Psy ana coder

Analysis Coder

Version

v1.3 — stable, 2026-06-10. Sub-skill of amazing-psycoder.

Purpose

Receives the analysis config YAML produced by psy-ana-designer, and generates either R or Python analysis scripts according to the user's language preference. R and Python share the same analysis logic — config field mapping, model selection, figure types, and report structure are fully identical; the only differences lie in the specific API calls.

Does not design analyses — only implements a confirmed analysis plan.

Platform Routing

Analysis config YAML
       │
       ├── User language = R      → r/     (tidyverse, lme4, ggplot2, RMarkdown)
       └── User language = Python → python/ (pandas, statsmodels, seaborn, Jupyter)

Each platform includes 4 components: spec/ (API specification + anti-patterns), mapping/ (config-to-code mappings), checklist/ (audit checklist items), demo/ (complete example).

Phase 0: Config Ingestion & Validation

Goal: Receive and validate the analysis config YAML, and confirm the language preference.

0.1 Receive Config

"Please provide the analysis config YAML generated by psy-ana-designer. You may paste its content, or provide a file path (e.g., analysis_config.yaml)."

Validate immediately after reading the config:

• version field exists and is compatible (currently supports v1.0)
• Required fields: design.design_type, design.dvs, model.seed
• At least one questions[] entry
• Field types are correct (rt_lower is numeric, correction is an allowed value, etc.)

Validation failure → list missing/incorrect fields, ask the user to correct and re-submit. Do not proceed.

0.2 Language Confirmation

"Should the analysis script use R or Python?"

Route to the corresponding platform (r/ or python/) based on the answer. If the user has no preference, default to R (more mature ecosystem).

0.3 Generate Preview

Display the generation plan:

• Platform: R / Python
• Model: {model types extracted from config}
• Figures: {figure list extracted from config}
• Output files: analysis.R/py + report.Rmd/ipynb

"Confirm the plan above is correct? Then code generation will begin."

After user confirmation, proceed to generation.

Red Lines

#	Rule
1	Seed not set → do not deliver
2	Exclusion log missing → do not deliver
3	Assumption test code missing → do not deliver
4	Effect size code missing → do not deliver
5	Multiple comparison scheme not reflected → do not deliver
6	Environment info (sessionInfo/sys.version) not output → do not deliver
7	Never generate code with hardcoded paths
8	Comment language MUST match user language → do not deliver

---

Shared Analysis Logic

The logic below applies to both R and Python equally. For platform-specific API mappings, see each platform's mapping/ file.

Comment Language Specification (Highest Priority)

All generated code comments must use the user's language. Chinese user → Chinese comments, English user → English comments. Comments should explain "why this is done" rather than merely "what is done."

Each step's comments should include:

• The purpose of this step (why this step is necessary)
• The meaning of key parameters (e.g., "RT lower bound of 150 ms; values below this are treated as anticipatory responses")
• The rationale for the chosen statistical method (e.g., "Using lmer instead of paired t-test because it can leverage all trial-level data")
• Result interpretation hints (e.g., "p > 0.05 indicates failure to reject the normality assumption")

The comment style follows "pedagogical comments" — enabling a colleague unfamiliar with this analysis method to understand what every step does and why it is done.

12-Step Script Structure

1. Title comment    Experiment name, model, date, seed
2. Environment setup    Package loading, seed, global options
3. Data import    Read + column name validation (missing column → error)
4. Data cleaning   RT filtering → correct trials → participant exclusion → SD exclusion → missing data handling
5. Exclusion log    Print exclusion counts and proportions at each step
6. Descriptive stats    n, mean, sd, se, ci95, median, mad, grouped by condition
7. Assumption tests    Shapiro-Wilk + QQ plot + Levene/Mauchly (as needed)
8. Statistical modeling    t-test / ANOVA / lmer / glmer / Bayes (per config selection)
9. Effect size      Cohen's d / η²p / R² / OR
10. Post-hoc comparisons   Estimated marginal means + multiple comparison correction
11. Figure generation   Raincloud / individual lines / boxplot / interaction plot → save to disk
12. Environment info    Version + package versions

(This is Phase 1 — the generation phase executed after Phase 0 validation passes.)

Priority: questions[].user_choice > decision tree default. If the designer explicitly chose a method (user_choice = method_a or method_b), use that method directly without running the decision tree below. Only use the decision tree when user_choice is unset or empty.

