Skills

Method skills by domain

The 44 validated analysis methods Neutropic can run, grouped by domain. Ask for one by name inside an analysis or statistics request.

Method skills run inside a workflow — ask for the analysis you need ("run a CFA on these 12 items", "score SUS and compare versions") and the workflow calls the method with its assumption checks, effect sizes and figures. Every method has a reference page; switch any of them off under Customize → Skills.

Psychometrics & stats

SkillDescription
Descriptive statisticsMean, SD, median, range, skew/kurtosis, and Shapiro–Wilk normality per variable.
Reliability (Cronbach α · McDonald ω)Internal-consistency reliability: Cronbach's alpha (with CI), McDonald's omega, item-rest correlations.
Correlation analysisPairwise correlations (Pearson/Spearman) with p-values.
t-test (with assumptions)One- or two-sample t-test with normality & equal-variance checks, Cohen's d, and 95% CI.
One-way ANOVAOne-way ANOVA with Levene homogeneity test and η² effect size.
Exploratory Factor AnalysisEFA with KMO, Bartlett's test, rotated loadings, and variance explained.
Confirmatory Factor AnalysisCFA of a measurement model with fit indices (CFI, TLI, RMSEA), loadings, and a path diagram.
Structural Equation ModelingFit an SEM (lavaan-style) with fit indices, path estimates, and a rendered path diagram.
Item Response TheoryIRT item parameters (difficulty, discrimination) for dichotomous items — 2PL or Rasch.
Mixed-effects modelsLinear mixed-effects (random intercept) via formula — for nested/repeated-measures designs.
Power analysisStatistical power / required sample size for t-tests and one-way ANOVA.
Regression (linear · logistic)Linear (OLS) or logistic regression from a formula, with coefficients and fit.
Bayesian estimation (Bayes factor)Bayes factors (BF10) for t-tests and correlations, with evidence interpretation.
Assumption checks & test selectionCheck normality/homogeneity/outliers/sphericity for an analysis, and recommend the statistical test (with alternatives) from the design and the data.
Effect size estimationCohen's d / Hedges' g / η² / r with 95% CI and interpretation — from data or from summary statistics (means, SDs, ns).
Multiple-comparison correctionAdjust a family of p-values (Bonferroni, Holm, Benjamini–Hochberg FDR) and report which results survive.
Test selection & execution (parametric ↔ non-parametric)Design- and assumption-driven test choice with execution: compare_groups (auto Student/Welch/Mann-Whitney, paired t/Wilcoxon, ANOVA/Welch ANOVA/Kruskal, RM-ANOVA/Friedman with effect sizes, CIs, post hoc, table + figure), explicit non-parametric tests, chi-square family, and factorial/repeated/mixed/ANCOVA designs.

Multimodal biosignals

SkillDescription
ECG / HRV (heart-rate variability)Heart-rate variability from RR/IBI intervals — SDNN, RMSSD, pNN50, mean HR (+LF/HF).
EDA / GSR (skin conductance)Electrodermal activity — tonic level (SCL) and phasic responses (SCR count & amplitude).
Eye tracking · Gaze heatmapEye-tracking analysis — fixations (I-DT) and a rendered gaze heatmap image.
EEG (band power)EEG spectral band powers (delta/theta/alpha/beta/gamma) via Welch PSD.
Facial expression / AUSummarize facial Action Unit intensities and infer likely emotions (EMFACS heuristic).
Voice emotion (acoustic features)Acoustic-prosodic features from a voice clip — pitch, intensity, voice quality, pitch contour.
Respiration (RSP)Respiration rate and variability from a raw respiratory signal (belt / thermistor / derived).
Power spectral densityWelch PSD of any periodic physiological signal with peak frequency and custom band powers.
EEG analysis (preprocess · band power · topomap · ERP · ERD/ERS · SSVEP)Multichannel EEG from a table (channels = numeric columns in µV, time_s, optional marker): artifact/line-noise removal, band power with peak-alpha and asymmetry, scalp topography, event-related potentials, event-related (de)synchronization and SSVEP frequency detection — deterministic, with tables and figures.
Speech-to-text (local whisper)Transcribe an uploaded audio file or a video's sound track to text with timestamps (faster-whisper, offline).
Facial expression (image · video, measured)Measured facial Action-Unit approximation from an image or video via mediapipe blendshapes (52) → FACS AUs, EMFACS emotion heuristic, and head pose (yaw/pitch/roll). Per-frame CSV and AU time-series figure for video.
Body pose (image · video, measured)Measured body pose from an image or video via mediapipe PoseLandmarker (33 landmarks): joint angles, posture class (standing/sitting/bent), arm position, torso tilt, facing, movement per frame; overlay image and CSV.
Remote PPG from face video (rPPG)Heart rate from a face video without contact — face-ROI mean RGB → POS / CHROM / GREEN (published open algorithms) → 0.7–4 Hz band → HR (bpm), SNR, peak-based IBI/RMSSD; pulse + spectrum figure and trace CSV.
Acoustic features (formants · MFCC)Formant frequencies (F1–F3), MFCC statistics, and spectral descriptors from a voice clip.

HCI & usability

SkillDescription
SUS (System Usability Scale)Standard SUS scoring (0–100) with grade and acceptability interpretation.
NASA-TLX (workload)Raw NASA-TLX workload index from 6 subscales, plus per-subscale means.
A/B testingCompare a metric across two groups — proportion (conversion) or continuous, with effect size.
UEQ (User Experience Questionnaire)Standard UEQ scoring — 6 UX scales (−3..+3) plus pragmatic/hedonic quality.

Emotion engineering

SkillDescription
Semantic Differential (SD)SD profile — mean rating per bipolar adjective scale, optionally by stimulus/product.
Conjoint analysisRatings-based conjoint: part-worth utilities and attribute importance via OLS.
Quantification Theory IHayashi Quantification Type I — predict a quantitative criterion from categorical items; category scores, item importance (partial correlation), and multiple R.

Qualitative & NLP

SkillDescription
Text overview (open-ended responses)Response counts, length distribution, and top TF-IDF terms + bigrams for a free-text column.
Topic modeling (LDA · NMF)Discover latent topics in open-ended responses (LDA or NMF) with top words per topic.
Thematic clustering (assisted coding)Cluster open-ended responses (TF-IDF + KMeans) into candidate themes with representative quotes — support for inductive thematic analysis / open coding.
Sentiment analysis (VADER)Lexicon-based sentiment (positive/neutral/negative) for open-ended responses; English-optimized.

Data preparation

SkillDescription
Data inspection & validationMissing-value audit, variable-type detection, and rule-based dataset validation before analysis.
Dataset transformationReverse-score items, build composite scores, z-score, recode, filter, rename, drop — saved as a new dataset with the recipe recorded for reproducibility.

Domain guides