Contains 8,000 users and 13 features
Time window is not explicitly documented in the source dataset
Presence of weekly aggregated variables suggests a short-term behavioral snapshot, but exact duration is unknown
We evaluate a hypothetical product change aimed at increasing user engagement on the platform.
We conduct an A/B test to determine whether the treatment has a meaningful impact on listening time.
Contains 8,000 users and 13 features
Time window is not explicitly documented in the source dataset
Presence of weekly aggregated variables suggests a short-term behavioral snapshot, but exact duration is unknown
| Column | Description | Type |
|---|---|---|
| user_id | Unique user identifier | ID |
| gender | User gender | Categorical |
| age | Age of user | Numeric |
| country | Country of user | Categorical |
| subscription_type | Free / Premium subscription | Categorical |
| listening_time | Total listening time | Numeric |
| songs_played_per_day | Avg songs played per day | Numeric |
| skip_rate | Fraction of songs skipped | Numeric |
| device_type | Device used (mobile/desktop/etc.) | Categorical |
| ads_listened_per_week | Ads listened per week (Free users) | Numeric |
| offline_listening | Whether user uses offline mode | Binary |
| is_churned | Whether user churned | Binary |
| treatment_group | A/B test assignment (control,treatment) | Categorical |
We simulate an A/B test, evaluating the impact of an improved recommendation system on user engagement and churn.
We compare pre-treatment characteristics across control and treatment groups to validate the random assignment.
Looking at the most important characteristics, subscription and device type, we see no meaningful differens between control and treatment groups.
| treatment_group | Control | Treatment |
|---|---|---|
| subscription_type | ||
| Family | 23.6% | 24.4% |
| Free | 24.9% | 25.9% |
| Premium | 26.7% | 25.8% |
| Student | 24.7% | 23.9% |
| treatment_group | Control | Treatment |
|---|---|---|
| device_type | ||
| Desktop | 34.7% | 34.7% |
| Mobile | 32.6% | 32.3% |
| Web | 32.7% | 33.0% |
| Metric | Control | Treatment | Relative Lift |
|---|---|---|---|
| Listening Time | 155.1 | 151.8 | -2.1% |
We test whether the observed difference in listening time between groups is statistically significant.
| t-statistic | p-value |
|---|---|
| -1.59 | 0.11 |
We estimate a small negative change in listening time for the treatment group compared to control.
| Metric | Control | Treatment | Relative Lift |
|---|---|---|---|
| Listening Time | 155.1 | 151.8 | -2.1% |
| t-statistic | p-value |
|---|---|
| -1.59 | 0.11 |
We examine whether the treatment effect varies across subscription types.
| treatment_group | Control | Treatment | Relative Lift |
|---|---|---|---|
| subscription_type | |||
| Family | 153.2 | 146.4 | -4.4% |
| Premium | 157.0 | 152.0 | -3.2% |
| Free | 156.5 | 151.7 | -3.1% |
| Student | 153.3 | 157.2 | 2.6% |
We examine whether the treatment effect varies across device types.
| treatment_group | Control | Treatment | Relative Lift |
|---|---|---|---|
| device_type | |||
| Desktop | 158.4 | 151.5 | -4.4% |
| Web | 156.5 | 153.4 | -2.0% |
| Mobile | 150.1 | 150.5 | 0.3% |
We evaluate whether the treatment negatively impacts retention or engagement quality.
| treatment_group | Control | Treatment | Relative Lift |
|---|---|---|---|
| Churn rate | 25.3% | 27.2% | 7.5% |
| Skip rate | 30.1% | 29.9% | -0.5% |
The treatment does not demonstrate a meaningful impact on user engagement.
Based on the experiment results:
We compare pre-treatment characteristics across control and treatment groups to validate the random assignment.
Looking at the most important characteristics, gender and age, we see no large differens between control and treatment groups.
| treatment_group | Control | Treatment |
|---|---|---|
| gender | ||
| Female | 33.4% | 32.8% |
| Male | 33.2% | 34.7% |
| Other | 33.4% | 32.5% |
| treatment_group | Control | Treatment |
|---|---|---|
| age_group | ||
| <18 | 6.9% | 6.1% |
| 18-25 | 16.3% | 16.1% |
| 26-35 | 21.8% | 21.7% |
| 36-50 | 33.9% | 35.7% |
| 50+ | 21.0% | 20.4% |