1# Claude can read your data files directly
2claude "analyze the CSV file in ./data/experiment_results.csv"
3 
4# It can write scripts and save them
5claude "create a Python script to clean this dataset and save it as clean_data.py"
6 
7# It can even run your code and iterate on errors
8claude "run the analysis script and fix any errors"

1# Run analysis and get results
2claude "run a statistical analysis on my survey data and show me the results"
3 
4# Debug interactively
5claude "this script is throwing an error - help me fix it"

1# Get project-wide understanding
2claude "explain how the data pipeline in this project works"
3 
4# Find and modify across files
5claude "update all functions that use the old API format"

1# Ask Claude to help clean messy data
2# "Clean this survey data: handle missing values, standardize date formats,
3#  and remove duplicates"
4 
5import pandas as pd
6import numpy as np
7 
8def clean_survey_data(filepath):
9    """Clean and preprocess survey data."""
10    df = pd.read_csv(filepath)
11 
12    # Handle missing values
13    df['age'] = df['age'].fillna(df['age'].median())
14    df['response'] = df['response'].fillna('No Response')
15 
16    # Standardize dates
17    df['date'] = pd.to_datetime(df['date'], format='mixed')
18 
19    # Remove duplicates based on participant ID
20    df = df.drop_duplicates(subset=['participant_id'], keep='first')
21 
22    return df

1# "Run a regression analysis on my experiment data with proper diagnostics"
2 
3import pandas as pd
4import statsmodels.api as sm
5from scipy import stats
6 
7def analyze_experiment(df, dependent_var, independent_vars):
8    """Run regression with full diagnostics."""
9    X = df[independent_vars]
10    X = sm.add_constant(X)
11    y = df[dependent_var]
12 
13    model = sm.OLS(y, X).fit()
14 
15    # Print comprehensive results
16    print(model.summary())
17 
18    # Check assumptions
19    residuals = model.resid
20    print(f"\nNormality test (Shapiro-Wilk): {stats.shapiro(residuals)}")
21    print(f"Homoscedasticity test (Breusch-Pagan): {sm.stats.diagnostic.het_breuschpagan(residuals, X)}")
22 
23    return model

1# "Create publication-ready figures for my results"
2 
3import matplotlib.pyplot as plt
4import seaborn as sns
5 
6def create_publication_figure(df, x_var, y_var, group_var=None):
7    """Create a publication-ready scatter plot with regression line."""
8    plt.figure(figsize=(10, 6))
9 
10    if group_var:
11        sns.scatterplot(data=df, x=x_var, y=y_var, hue=group_var, alpha=0.7)
12    else:
13        sns.regplot(data=df, x=x_var, y=y_var, scatter_kws={'alpha':0.5})
14 
15    plt.xlabel(x_var.replace('_', ' ').title(), fontsize=12)
16    plt.ylabel(y_var.replace('_', ' ').title(), fontsize=12)
17    plt.title(f'{y_var.replace("_", " ").title()} vs {x_var.replace("_", " ").title()}',
18              fontsize=14, fontweight='bold')
19 
20    plt.tight_layout()
21    plt.savefig('figure.png', dpi=300, bbox_inches='tight')
22    plt.savefig('figure.pdf', bbox_inches='tight')  # Vector format for publications
23 
24    return plt.gcf()

1# "Build a classification model with cross-validation and feature importance"
2 
3from sklearn.model_selection import cross_val_score, train_test_split
4from sklearn.ensemble import RandomForestClassifier
5from sklearn.preprocessing import StandardScaler
6from sklearn.pipeline import Pipeline
7import pandas as pd
8 
9def build_classifier(df, target_col, feature_cols):
10    """Build and evaluate a classification pipeline."""
11    X = df[feature_cols]
12    y = df[target_col]
13 
14    # Create pipeline
15    pipeline = Pipeline([
16        ('scaler', StandardScaler()),
17        ('classifier', RandomForestClassifier(n_estimators=100, random_state=42))
18    ])
19 
20    # Cross-validation
21    cv_scores = cross_val_score(pipeline, X, y, cv=5, scoring='accuracy')
22    print(f"Cross-validation accuracy: {cv_scores.mean():.3f} (+/- {cv_scores.std()*2:.3f})")
23 
24    # Fit and get feature importance
25    pipeline.fit(X, y)
26    importance = pd.DataFrame({
27        'feature': feature_cols,
28        'importance': pipeline.named_steps['classifier'].feature_importances_
29    }).sort_values('importance', ascending=False)
30 
31    print("\nFeature Importance:")
32    print(importance.to_string(index=False))
33 
34    return pipeline, importance

