Skip to content

virufy/paper-vector-career

BranchesTags

Repository files navigation

Project VECTOR: Volunteer Career Outcomes Analysis

Skill & Network Development as Predictors of Career Impact

A Quantitative Analysis of Virufy Volunteers (N = 78)

Amil Khanzada — Graduate Research in Career Outcomes & Development

R 4.0+ Data Audited Reproducible

Abstract

This repository implements a reproducible quantitative supplement for the paper "From Volunteer to Vocation: The Career Impact of Skill and Network Development in a Global Tech Nonprofit." Using relative importance analysis (LMG decomposition), we decompose the career-outcome variance attributable to seven skill and network predictors across 78 Virufy volunteers. The full-sample model achieves R² = 0.575, with Leadership Skills (q3) emerging as the strongest predictor (17.2% contribution), followed by Communication Skills (q2) (16.1%) and Network Quality (q6) (15.7%). Subgroup analyses reveal role-specific and career-stage-specific patterns, with students showing stronger leadership effects (23.3%) than professionals (10.4%). SEM fit is mixed but generally acceptable (CFI = 0.996, TLI = 0.994, SRMR = 0.030, RMSEA = 0.083).

Keywords: Career Development · Volunteer Outcomes · Relative Importance Analysis · Psychometric Modeling · Tech Nonprofit

1. Analytical Framework: Skill & Network Decomposition

This supplement decomposes six months of Virufy volunteer survey data (April 2025 – September 2025) into three interpretable components:

┌────────────────────────────────────────────────────────────┐
│           CAREER OUTCOMES PREDICTION PIPELINE              │
│                                                            │
│  ┌────────────────────┐  ┌──────────────────────┐        │
│  │  Skill Predictors  │  │  Network Predictors  │        │
│  │                    │  │                      │        │
│  │  • q1: Technical   │  │  • q5: Size          │        │
│  │  • q2: Comm.       │  │  • q6: Quality       │        │
│  │  • q3: Leadership  │  │  • q7: Access        │        │
│  │  • q4: Time Mgmt   │  │                      │        │
│  └─────────┬──────────┘  └──────────┬───────────┘        │
│            │                        │                     │
│            └────────────┬───────────┘                     │
│                         │                                 │
│                   Feature Engineering                     │
│                Scaling · Missingness Audit                │
│                Complete-Case Deletion (n=78)              │
│                         │                                 │
│         ┌───────────────┴───────────────┐                │
│         │   OLS Regression (LMG)        │                │
│         │   + Bootstrap Confidence      │                │
│         │   + VIF Diagnostics          │                │
│         │   + SEM Construct Validation │                │
│         └───────────────┬───────────────┘                │
│                         │                                 │
│    ┌────────────────────┼────────────────────┐           │
│    │                    │                    │           │
│  ┌─▼──────┐  ┌──────────▼────────┐  ┌───────▼─┐        │
│  │ Full   │  │ Subgroup Ranking  │  │ SEM     │        │
│  │ Rank   │  │ (Role, Stage, Geo)│  │ Structure│        │
│  │ Order  │  │                   │  │ Validity │        │
│  └────────┘  └───────────────────┘  └─────────┘        │
│                                                            │
│         ──► relative_importance_results.csv               │
│         ──► subgroup_analysis_results.csv                │
│         ──► sem_fit_indices.csv                          │
│         ──► paper_claim_check.csv                        │
└────────────────────────────────────────────────────────────┘

2. Key Results

Metric Full Sample Students Professionals
n 78 46 32
R² (OLS) 0.575 0.709 0.484
Top Predictor q3: Leadership (17.2%) q3: Leadership (23.3%) q6: Network Quality (21.5%)
#2 Predictor q2: Communication (16.1%) q4: Time Mgmt (17.1%) q1: Technical (18.6%)
#3 Predictor q6: Network Quality (15.7%) q2: Communication (15.3%) q2: Communication (15.5%)
SEM CFI 0.996 — —
SEM RMSEA 0.083 — —

Role-Type Results:

Predictor Tech (n=51) Non-Tech (n=27)
q1: Technical 12.7% 21.7% ⭐
q2: Communication 15.3% 12.9%
q3: Leadership 18.4% ⭐ 13.6%
q6: Network Quality 16.4% 15.2%

3. Reproducibility & Verification

3.1 Complete-Case Accounting

From output/participant_flow.csv:

  • Input rows: 80
  • Complete-case rows (q1–q11): 78
  • Excluded (missing core items): 2

3.2 Running the Analysis on a New Machine

To reproduce all results from scratch on a clean machine:

3.2 Quick Start: Run the Analysis in 3 Steps

To reproduce all results on your machine:

Step 1: Clone the Repository

git clone https://github.com/virufy/paper-career-supplement.git
cd paper-career-supplement

Step 2: Install R Dependencies (Once)

Rscript --vanilla install_dependencies.R

On Linux (Ubuntu/Debian), you may first need system tools:

sudo apt update && sudo apt install -y build-essential r-base-dev libcurl4-openssl-dev libxml2-dev libssl-dev

Step 3: Run the Consolidated Analysis Pipeline

Rscript --vanilla run_analysis.R

This single script:

  • Auto-detects data source: Uses input/vector_survey_responses.csv if available (real data), otherwise uses example data for demonstration
  • Executes full 6-step pipeline: Data audit → Descriptive stats → LMG analysis → Subgroup analysis → SEM → Paper claim verification
  • Generates 29 output files: CSV/HTML tables, PNG/SVG visualizations, and session metadata in the output/ directory
  • Takes ~2-5 minutes depending on your machine (bootstrap iterations: 1,000)

