Lighthouse

Level 6 · Boss: the channel that looks worse

From Jordan Reyes, Head of Marketing

“Customers who find us on Pinterest spend about $100 more per order than Google Ads customers, so I want to move Google Ads budget to Pinterest. Before I do, can you check that the comparison is fair?”

Your goal

Step 1 of 3. Set result to a dict with the mean order_value per referral source, rounded to 2 decimals: keys "google_ads" and "pinterest".

Use the customers DataFramenot done yet
Use the orders DataFramenot done yet
Store a dict in resultnot done yet

Average completed-order value by referral source

Overall, pinterest is ahead. Inside each segment, google_ads is ahead. Both are true at the same time.

Show the numbers

	google_ads	pinterest
All orders	$535.22	$632.14
Business orders	$1,172	$1,027
Consumer orders	$374.4	$350.71

Step 1 · your code

Python

Loading the editor…

Reference solution

ResultPreview of customers · press Run to see yours

Preview of customers
customer_id	first_name	last_name	email	city
1	Owen	Hudson	owen.hudson@icloud.com	San Francisco
2	Douglas	Mendoza	dmendoza69@gmail.com	Atlanta
3	Henry	Brooks	hbrooks@lakesidecafe.com	Los Angeles
4	Brian	Graham	bgraham@bluecedarhomes.com	Seattle

Mote checks your result

Press Run. Each row that matches the goal becomes a firefly in your lantern.

Step 2 · your code

Python

Loading the editor…

Reference solution

ResultPreview of customers · press Run to see yours

Preview of customers
customer_id	first_name	last_name	email	city
1	Owen	Hudson	owen.hudson@icloud.com	San Francisco
2	Douglas	Mendoza	dmendoza69@gmail.com	Atlanta
3	Henry	Brooks	hbrooks@lakesidecafe.com	Los Angeles
4	Brian	Graham	bgraham@bluecedarhomes.com	Seattle

Mote checks your result

Press Run. Each row that matches the goal becomes a firefly in your lantern.

Step 3 · your code

Python

Loading the editor…

Reference solution

ResultPreview of customers · press Run to see yours

Preview of customers
customer_id	first_name	last_name	email	city
1	Owen	Hudson	owen.hudson@icloud.com	San Francisco
2	Douglas	Mendoza	dmendoza69@gmail.com	Atlanta
3	Henry	Brooks	hbrooks@lakesidecafe.com	Los Angeles
4	Brian	Graham	bgraham@bluecedarhomes.com	Seattle

Mote checks your result

Press Run. Each row that matches the goal becomes a firefly in your lantern.

DataFrames

Loaded for you as pandas DataFrames, plus pd and np.

customers2,018 rows · 9 columns

customer_idinteger
first_namevarchar
last_namevarchar
emailvarchar
cityvarchar
statevarchar
signup_datedate
segmentvarchar
referral_sourcevarchar

First 5 rows of customers
customer_id	first_name	last_name	email	city	state	signup_date	segment	referral_source
1	Owen	Hudson	owen.hudson@icloud.com	San Francisco	CA	2022-06-02	consumer	search
2	Douglas	Mendoza	dmendoza69@gmail.com	Atlanta	GA	2022-06-02	consumer	NaN
3	Henry	Brooks	hbrooks@lakesidecafe.com	Los Angeles	CA	2022-06-04	business	NaN
4	Brian	Graham	bgraham@bluecedarhomes.com	Seattle	WA	2022-06-06	business	pinterest
5	Kyle	Jimenez	kjimenez27@yahoo.com	Indianapolis	IN	2022-06-07	consumer	NaN

orders12,000 rows · 7 columns

order_idinteger
customer_idinteger
order_datetimestamp
statusvarchar
channelvarchar
shipping_feedecimal
coupon_codevarchar

First 5 rows of orders
order_id	customer_id	order_date	status	channel	shipping_fee	coupon_code
1001	125	2023-01-01 20:51:19	completed	web	0.00	NaN
1002	156	2023-01-02 07:58:46	completed	web	0.00	NaN
1003	22	2023-01-02 10:25:27	completed	marketplace	4.99	WELCOME10
1004	48	2023-01-02 11:21:51	completed	marketplace	4.99	WELCOME10
1005	82	2023-01-02 11:43:36	completed	web	0.00	WELCOME10

order_items30,338 rows · 6 columns

order_item_idinteger
order_idinteger
product_idinteger
quantityinteger
unit_pricedecimal
discount_pctdecimal

First 5 rows of order_items
order_item_id	order_id	product_id	quantity	unit_price
1	1001	121	9	31.99
2	1001	154	3	11.99
3	1002	154	4	11.99
4	1002	194	1	149.00
5	1002	193	2	19.99

products120 rows · 8 columns

product_idinteger
namevarchar
categoryvarchar
subcategoryvarchar
unit_pricedecimal
unit_costdecimal
launched_ondate
is_activeboolean

