Subscription price test

A randomized price test for the monthly plan of a fictional language-learning app, Sadeem Lingo. About 9 000 visitors to the pricing page were randomly shown one of four monthly prices (49, 59, 69 or 79 SAR). Each row records the variant, traffic source, device, whether the visitor subscribed and the first-month revenue. Designed for comparing several proportions at once and for the multiple-comparisons problem: which price maximises revenue, and which differences are real?

Open in the lab Download CSV

Rows: 9,000
Columns: 7
One row is: one visitor to the pricing page
Version: v1

Data dictionary

Column	Type	Unit	Summary	Description
`visitor_id` Visitor	id			Anonymous visitor identifier.
`variant` Price variant	category		4 levels: P49, P59, P69, P79	Price shown: P49 = 49 SAR/month … P79 = 79 SAR/month.
`price_sar` Monthly price	integer	SAR	mean 63.95, median 59, SD 11.19, range 49–79	Monthly price shown to the visitor.
`traffic_source` Traffic source	category		4 levels: search, social, direct, email	How the visitor arrived.
`device` Device	category		2 levels: mobile, desktop	Device type.
`converted` Subscribed	integer		mean 0.048, median 0, SD 0.2138, range 0–1	1 if the visitor subscribed, else 0.
`revenue_sar` First-month revenue	number	SAR	mean 3.087, median 0, SD 14.01, range 0–94	Price plus optional 15 SAR add-on for subscribers; 0 otherwise.

What it is good for

Compare four conversion rates with a chi-square test.
Run pairwise comparisons and adjust for multiple testing (Bonferroni, Holm).
Compare revenue per visitor (not just conversion) across prices.
Explain why the 'winning' variant from many comparisons may be a false positive.

Methods: chi-square test, two-proportion z-test, Bonferroni correction, Holm correction, confidence intervals, ANOVA on revenue per visitor.

Data quality notes

Conversion falls with price, but revenue per visitor peaks at a middle price — the differences in revenue are small and noisy.
In this particular sample the observed revenue per visitor is highest at 49 SAR, although the true optimum (by construction) is 59 SAR — a concrete example of how noisy a 'winner' can be.
Variants are not exactly equal in size (random assignment).
No missing values.

Provenance and licence

Visitors are assigned to the four prices with equal probability. True conversion rates fall with price: 6.0%, 5.4%, 4.4%, 3.6%, adjusted by traffic source (email 1.5×, search 1.1×, social 0.7×, direct 1.0×, averaging ≈1). A subscriber adds an optional 15 SAR add-on with probability 0.2. True revenue per visitor is therefore highest at 59 SAR, but the differences between neighbouring prices are small relative to sampling noise.

Synthetic data · CC0-1.0 (synthetic data generated by Statistics Quest) · generated deterministically (seed 20241006). It does not describe real people or organisations; any resemblance is coincidental.

Preview

Dataset

Subscription price test

9,000 rows · one row = one visitor to the pricing page · synthetic data

About this dataset Open in the lab

Visitor	Price variant	Monthly price (SAR)	Traffic source	Device
V0001	P59	59	search	mobile
V0002	P59	59	search	mobile
V0003	P59	59	social	mobile
V0004	P59	59	email	mobile
V0005	P69	69	direct	mobile
V0006	P49	49	email	mobile
V0007	P59	59	search	desktop
V0008	P49	49	search	desktop
V0009	P69	69	email	mobile
V0010	P59	59	email	mobile
V0011	P59	59	direct	mobile
V0012	P69	69	social	mobile

Used in

Multiple variants · Experiments and A/B Testing
Credible intervals · Advanced Foundations