Output files are written to output/ (git-ignored, generated freshly each run):

relative_importance_results.csv       ← Main LMG rankings (Table 2)
subgroup_analysis_results.csv        ← Stratified findings (role, stage, geography)
sem_fit_indices.csv                  ← SEM model validation
paper_claim_check.csv                ← Automated paper reproducibility audit
correlation_heatmap.png              ← Visual predictor correlations
relative_importance_barplot.png      ← Visual LMG rankings
subgroup_top_predictors_comparison.png ← Subgroup comparison
[11 additional CSV audit files]

To Run with Your Own Data

If you have collected your own survey data using the standardized instrument:

  1. Ensure your CSV has the same structure: ≥18 columns with Likert items in columns 8–18
  2. Save it as input/vector_survey_responses.csv
  3. Run: Rscript --vanilla run_analysis.R
  4. Script automatically detects: "✓ Using real data: input/vector_survey_responses.csv"

Troubleshooting

Issue Solution
"Missing package 'X'" Run install_dependencies.R again or ensure internet connectivity
"Data file not found" Verify your CSV is at input/vector_survey_responses.csv (or use example)
Permission errors on Linux Try: chmod +x *.R && Rscript --vanilla install_dependencies.R
Very slow on large N Edit run_analysis.R line ~220: change R = 1000 to R = 500 for bootstrap iterations

5. Input Data Specification

File: input/vector_survey_responses.csv

Field Specification
Format CSV, comma-separated
Encoding UTF-8
Required columns Minimum 18 (see DATA_DICTIONARY.md)
Core Likert items Columns 8–18 → mapped to q1–q11
Missing data handling Complete-case deletion: rows with any NA in q1–q11 excluded
Primary outcome q10 (Job/Promotion Success)

6. Repository Structure

paper-vector-career/
├── README.md                                (this file)
├── DATA_DICTIONARY.md                       (variable mappings)
├── SUPPLEMENT.md                            (academic methods supplement)
├── VERIFICATION_REPORT.md                   (reproducibility audit)
├── install_dependencies.R                   (install R packages)
├── run_analysis.R                           (main analysis pipeline)
├── generate_figures.R                       (publication figures 1–4)
├── generate_tables.R                        (publication tables 1–5)
├── input/
│   └── vector_survey_responses.csv          (survey data — not in git for privacy)
├── statistical_appendix/
│   ├── README.md
│   ├── reproduce_analysis.R
│   └── vector_survey_responses_example.csv  (anonymised example dataset, N=30)
└── output/                                  (generated — not in git)
    ├── fig1_correlation_matrix.png
    ├── fig2_sem_path.png
    ├── fig3_lmg_forest.png
    ├── fig4_subgroup_comparison.png
    ├── table1_geography.html  … table5_convergence.html
    ├── relative_importance_results.csv
    ├── subgroup_analysis_results.csv
    ├── sem_fit_indices.csv
    ├── paper_claim_check.csv
    └── [additional CSV diagnostics]

7. Main Output Files

Data Quality

  • output/data_audit_summary.csv — input rows, complete rows, excluded rows
  • output/participant_flow.csv — stage-by-stage participant counts
  • output/core_item_missingness.csv — per-item missing data proportion

Full-Sample Analysis

  • output/relative_importance_results.csv — LMG rankings with 95% bootstrap CIs
  • output/full_model_diagnostics.csv — R², VIF, Shapirop, Breusch-Pagan p
  • output/correlation_matrix.csv — Spearman correlation heatmap data
  • output/correlate_heatmap.png — visual correlation matrix
  • output/relative_importance_barplot.png — LMG contribution chart

Subgroup Analysis

  • output/subgroup_counts.csv — sample sizes per demographic group
  • output/subgroup_analysis_results.csv — LMG rankings by role/stage/geography
  • output/subgroup_top_predictors_comparison.png — visual comparison

SEM & Validation

  • output/sem_fit_indices.csv — CFI, TLI, RMSEA, SRMR, composite correlations
  • output/paper_claim_check.csv — automated claim-by-claim verification

8. Statistical Methods

Method Package Purpose
LMG Relative Importance relaimpo Decompose R² across predictors; handles multicollinearity
OLS Diagnostics car, lmtest VIF, Shapiro-Wilk normality, Breusch-Pagan heteroscedasticity
Bootstrap Confidence Intervals boot 1,000-iteration BCa CI for LMG contributions
Spearman Correlation base R Non-parametric association for ordinal Likert items
Structural Equation Modeling lavaan WLSMV Latent construct validation (Skill_Development, Networking, Career_Outcomes)
Subgroup Analysis Custom Split samples by role type, career stage, geography

LMG Decomposition

The Lindeman-Merenda-Gold method decomposes the explained variance (R²) into contributions that account for predictor order and multicollinearity:

$$\text{LMG}_j = \frac{1}{P!} \sum_{\text{orderings } \pi} [R^2_{\pi_j} - R^2_{\pi_j^{-1}}]$$

where P is the number of predictors and $\pi_j^{-1}$ indicates the permutation without predictor j.

9. Reproducibility & Scientific Integrity

This repository adheres to the FAIR principles (Findable, Accessible, Interoperable, Reusable):

✓ Findable: versioned, publicly available, documented metadata
✓ Accessible: open-source code, example data included, dependency specifications
✓ Interoperable: standard CSV outputs, standard R ecosystem
✓ Reusable: standalone reproducible scripts

All numeric claims in the paper are automatically verified against code output in output/paper_claim_check.csv. See VERIFICATION_REPORT.md for full audit trail.

About

No description, website, or topics provided.

Resources

Readme
Activity
Custom properties

Stars

0 stars

Watchers

0 watching

Forks

1 fork
Report repository

Releases 1

Submission v1: From Volunteer to Vocation Latest
Apr 29, 2026

Packages

 
 
 

Contributors

Languages