950s | 135 | 6 | 32.3 | Town Plan | 0% |
| 1960s | 376 | 37 | 73.8 | System | 0% |
| 1970s | 620 | 62 | 107.8 | Construction | 0% |
| 1980s | 1,212 | 91 | 90.4 | Technic | 0% |
| 1990s | 2,041 | 170 | 124.7 | Technic | 1.0% |
| 2000s | 3,831 | 279 | 166.4 | Star Wars | 9.7% |
| 2010s | 6,813 | 254 | 195.6 | Star Wars | 22.2% |
| 2020s | 677 | 82 | 258.6 | Friends | 30.9% |
Three findings stand out. First, average part count was actually lower in the 1980s (90.4) than the 1970s (107.8) — reflecting the explosion of small City and Service Pack sets in that decade pulling the mean down even as flagship sets grew larger. Second, theme count peaked in the 2000s at 279 and declined in the 2010s to 254 — consolidation, not expansion, defined the 2010s. Third, the 2020s already show the highest average part count at 258.6 parts despite the decade being only two years complete in this dataset — LEGO's complexity floor has risen permanently.
| Decade | Total Sets | Unique Themes | Avg Parts | Most Popular Theme | % Licensed | |
|---|---|---|---|---|---|---|
| 0 | 1940s | 5 | 2 | 99.6 | System | 0.0% |
| 1 | 1950s | 135 | 6 | 32.3 | Town Plan | 0.0% |
| 2 | 1960s | 376 | 37 | 73.8 | System | 0.0% |
| 3 | 1970s | 620 | 62 | 107.8 | Construction | 0.0% |
| 4 | 1980s | 1212 | 91 | 90.4 | Technic | 0.0% |
| 5 | 1990s | 2041 | 170 | 124.7 | Technic | 1.0% |
| 6 | 2000s | 3831 | 279 | 166.4 | Star Wars | 9.7% |
| 7 | 2010s | 6813 | 254 | 195.6 | Star Wars | 22.2% |
| 8 | 2020s | 677 | 82 | 258.6 | Friends | 30.9% |
Finding: The table reveals three distinct eras. A low-volume origins phase (1940s–1970s): fewer than 700 sets per decade, minimal theme diversity, simple products averaging under 110 parts, 0% licensed. An expansion phase (1980s–1990s): set volume tripled each decade, themes multiplied, average complexity grew modestly, licensed share still near zero (1.0% in the 1990s). And a licensed IP era (2000s–2020s): volume scaled to 3,831 then 6,813 sets per decade, licensed share jumped to 9.7% then 22.2%, and average part count climbed from 166 to 196 to 259.
The 2000s are the inflection decade across every metric simultaneously — not just licensed share. Set count doubled, 53 new colours entered the palette, minifigure density climbed sharply, and average part count grew faster than in any previous decade. The franchise deals that began in 1999 did not just add licensed sets to the catalogue. They changed the economics of the whole product line: higher average complexity, broader colour requirements, and greater pressure to release multiple sets per franchise year rather than a single flagship product.
Analysis 6 — Minifigures and the Collector Economy¶
Skills demonstrated: multi-table joins · feature derivation · hypothesis testing
The previous analyses establish that licensed IP grew, complexity grew, and the colour palette expanded — all moving together from 1999 onward. But they do not fully explain the mechanism. Why do licensed sets command premium pricing even when their part count is comparable to an original-IP set of the same size?
The answer is minifigures. A generic City police officer is a throwaway accessory. Luke Skywalker in his Episode IV outfit is a collectible. LEGO discovered that in licensed themes, the character is the product — the bricks are the packaging. If that hypothesis is correct, we should see two measurable signals in the data: (1) minifigure density (figures per 100 parts) rising over time as sets became more character-focused, and (2) licensed sets containing more unique character types per set than original-IP sets of comparable size.
inventory_minifigs.csv records which minifigures appear in which inventory, with quantities. The join chain mirrors the colour evolution analysis: inventory_minifigs → inventories → sets → year and theme.
Business question: Did the licensed IP era structurally shift LEGO sets from brick-centric construction toward character-collecting experiences — and is that shift measurable in the data?
Sets with at least one minifig: 6,204 (39.5% of all sets)
Licensed vs Original — avg unique fig types per set:
unique_fig_types total_figs figs_per_100_parts
ip_type
Licensed 2.95 3.54 4.33
Original 2.33 2.78 4.91
Licensed avg unique fig types: 2.95 Original avg unique fig types: 2.33 Licensed premium: 27.0%
6,204 of 15,710 sets contain at least one minifigure — 39.5% of the full catalogue. The majority of sets (60.5%) have no minifigures at all: these are purely architectural or vehicle builds, bulk brick sets, accessories, and Technic/Mindstorms sets that use no standard figures. The analysis below focuses on the 39.5% that do have figures, where the character-density signal is meaningful.
