Documents how I constructed the three datasets, their differences, inclusion criteria, and intended research applications.

02-Data/03-Analysis_Ready/README.md

@@ -0,0 +1,424 @@
+# Analysis-Ready Datasets — AffLex-Provo
+
+## 1. Overview
+
+This directory contains three research datasets constructed
+through the AffLex-Provo computational data preparation and
+triangulation pipeline.
+
+The datasets were prepared by Mohit Gupta as part of the
+AffLex-Provo research program at the University of Illinois
+Urbana-Champaign.
+
+The construction process integrates three established
+psycholinguistic research resources:
+
+1. Provo Corpus — Eye-tracking and predictability information.
+2. SUBTLEX-US — Lexical frequency information.
+3. Warriner et al. (2013) — Valence, arousal, and dominance norms.
+
+The resulting datasets support investigations of lexical
+and affective influences on eye-movement behavior during
+natural reading.
+
+The three datasets represent different levels of predictor
+availability and analytical eligibility.
+
+---
+
+## 2. Dataset Inventory
+
+| Dataset | Description | Intended Research Application |
+|---|---|---|
+| `01-Master_Provo.csv` | Complete triangulated Provo dataset, retaining observations regardless of lexical or affective norm availability. | General corpus research and coverage analyses |
+| `02_Lexical_Provo.csv` | Subset with complete mandatory lexical predictors. | Lexical Provo research program |
+| `03_Affective_Provo.csv` | Subset with complete mandatory lexical and affective predictors. | Affective Provo research program |
+
+All three datasets were generated through the
+AffLex-Provo computational preparation workflow.
+
+The datasets share a common source corpus but differ
+in their predictor-completeness requirements.
+
+---
+
+## 3. Source Research Datasets
+
+### 3.1 Provo Corpus
+
+The Provo Corpus provides eye-tracking observations
+and word predictability information collected during
+natural reading.
+
+Original publication:
+
+Luke, S. G., & Christianson, K. (2018).
+
+The Provo Corpus: A large eye-tracking corpus with
+predictability norms.
+
+*Behavior Research Methods, 50*, 826–833.
+
+https://doi.org/10.3758/s13428-017-0908-4
+
+### 3.2 SUBTLEX-US
+
+SUBTLEX-US provides lexical frequency information
+derived from American English subtitles.
+
+Original publication:
+
+Brysbaert, M., & New, B. (2009).
+
+Moving beyond Kučera and Francis: A critical evaluation
+of current word frequency norms and the introduction
+of a new and improved word frequency measure for
+American English.
+
+*Behavior Research Methods, 41*, 977–990.
+
+https://doi.org/10.3758/BRM.41.4.977
+
+### 3.3 Warriner Affective Norms
+
+Warriner et al. (2013) provide affective ratings
+for English lemmas across three dimensions:
+
+- Valence
+- Arousal
+- Dominance
+
+Original publication:
+
+Warriner, A. B., Kuperman, V., & Brysbaert, M. (2013).
+
+Norms of valence, arousal, and dominance for
+13,915 English lemmas.
+
+*Behavior Research Methods, 45*, 1191–1207.
+
+https://doi.org/10.3758/s13428-012-0314-x
+
+---
+
+## 4. Computational Dataset Construction
+
+The datasets were constructed using the AffLex-Provo
+Python preprocessing and triangulation pipeline.
+
+The workflow includes:
+
+1. Loading the original research datasets.
+2. Standardizing lexical identifiers.
+3. Evaluating vocabulary overlap.
+4. Identifying missing lexical entries.
+5. Applying lexical normalization and lemmatization.
+6. Resolving duplicate lexical entries before merging.
+7. Integrating SUBTLEX-US frequency information.
+8. Integrating Warriner affective norms.
+9. Generating lexical coverage indicators.
+10. Validating dataset integrity.
+11. Evaluating predictor completeness.
+12. Exporting the three research datasets.
+
+The corresponding computational notebook is:
+
+`02-Build_Master+Lexical+Affective_Provo(2).ipynb`
+
+The notebook documents the computational procedures
+used to generate these datasets.
+
+---
+
+## 5. Dataset 1 — Master Provo
+
+**Filename:** `01-Master_Provo.csv`
+
+### Purpose
+
+The Master Provo dataset provides the complete
+triangulated research resource.
+
+It combines Provo eye-tracking observations with
+available lexical frequency and affective norming
+information.
+
+### Construction
+
+The master dataset is generated using validated
+many-to-one lexical merges.
+
+The primary standardized lexical identifier is:
+
+`Word_Norm_lower_lemma`
+
+The merge procedure retains the original Provo
+token-level observations.
+
+### Coverage Indicators
+
+The dataset includes indicators documenting
+lexical norm availability:
+
+| Variable | Meaning |
+|---|---|
+| `Has_SUBTLEX` | Availability of SUBTLEX-US frequency information |
+| `Has_VAD` | Availability of Warriner affective norms |
+| `Has_FULL_NORMS` | Availability of both SUBTLEX-US and VAD information |
+
+### Filtering
+
+No analysis-specific filtering is applied.
+
+Observations are retained even when lexical
+or affective information is unavailable.
+
+This dataset is appropriate for:
+
+- Corpus-level descriptive analyses.
+- Vocabulary coverage investigations.
+- Missingness analyses.
+- Alternative preprocessing strategies.
+- Construction of additional analytical subsets.
+
+---
+
+## 6. Dataset 2 — Lexical Provo
+
+**Filename:** `02_Lexical_Provo.csv`
+
+### Purpose
+
+The Lexical Provo dataset supports investigations
+of conventional lexical predictors of eye movements.
