PSTAT 131 Final Project — University of California, Santa Barbara
Author: Isaiah Singer
Date: May 2025
This study investigates whether age and gender significantly influence the amount of time people spend on their smartphones.
Using a dataset containing variables such as Daily_Screen_Time_Hours, Age, Gender, Location, Total_App_Usage_Hours, and Number_of_Apps_Used, I modeled and visualized patterns in smartphone engagement across demographics.
The central hypothesis was that younger generations are not necessarily more glued to their phones than older ones, contrary to popular belief.
- Test whether age and gender are significant predictors of daily screen time.
- Visualize trends across age ranges, genders, and cities.
- Fit and compare multiple predictive models to evaluate which best explains variance in smartphone use.
- R packages: tidymodels, ggplot2, dplyr, ggcorrplot, xgboost, ranger
- Models Tested: Linear Regression, Pruned Decision Tree, Random Forest, Gradient Boosted Tree
- Validation: 5-fold cross-validation using RMSE and R² metrics
- Feature Engineering: dummy encoding, correlation filtering (0.9 threshold), normalization
- Average daily screen time was roughly consistent across age groups, with a mild peak around ages 20–30.
- Gender showed almost no difference in daily screen time.
- Location had minor effects — for example, users in Houston had slightly higher variance than those in Los Angeles.
- Correlation heatmap confirmed no strong correlation between age and screen time or app usage hours.
| Model | RMSE | R² (approx.) | Notes |
|---|---|---|---|
| Linear Regression | ~3.7 | 0.0002 | Best performing overall, though variance explained was minimal. |
| Gradient Boosted Tree | ~3.8 | 0.00018 | Slightly worse after tuning; not significant improvement. |
| Random Forest | ~3.74 | 0.006 | Marginally better RMSE, still poor explanatory power. |
| Pruned Decision Tree | ~3.79 | 0.00 | No discernible structure found. |
Despite model tuning, no model could explain even 1.5% of variance — reinforcing the finding that age is not a meaningful predictor of smartphone screen time.
- Age and gender are not statistically significant factors in predicting daily screen time.
- The popular narrative that “younger people are always on their phones” is not supported by the data.
- The best-fitting model was a linear regression, though its predictive power was very weak — a result consistent with the study’s original hypothesis.
This project demonstrated practical application of:
- Building full machine learning workflows with
tidymodels - Cross-validation and model tuning
- Translating quantitative results into clear conclusions
Ultimately, the study suggests that smartphone overuse is a universal behavior, not limited by age or gender — a finding that emphasizes collective digital awareness rather than generational blame.
Created by Isaiah Singer — Data Analyst/Scientist | R, Python, SQL | UCSB Statistics & Data Science