VR Personalization & Distraction Analysis

BBA Thesis – Over-Personalization in Virtual Reality

This repository contains the full statistical analysis pipeline for my thesis examining how personalization in virtual reality (VR) influences task performance, efficiency, and user frustration. The work focuses on the distinction between task-relevant and non-task-relevant personalization and investigates when personalization becomes distracting rather than beneficial.

---

Research Context

Personalization is often assumed to improve user experience in immersive systems. However, VR introduces high cognitive load, and excessive or poorly aligned personalization may impair task execution.
This study empirically tests whether personalization:

• Improves task performance when it is task-relevant
• Increases distraction when it is non-task-relevant
• Exhibits a non-linear “sweet spot”, beyond which performance declines

---

Dataset

• Source: Controlled VR experiment
• Format: .xlsx (Excel)
• Unit of analysis: Individual participant
• Design: Between-subjects experimental study
• Assumption: Small-N, non-normal distributions

Key Variables

• total_time_sec – task completion time
• dfop – path deviation (inefficiency)
• frustration – self-reported frustration
• Task-relevant interaction counts (map, arrow)
• Non-task-relevant interaction counts (decor, UI, objects)

Derived variables:

• total_task_relev
• total_non_relev
• total_interactions
• efficiency = op / total_time_sec

---

Analysis Pipeline

The script is structured into ten clear sections:

1. Imports

Standard Python data analysis and statistics libraries:

• pandas, numpy
• matplotlib, seaborn
• scipy.stats
• statsmodels

---

2. Data Loading

Loads the VR experiment dataset directly from Excel into a Pandas DataFrame.

---

3. Data Preparation

• Aggregates interaction counts into task-relevant and non-task-relevant totals
• Computes total personalization interactions
• Calculates an efficiency ratio to normalize performance by time

---

4. Exploratory Data Analysis (EDA)

• Descriptive statistics
• Shapiro–Wilk normality tests
• Histograms, KDEs, and Q-Q plots

Outcome: Normality assumptions are violated, motivating non-parametric testing.

---

5. Participant Grouping (Percentile-Based)

Participants are categorized into:

• Low
• Moderate
• High

Groups are created separately for:

• Task-relevant interactions
• Non-task-relevant interactions

Thresholds are based on the 33rd and 66th percentiles.

---

6. Threshold Sensitivity Check

A simplified grouping approach using fixed interaction thresholds is applied to verify robustness of grouping logic.

---

7. Hypothesis 1 – Task-Relevant Personalization

Tests whether increased task-relevant personalization affects:

• Completion time
• Path deviation
• Frustration

Methods:

• Kruskal–Wallis tests
• Mann–Whitney U pairwise comparisons
• Effect size estimation
• Boxplot visualizations

---

8. Hypothesis 2 – Non-Task-Relevant Personalization

Replicates Hypothesis 1 using non-task-relevant interaction groups to assess distraction effects.

---

9. Hypothesis 3 – Over-Personalization (Quadratic Regression)

Explores non-linear effects of total personalization using quadratic regression:

[ Y = \beta_0 + \beta_1(\text{interactions}) + \beta_2(\text{interactions}^2) ]

Models are estimated for:

• Efficiency
• Path deviation
• Frustration

Scatter plots with fitted curves visualize potential inverted-U relationships.

---

10. Sweet Spot Identification

• Median and mean comparisons across personalization levels
• Line plots highlight optimal moderation points
• Visual markers indicate performance-efficient personalization ranges

---

Statistical Rationale

• Non-parametric tests are used due to small sample size and non-normality
• Effect sizes accompany all inferential tests
• Regression models are exploratory and theory-driven, not predictive

---

Key Contribution

This analysis demonstrates that:

• Personalization is not inherently beneficial in VR
• Task relevance matters more than interaction volume
• Excessive personalization can increase cognitive load, reduce efficiency, and raise frustration

The findings support bounded and selective personalization as a design principle for immersive systems.

---

Reproducibility Notes

• File paths may need adjustment
• No stochastic elements or random seeds are used
• All plots are generated inline

---

Author

Mahika Vats
BBA Thesis – Virtual Reality, Personalization & Cognitive Load