Licensed sets include measurably more distinct characters than original-IP sets:
| Unique fig types | Total figs | |
|---|---|---|
| Licensed | 2.95 | 3.54 |
| Original | 2.33 | 2.78 |
The difference in unique figure types (2.95 vs 2.33, a 27% premium) is the key signal. It means licensed sets are not just including more figures — they are including more distinct characters. That distinction drives collectibility: a generic knight can be replaced; Luke Skywalker in his Episode IV outfit cannot. Raw figures per 100 parts is not a meaningful differentiator between the two groups — the licensed advantage sits entirely in character variety, not figure volume.
decade 1970 3.50 1980 3.16 1990 3.18 2000 4.44 2010 5.48 2020 4.95
1970s density: 3.50 figs per 100 parts 2010s density: 5.48 figs per 100 parts Growth 1970s → 2010s: 56.6%
Avg unique fig types by licensed franchise (all, sorted by avg):
avg_unique_figs set_count
theme_name
Stranger Things 8.0 1
Ghostbusters 7.5 2
Ultimate Collector Series 6.4 25
Disney 5.0 3
Pirates of the Caribbean 4.7 14
Avatar 4.5 2
Harry Potter 4.2 80
Indiana Jones 4.2 17
Justice League 4.0 15
The Lord of the Rings 4.0 15
The Hobbit 3.9 20
The Lone Ranger 3.9 8
Scooby-Doo 3.8 6
Prince of Persia 3.8 6
Teenage Mutant Ninja Turtles 3.6 19
The LEGO Movie 3.5 28
X-Men 3.5 2
Avengers 3.4 49
Spider-Man 3.2 46
Star Wars 3.2 482
Minions 3.0 2
Batman 3.0 63
DC Comics Super Heroes 3.0 4
Fantastic Beasts 3.0 2
Minecraft 2.9 53
Jurassic World 2.9 28
SpongeBob SquarePants 2.9 14
The LEGO Movie II 2.8 37
Trolls: World Tour 2.8 9
Cars 2.7 31
Overwatch 2.7 6
Guardians of the Galaxy 2.7 9
Super Heroes Marvel 2.5 32
UCS 2.5 2
The Powerpuff Girls 2.5 2
Jurassic Park III 2.5 2
Superman 2.4 7
Iron Man 2.4 5
Jurassic World: Fallen Kingdom 2.3 3
Toy Story 2.3 21
Marvel Super Heroes 2.2 4
The LEGO Batman Movie 2.1 76
Angry Birds 2.0 2
Frozen II 2.0 8
Disney Princess 1.8 57
DC Super Hero Girls 1.7 10
Super Heroes DC 1.6 27
Harry Potter and Fantastic Beasts Series 1 1.0 22
Harry Potter and Fantastic Beasts Series 2 1.0 16
Disney Series 1 1.0 18
The LEGO Movie Series 1.0 16
Disney Series 2 1.0 18
DC Super Heroes 1.0 16
The LEGO Movie Series II 1.0 20
Finding: Minifigure density grew 57% from 3.5 figures per 100 parts in the 1970s to 5.5 in the 2010s — a structural shift, not statistical noise. The trend is not monotonic: density dipped to 3.2 in the 1980s before the 2000s surge. The steepest single-decade increase occurs between the 1990s (3.2) and the 2000s (4.4): the decade in which Star Wars, Harry Potter, and a dozen other franchises entered the catalogue simultaneously.
The licensed advantage sits in character variety, not raw density. Licensed sets average 2.95 unique character types per set versus 2.33 for original IP — a 27% premium. The effect is strongest in narrative-driven franchises where character identity is the primary purchase driver. Stranger Things (8.0) and Ghostbusters (7.5) show the highest averages, but both represent single-set franchises in this dataset — the figure reflects one set's character roster, not a sustained product strategy. Among franchises with multiple sets, the Ultimate Collector Series averages 6.4 unique character types, Pirates of the Caribbean 4.7, Harry Potter 4.2, and Indiana Jones 4.2.
Together, the six analyses close the argument. Licensed IP never represented more than 31.5% of annual releases — it was never the majority of what LEGO made. But it changed what LEGO making a set meant: higher parts count, broader colour requirements, and above all, sets engineered around collectible characters rather than generic brick constructions. The turnaround was not just about adding franchise logos to boxes. It was about discovering a product format — the character-dense licensed set — that created a category of buyer LEGO had never previously reached.