+
+### Mandatory Predictors
+
+| Variable | Description |
+|---|---|
+| `Word_Length` | Word length |
+| `Response_Proportion` | Contextual word predictability |
+| `Lg10WF` | Log-transformed lexical frequency |
+
+### Inclusion Criteria
+
+Observations must satisfy:
+
+1. `Has_SUBTLEX == True`
+2. Nonmissing `Word_Length`
+3. Nonmissing `Response_Proportion`
+4. Nonmissing `Lg10WF`
+
+### Important Filtering Policy
+
+Eye-tracking outcomes are not used to determine
+inclusion in the exported Lexical Provo dataset.
+
+The dataset retains:
+
+- Skipped observations.
+- Zero-dwell-time observations.
+- Observations with missing eye-tracking outcomes.
+
+Outcome-specific filtering must be performed
+during subsequent statistical analysis.
+
+---
+
+## 7. Dataset 3 — Affective Provo
+
+**Filename:** `03_Affective_Provo.csv`
+
+### Purpose
+
+The Affective Provo dataset supports investigations
+of lexical and affective influences on eye movements.
+
+### Mandatory Lexical Predictors
+
+| Variable | Description |
+|---|---|
+| `Word_Length` | Word length |
+| `Response_Proportion` | Contextual word predictability |
+| `Lg10WF` | Log-transformed lexical frequency |
+
+### Mandatory Affective Predictors
+
+| Variable | Description |
+|---|---|
+| `V.Mean.Y` | Mean valence rating |
+| `A.Mean.Y` | Mean arousal rating |
+| `D.Mean.Y` | Mean dominance rating |
+
+### Inclusion Criteria
+
+Observations must satisfy:
+
+1. `Has_FULL_NORMS == True`
+2. Complete mandatory lexical predictors.
+3. Complete mandatory affective predictors.
+
+### Important Filtering Policy
+
+The exported dataset retains eye-tracking observations
+regardless of their outcome-specific eligibility.
+
+This includes:
+
+- Skipped observations.
+- Zero-dwell-time observations.
+- Missing eye-tracking measurements.
+
+The appropriate filtering procedures are applied
+separately for each statistical outcome.
+
+---
+
+## 8. Eye-Tracking Analysis Considerations
+
+The exported datasets are predictor-complete
+research resources.
+
+They are not automatically the final datasets
+used in every statistical model.
+
+### Fixation Duration Analyses
+
+Duration analyses may require:
+
+- Valid outcome measurements.
+- Positive fixation or dwell-time values.
+- Complete model predictors.
+- Appropriate grouping identifiers.
+
+### Skipping Analyses
+
+Skipping analyses must preserve both skipped
+and non-skipped observations.
+
+Filtering the dataset to positive dwell times
+before modeling skipping behavior would remove
+important observations.
+
+### Regression Analyses
+
+Regression-related analyses require eligibility
+criteria appropriate to the specific regression
+outcome and statistical model.
+
+### Reporting Requirements
+
+Researchers should report the actual number of
+observations included in each statistical model.
+
+The total number of rows in an exported dataset
+should not automatically be reported as the
+sample size for every analysis.
+
+---
+
+## 9. Dataset Integrity and Validation
+
+The construction pipeline includes procedures
+designed to preserve structural integrity.
+
+These include:
+
+- Standardized lexical matching.
+- Duplicate resolution.
+- Validated many-to-one merges.
+- Row-count verification.
+- Lexical coverage diagnostics.
+- Predictor completeness checks.
+- Missingness evaluation.
+- Final column standardization.
+
+The master dataset retains the original
+token-level structure during lexical integration.
+
+The Lexical Provo and Affective Provo datasets
+apply additional predictor-completeness criteria.
+
+---
+
+## 10. Reproducibility
+
+The three datasets were generated using the
+AffLex-Provo data preparation pipeline.
+
+Researchers seeking to reproduce their construction
+should consult the corresponding preprocessing
+notebooks and source documentation.
+
+Reproduction requires access to the applicable
+original research datasets and compatible
+computational dependencies.
+
+The preparation workflow was executed using
+personally owned computing resources.
+
+Subsequent statistical analyses used a combination
+of personal computing resources and Illinois Computes.
+
+Those statistical computing environments are
+documented separately.
+
+---
+
+## 11. Licensing and Attribution
+
+The datasets contain information originating from
+previously published research resources.
+
+The AffLex-Provo project contributes computational
+preprocessing, integration, validation, and
+dataset construction procedures.
+
+Original contributions may be licensed separately
+where the contributors possess the necessary rights.
+
+Underlying third-party data remain subject to
+their applicable original licenses and
+redistribution permissions.
+
+See:
+
+[Analysis-Ready Dataset Licensing](AnalysisReadyDataLicense.md)
+
+[Original Dataset Citations](../01-Raw/MandatoryCitations.md)
+
+---
+
+## 12. Scholarly Citation
+
+Researchers using original AffLex-Provo contributions
+are encouraged to cite the corresponding publication.
+
+Gupta, M., & Christianson, K. (in press).
+
+Beyond word length, frequency, and predictability:
+Affective properties predict eye movement in reading.
+
+*Reading Research Quarterly.*
+
+DOI: Pending assignment.
+
+Researchers should additionally cite the original
+data sources relevant to their analyses.
+
+---
+
+## 13. Repository Information
+
+**Repository:** AffLex-Provo
+
+**Dataset Preparation:** Mohit Gupta
+
+**Institution:** University of Illinois Urbana-Champaign
+
+**Repository URL:**
+https://huggingface.co/datasets/mohitg2/AffLex-Provo
+
+**Documentation Version:** 1.0
+
+**Last Updated:** October 2026