Case study 03 Secondary data analysis · Open Play dataset · 2026

Gaming & Psychological Wellbeing

Does playing more make you feel better — or does it matter more why you play?

Dataset
Open Play · Ballou et al. (2025)
Sample
6,844 observations · multi-platform
Tools
R · ggplot2 · lm()
Role
Sole analyst
Secondary data analysis R · ggplot2 Regression analysis Longitudinal dataset Gaming behavior Psychological wellbeing
01

Context & research question

The relationship between video games and mental health is one of the most contested questions in behavioral science. Popular narratives range from "gaming is harmful" to "gaming is therapeutic" — yet most research has focused on a single, easily measurable variable: how many hours people play. The assumption built into decades of research and policy is that more play means worse outcomes.

Recent work from Oxford's Internet Institute challenges this assumption. Ballou et al. (2025) found that among Nintendo Switch players, perceived value of gaming — not hours played — predicted mental wellbeing. This analysis independently replicates and extends that finding using the Open Play dataset, a large-scale longitudinal resource combining objective platform-level play data with validated psychological measures across Nintendo Switch, Steam, and Xbox players.

Primary question: Does the perceived value players derive from gaming predict psychological wellbeing above and beyond time spent playing — and if so, by how much?

Portfolio thread: This study connects directly to the other research in this portfolio. The thesis showed that perceived risk — not objective crime rates — drives fear. AlerTT showed that perceived value of an app — not its feature set — drives trust. Here, perceived gaming value — not playtime — predicts wellbeing. A consistent pattern: subjective appraisal matters more than objective behavior. That's a research identity, not a coincidence.

02

Approach

Rather than collecting new data, this study used the Open Play dataset — a publicly available longitudinal resource produced by researchers at Oxford, Tilburg, and Karolinska Institute. Using existing data allowed the analysis to focus on replication and extension of a finding with theoretical importance, rather than basic data collection.

6,844
Biweekly survey observations used in analysis
164
Variables in the biweekly survey dataset
5
Gaming value dimensions averaged into composite
Why secondary data
Replication is undervalued in UXR
Testing whether a finding holds in a different context is a genuine research contribution. Using an existing high-quality dataset allowed for faster analysis with more statistical power than a new survey would provide.
Why hierarchical regression
Isolating the unique contribution of each predictor
By entering playtime in Block 1 and gaming value in Block 2, the analysis directly tests whether gaming value explains additional variance in wellbeing beyond what playtime already accounts for.
Key predictor — gaming value
A composite of five dimensions
Gaming value was computed as the mean of five subscales: work (productive value), social (connection), cognitive (mental stimulation), emotion (emotional regulation), and routines (daily structure). This captures why people play, not just how much.
Key outcome — wellbeing
WEMWBS — a validated scale
Mental wellbeing was measured using the Warwick-Edinburgh Mental Wellbeing Scale (WEMWBS), a validated 7-item instrument used in large-scale population health research. Scores were averaged across items to form a composite.
03

Execution

The analysis was conducted entirely in R using tidyverse, ggplot2, and base R regression functions. The Open Play dataset was downloaded from Zenodo and loaded directly from compressed .csv.gz files — no preprocessing outside of R was required.

Composite scores were built using rowMeans() across validated subscale items. Hierarchical regression was run using lm() with two nested models. Visualizations were produced in ggplot2 and exported at 300dpi.

# Composite score construction analysis <- biweekly %>% mutate( gaming_value = rowMeans(select(., gaming_value_work, gaming_value_social, gaming_value_cognitive, gaming_value_emotion, gaming_value_routines), na.rm = TRUE), wellbeing = rowMeans(select(., wemwbs_1:wemwbs_7), na.rm = TRUE), playtime = self_reported_weekly_play ) # Hierarchical regression model1 <- lm(wellbeing ~ playtime, data = analysis) model2 <- lm(wellbeing ~ playtime + gaming_value, data = analysis)
04

Findings

Finding 01

Perceived gaming value strongly predicts wellbeing — playtime barely does

Pearson correlations reveal a clear pattern. Gaming value shows a meaningful positive relationship with mental wellbeing (r = 0.277) and mood (r = 0.223). Playtime, by contrast, shows near-zero relationships with both — and is actually weakly negatively associated with life satisfaction (r = −0.144).

Pearson correlations with wellbeing outcomes
Gaming value → mental wellbeingr = 0.277
Gaming value → mood (affective valence)r = 0.223
Gaming value → life satisfactionr = 0.129
Playtime → mental wellbeingr = −0.039
Playtime → life satisfactionr = −0.144
Finding 02

Adding gaming value increases explanatory power by 50x

The hierarchical regression tells the decisive story. Playtime alone explains just 0.16% of variance in wellbeing (R² = .002) — statistically significant but practically negligible. When gaming value is added in Block 2, explanatory power jumps to 8.2% (R² = .082). Gaming value's standardized coefficient is β = 0.288 (p < .001) — highly significant and meaningful. Playtime's coefficient remains effectively zero.

Metric Playtime alone + Gaming value
R² (variance explained) 0.002 (0.2%) 0.082 (8.2%)
Gaming value β 0.288***
Playtime β −0.00003** −0.00003**
F-statistic 8.22 (df=1) 230.0 (df=2)***
Finding 03 · Key insight

The two charts say everything

The scatter plots produced in ggplot2 make the finding visually undeniable. The gaming value chart shows a clear upward trend with a tight confidence band — as perceived value increases, wellbeing rises consistently. The playtime chart shows a nearly flat line trending slightly downward — hours played tells you almost nothing about how a player feels. The contrast between these two charts is the core of the finding.

GAMING VALUE → WELLBEING
Clear upward trend · r = 0.277
PLAYTIME → WELLBEING
Effectively flat · r = −0.039
05

Impact & recommendation

This finding has direct implications for how game studios think about player wellbeing — and how they measure it.

What this means for product teams at game studios

If player wellbeing is driven by perceived value — not playtime — then optimizing for session length is measuring the wrong thing. A player who logs 20 hours a week but derives little meaning from the experience will report lower wellbeing than one who plays 5 hours a week with high engagement and purpose.

Studios that use daily active users and average session length as their primary health metrics may be blind to the actual psychological experience of their players. The more relevant metrics would capture perceived value: did this session feel worthwhile? Did players feel connected, challenged, or relaxed?

For Riot specifically — a studio whose games are built around competitive intensity and community — this suggests that social connection and cognitive challenge (two of the five gaming value dimensions) are likely the primary drivers of player wellbeing. Design decisions that strengthen those dimensions — not ones that extend session time — are the ones most likely to produce players who feel good about their relationship with the game.

06

Reflection

8.2% of variance is meaningful but not large. The model explains a real portion of wellbeing variance — but the majority remains unaccounted for. Future analyses could add demographic controls, game genre, and platform type as covariates to improve explanatory power and identify which player segments show the strongest gaming value effects.

The gaming value composite flattens important distinctions. Averaging across five dimensions — work, social, cognitive, emotion, routines — produces a clean composite but loses information. A follow-up analysis examining which specific dimensions drive wellbeing most strongly would be more actionable for product teams.

This analysis is correlational. The regression shows that gaming value predicts wellbeing — but causal direction is not established. It's plausible that people who feel better generally also derive more value from gaming, rather than gaming value causing wellbeing. Longitudinal modelling using the full time-series data in the Open Play dataset would help untangle this.

R was new to me for this analysis. This study was conducted using R for the first time, moving from SPSS. The transition surfaced important lessons about data type handling, pipe-based workflows, and visualization with ggplot2 — skills now being applied to future analyses.