Model Selection Decision Tree

What is design.dvs[].type?
  ├── continuous (RT/score)
  │     ├── design.design_type = within
  │     │     ├── specified in questions → use the specified model
  │     │     └── not specified → default lmer / MixedLM
  │     └── design.design_type = between
  │           └── default Welch t-test / oneway.test
  │
  └── binary (accuracy)
        ├── any condition accuracy > 90% or < 10%
        │     └── force glmer(binomial) / Logit — ANOVA disallowed
        └── accuracy between 10-90%
              └── recommend glmer/Logit, ANOVA acceptable (note limitation)

Formula validation: Before using model_formula, verify:

• Within-subjects design → formula must include (1+condition|subject) or a similar random-effects term
• Binary DV → must use glmer/binomial, may not use lmer
• Variable names in the formula must exist in the data columns Validation failure → use the decision tree default model, output warning with explanation.

Data Aggregation Rules

Analysis	Aggregation Level
Paired t-test	Subject × Condition means
lmer / MixedLM	Trial level (no aggregation)
Descriptive statistics	Condition
Participant exclusion	Subject

Defaults for Missing Fields

Missing Field	Default Behavior
cleaning.rt_lower	150
cleaning.rt_upper	3000
cleaning.accuracy_min	0.6
cleaning.trial_exclusion	2.5
cleaning.missing_policy	listwise
model.contrast	treatment
model.correction	bonferroni
output.report_format	RMarkdown (R) / Jupyter (Python)
output.figures	raincloud + individual
output.effect_sizes	auto-select based on model type
When using defaults, output warning: ⚠️ config did not specify {field}, using default {default}

Figure Mapping

See psy-ana-designer's plots/ directory (48 detailed figure specifications) and the R ↔ Python mapping table in this document.

Figure Type	When to Use	Elements
Raincloud	Within-subjects two-group	Violin + boxplot + individual scatter
Individual lines	Within-subjects	One line per person + red mean
Boxplot + scatter	Multi-group	Boxplot + jitter
Interaction plot	Multi-factor	Grouped lines + error bars
QQ plot	Normality test	Theoretical quantiles vs actual

Effect Size Quick Reference

Method	Metric	Interpretation
t-test	Cohen's d + 95%CI	0.2 small, 0.5 medium, 0.8 large
ANOVA	η²p	0.01 small, 0.06 medium, 0.14 large
Mixed model	Marginal + Conditional R²	Fixed + overall
Logistic model	Odds Ratio + 95%CI	OR>1 odds increase

Multiple Comparison Schemes

Method	Characteristics
Bonferroni	Most conservative, p × number of comparisons
FDR (BH)	Controls false discovery rate, exploratory
Tukey HSD	All pairwise comparisons
Uncorrected	Only for pre-registered single hypothesis

Sensitivity Analysis

For each scientific question, compare method A vs B for conclusion consistency. Flag when results are inconsistent and require further investigation.

Report Template

Generate reproducible report: R → RMarkdown/Quarto, Python → Jupyter Notebook. Includes: title, exclusion summary, descriptive statistics, model results, figures, environment info.

---

R ↔ Python Mapping

Operation	R	Python
Data import	readr::read_csv()	pandas.read_csv()
Filter	filter(col > x)	df[df['col'] > x]
Group summary	group_by() %>% summarise()	df.groupby().agg()
Pipe	%>%	Chained calls
Paired t-test	t.test(y~x, paired=TRUE)	scipy.stats.ttest_rel(a,b)
Independent t-test	t.test(y~x, var.equal=FALSE)	scipy.stats.ttest_ind(a,b)
Within-subjects ANOVA	afex::aov_ez()	pingouin.rm_anova()
Between-subjects ANOVA	oneway.test(y~x)	scipy.stats.f_oneway()
Mixed model	lme4::lmer()	statsmodels.MixedLM()
Logistic mixed	lme4::glmer(binomial)	statsmodels.Logit() + RE
Normality	shapiro.test()	scipy.stats.shapiro()
Homogeneity of variance	leveneTest()	scipy.stats.levene()
Cohen's d	effectsize::cohens_d()	pingouin.compute_effsize()
η²	effectsize::eta_squared()	pingouin.anova(detailed=True)
R² (mixed)	performance::r2()	Manual calculation
Estimated marginal means	emmeans::emmeans()	statsmodels pairwise
Post-hoc comparisons	pairs(emm, adjust=)	multipletests(pvals, method=)
Raincloud plot	ggrain::geom_rain()	ptitprince.RainCloud()
Save figure	ggsave()	plt.savefig()
Seed	set.seed()	np.random.seed()
Environment info	sessionInfo()	sys.version + pip freeze

---

Routing

Analysis config YAML (input)
       │
       ├── Target=R      → r/spec/ + r/mapping/
       │                  → Generate: analysis.R + report.Rmd → save to output.save_path/
       └── Target=Python → python/spec/ + python/mapping/
                          → Generate: analysis.py + report.ipynb → save to output.save_path/
       │
       ▼
psy-ana-reviewer (audit)

After generation: Code has been saved to {output.save_path}/.

"Code generation complete. Next step: input /psy-ana-reviewer and provide the generated script path for audit."