Michal KuderskiDATA ANALYST · B.S. STATISTICS
← Selected work

03 / Honors capstone · Research design & simulation

Design the study.
Distinguish the evidence.

Dreaming in Different Cultures explores a cross-cultural research question. The recovered materials document a proposed DreamBank analysis and a synthetic count-model demonstration; empirical cultural findings remain unverified.

R · simulation · Poisson GLM · research designNIU Honors capstone · May 2026Evidence review updated October 2026

The research question

How would we compare language in dream narratives?

My capstone proposed comparing American, Japanese, and Peruvian DreamBank narratives through text preparation, pronoun rates, topic exploration, and count models. It asks how to make language features comparable when report lengths and sample sizes differ.

The recovered code establishes a narrower completed demonstration: generating synthetic counts and fitting a length-adjusted Poisson model. It does not establish an observed cultural difference.

Demonstrated in source
Reproducible synthetic-data generation, a count model with a word-count offset, model-derived intervals, and diagnostic plotting.
Research design
A separate draft proposes DreamBank retrieval and LDA. That draft is incomplete; the newly recovered file uses mock topic weights rather than fitting LDA.
Evidence status
No frozen empirical feature table, verified observed cohort counts, empirical model outputs, permutation results, or fitted topic outputs were recovered.
1,600SYNTHETIC ROWS IN THE DEMONSTRATORNot a verified count of observed dream reports
1,000 / 300 / 300ASSIGNED GROUP SIZESSimulation inputs for USA, Japan, and Peru labels
Empirical validationINCOMPLETEThe original cultural conclusions are not validated here

Evidence status

The group difference is an input.

The recovered USE_THIS_CODE.Rmd sets a seed, simulates report lengths, and generates first-person counts with different rates assigned to the group labels. It does not import dream narratives or an observed feature table.

Values explicitly assigned in the recovered code; these are assumptions, not empirical estimates.
Group labelGenerated rowsImposed expected count per wordMeaning
USA1,0000.050Assigned higher rate
Japan3000.020Assigned lower rate
Peru3000.025Assigned lower rate

Interpretation: A fitted difference on these generated data reflects the simulation design. Its coefficients, intervals, and residual plots cannot validate a finding about actual American, Japanese, or Peruvian students.

Count-model demonstration

The code fits I_Count ~ Culture + offset(log(Total_Words)) with a Poisson family. The offset models counts relative to report length. This is a useful statistical example when clearly identified as synthetic.

Topic illustration

The object lda_mock combines manually selected words with runif() weights. It does not call an LDA fitting routine. Academic, Familial, and Nature & Travel are illustration labels, not recovered fitted topics.

Paper versus provenance

The final paper describes empirical DreamBank findings. Its figure titles, styling, and qualitative rate pattern correspond to the supplied simulation code. Without the empirical inputs and final run, those figures are treated as simulation-associated illustrations, not empirical evidence.

Validation still needed

Neither recovered CURE script supplies a permutation-test implementation or saved permutation output. The DreamBank draft also has acquisition, syntax, cohort-selection, and topic-summary dependencies that prevent an independent empirical run.

Inspect the paper’s figures as simulation-associated illustrations

All four figures below are illustrative and are not used to support cultural conclusions. They are retained unchanged from the source paper. The exact R-generated images and fitted numbers have not been regenerated in this environment.

SIMULATION-ASSOCIATED ILLUSTRATION · EMPIRICAL PROVENANCE UNVERIFIED

Synthetic-pronoun distribution illustration; not validated empirical evidence about cultures.
Synthetic-pronoun distribution illustrationThe supplied generator imposes different group rates. The visible ordering is consistent with those inputs; it is not evidence of an observed cultural ordering.

SIMULATION-ASSOCIATED ILLUSTRATION · EMPIRICAL PROVENANCE UNVERIFIED

Synthetic-model rate-ratio illustration; not validated empirical evidence about cultures.
Synthetic-model rate-ratio illustrationIntervals from a model fitted to synthetic counts describe that simulation. They are not empirical confidence intervals for DreamBank cohorts.

SIMULATION-ASSOCIATED ILLUSTRATION · EMPIRICAL PROVENANCE UNVERIFIED

Mock-topic keyword illustration; not validated empirical evidence about cultures.
Mock-topic keyword illustrationThe supplied code manually lists terms and draws random weights. The axis called “Significance within Theme” is not a p-value or a fitted topic probability.

SIMULATION-ASSOCIATED ILLUSTRATION · EMPIRICAL PROVENANCE UNVERIFIED

Synthetic-model residual illustration; not validated empirical evidence about cultures.
Synthetic-model residual illustrationThe recovered generator makes this kind of diagnostic from simulated counts. A residual plot cannot establish the provenance or adequacy of an empirical analysis.

Research design & completion requirements

What would make the empirical claim testable?

The proposal outlines text cleaning, LDA, pronoun comparisons, count modeling, and resampling. Those are planned analytical steps. The synthetic demonstration must remain separate from any future analysis of observed reports.

Current boundary: This page demonstrates research planning and a statistical simulation. It does not claim a completed empirical NLP study, cultural prediction model, or validated difference in self-focus.

Other supplied materials and duplicates

Honors SCP.Rmd is byte-identical to the earlier Updated_Code.Rmd. It is not new independent evidence. Both previously supplied DreamSAT workbooks contain zero project totals and division errors; their norm columns are reference values. They remain excluded as results evidence.

The paper’s “final” filename and the proposal’s intended sample sizes do not substitute for observed cohort counts or executed empirical outputs.

Inspect the materials

Source roles are explicit.

Paper and proposal review status

The supplied final paper remains a reporting artifact with unverified empirical claims; the proposal describes intended work. Their evidence roles are summarized in the provenance review. The newly supplied full PDFs are not republished.