First 5 rows of products
product_id	name	category	subcategory	unit_price	unit_cost	launched_on	is_active
101	Vale Velvet Loveseat	Furniture	Sofas	1659.00	NaN	2023-09-11	true
102	Alder Oak Coffee Table	Furniture	Tables	919.00	503.50	2024-05-16	true
103	Orchard Rattan Dining Chair	Furniture	Chairs	789.00	429.87	2023-05-27	true
104	Upland Oak Bookcase	Furniture	Storage	429.00	214.70	2022-12-20	true
105	Avery Leather Bed Frame	Furniture	Beds	449.00	232.70	2021-08-31	true

returns1,673 rows · 5 columns

return_idinteger
order_item_idinteger
returned_ondate
reasonvarchar
refund_amountdecimal

First 5 rows of returns
return_id	order_item_id	returned_on	reason	refund_amount
1	14	2023-01-10	damaged	83.69
2	42	2023-01-10	wrong_item	19.99
3	105	2023-01-14	wrong_item	10.99
4	24	2023-01-16	changed_mind	26.99
5	126	2023-01-16	late	20.99

employees60 rows · 8 columns

employee_idinteger
first_namevarchar
last_namevarchar
departmentvarchar
titlevarchar
manager_idinteger
hire_datedate
salaryinteger

First 5 rows of employees
employee_id	first_name	last_name	department	title	manager_id	hire_date	salary
1	Catherine	Hale	Executive	Chief Executive Officer	NaN	2016-03-01	285000
2	Robert	Kim	Executive	Chief Operating Officer	1	2017-01-09	235000
3	Megan	Collins	Sales	Head of Sales	1	2018-04-16	172000
4	Chloe	Gutierrez	Sales	Account Executive	3	2020-01-21	72000
5	Frank	Bell	Sales	Account Executive	3	2023-05-21	73000

web_sessions40,000 rows · 7 columns

session_idinteger
customer_idinteger
started_attimestamp
devicevarchar
sourcevarchar
pages_viewedinteger
convertedboolean

First 5 rows of web_sessions
session_id	customer_id	started_at	device	source	pages_viewed	converted
1	NaN	2023-01-01 23:02:24	mobile	paid	1	false
2	18	2023-01-01 15:07:49	desktop	paid	1	false
3	NaN	2023-01-01 16:55:46	desktop	direct	7	false
4	NaN	2023-01-01 12:43:28	mobile	organic	3	false
5	NaN	2023-01-01 13:06:21	mobile	social	2	false

marketing_spend3,276 rows · 3 columns

spend_datedate
channelvarchar
amountdecimal

First 5 rows of marketing_spend
spend_date	channel	amount
2023-01-01	paid	492.99
2023-01-01	email	37.38
2023-01-01	social	200.53
2023-01-02	paid	709.13
2023-01-02	email	64.13

The lesson

Simpson’s paradox

Simpson’s paradox is when a comparison points one way in every subgroup and the other way when the subgroups are pooled. It sounds like a trick. It is ordinary, and it happens whenever two groups contain different mixes of something that strongly affects the outcome.

At Harbor & Pine that something is the segment. Business customers’ completed orders average about $1,050; consumer orders about $350. Any group with more business customers in it will have a higher average order value, whoever those customers are.

The mix behind the overall gap

pd.crosstab(two["referral_source"], two["segment"], normalize="index")

referral_source	business	consumer
google_ads	20%	80%
pinterest	42%	58%

Pinterest brings twice the share of business customers. Its overall average is pulled up by that mix, not by its customers spending more than comparable customers from Google Ads. Inside each segment, the comparison runs the other way.

Holding the mix equal

To compare fairly, give both sources the same mix and recompute. Take one common set of weights (the segment shares of all orders from both sources, 31% business and 69% consumer), and average each source’s segment means with those weights:

mix = two["segment"].value_counts(normalize=True)
by_segment.mul(mix, axis=0).sum()

This is called standardization. It answers: “what would each source’s average be if both had the same customers by segment?”

How sure is the reversal?

Not very, and saying so is part of the job. Order values vary a lot (a standard deviation around $1,150 for business orders), and there are only 155 google_ads business orders. The within-segment gaps of $145 and $24 are well inside what chance produces with groups this size. The solid finding is the negative one: the $97 overall gap is a mix effect, so it is no reason to move budget.

Common mistakes

Comparing pooled averages of groups with different compositions. Always ask what drives the outcome, then check whether the groups differ on it.
Declaring the reversed result the winner. The reversal can be noise; the mix effect is the finding.
Standardizing with one group’s own mix. Use a common mix for both, or the comparison is still tilted.
Splitting by every column until something flips. Choose the variable that drives the outcome for a reason you can explain, before you look.

Level 6 · Boss: the channel that looks worse

Your goal

Step 1 · your code

Reference solution

Mote checks your result

Step 2 · your code

Reference solution

Mote checks your result

Step 3 · your code

Reference solution

Mote checks your result

What Jordan gets, and why

DataFrames

The lesson

Simpson’s paradox

The mix behind the overall gap

Holding the mix equal

How sure is the reversal?

Common mistakes

Lantern lit!

Level 6 · Boss: the channel that looks worse

The request

Your goal

Step 1 · your code

Reference solution

Mote checks your result

Step 2 · your code

Reference solution

Mote checks your result

Step 3 · your code

Reference solution

Mote checks your result

What Jordan gets, and why

DataFrames

The lesson

Simpson’s paradox

The mix behind the overall gap

Holding the mix equal

How sure is the reversal?

Common mistakes

Cookie choices