1# "Clean and reshape my longitudinal study data"
2 
3library(tidyverse)
4 
5clean_longitudinal_data <- function(filepath) {
6  data <- read_csv(filepath)
7 
8  cleaned <- data %>%
9    # Handle missing values
10    mutate(across(where(is.numeric), ~replace_na(., median(., na.rm = TRUE)))) %>%
11    # Standardize variable names
12    rename_with(tolower) %>%
13    rename_with(~str_replace_all(., " ", "_")) %>%
14    # Convert to long format for repeated measures
15    pivot_longer(
16      cols = starts_with("time_"),
17      names_to = "timepoint",
18      values_to = "measurement",
19      names_prefix = "time_"
20    ) %>%
21    # Create factors
22    mutate(
23      timepoint = factor(timepoint, levels = c("1", "2", "3")),
24      group = factor(group)
25    )
26 
27  return(cleaned)
28}

1# "Run a mixed-effects model for my repeated measures design"
2 
3library(lme4)
4library(lmerTest)
5library(emmeans)
6 
7analyze_repeated_measures <- function(data) {
8  # Fit mixed-effects model
9  model <- lmer(
10    measurement ~ group * timepoint + (1 | participant_id),
11    data = data
12  )
13 
14  # Print summary with p-values
15  print(summary(model))
16 
17  # Post-hoc comparisons
18  emm <- emmeans(model, ~ group | timepoint)
19  print(pairs(emm, adjust = "tukey"))
20 
21  # Effect sizes
22  print(effectsize::eta_squared(model))
23 
24  return(model)
25}

1# "Create a figure showing group differences over time with error bars"
2 
3library(ggplot2)
4library(ggpubr)
5 
6create_longitudinal_plot <- function(data) {
7  # Calculate summary statistics
8  summary_data <- data %>%
9    group_by(group, timepoint) %>%
10    summarise(
11      mean = mean(measurement, na.rm = TRUE),
12      se = sd(measurement, na.rm = TRUE) / sqrt(n()),
13      .groups = "drop"
14    )
15 
16  # Create plot
17  p <- ggplot(summary_data, aes(x = timepoint, y = mean, color = group, group = group)) +
18    geom_line(linewidth = 1) +
19    geom_point(size = 3) +
20    geom_errorbar(aes(ymin = mean - se, ymax = mean + se), width = 0.1) +
21    labs(
22      x = "Time Point",
23      y = "Measurement (Mean +/- SE)",
24      color = "Group",
25      title = "Change in Measurement Over Time by Group"
26    ) +
27    theme_pubr() +
28    theme(
29      legend.position = "bottom",
30      plot.title = element_text(hjust = 0.5, face = "bold")
31    ) +
32    scale_color_brewer(palette = "Set1")
33 
34  # Save in multiple formats
35  ggsave("figure.png", p, width = 8, height = 6, dpi = 300)
36  ggsave("figure.pdf", p, width = 8, height = 6)
37 
38  return(p)
39}

1# Start a conversation with Claude Code
2claude
3 
4# Then interact naturally:
5> "I have survey responses in data/survey.csv. Can you:
6   1. Load and explore the data
7   2. Check for missing values and outliers
8   3. Run reliability analysis on the Likert scale items
9   4. Create summary statistics by demographic groups
10   5. Export a clean dataset for further analysis"

claude "I have 50 CSV files in the experiments/ folder, each from a different
participant. Combine them into a single dataframe, add a participant ID column
based on filename, and calculate summary statistics per participant."

claude "Read the methods section in paper_notes.md and help me implement
the same statistical analysis on my replication data in data/replication.csv"

claude "My analysis script analysis.py is throwing a memory error when
processing large files. Help me optimize it to handle datasets over 10GB."

1IMPORTANT: Claude Code sends your prompts and relevant file contents to
2Anthropic's servers for processing.
3 
4- Do NOT use with sensitive/identifiable participant data
5- Do NOT include API keys, passwords, or credentials in files
6- Consider using anonymized or synthetic data for development
7- Check your institution's policies on AI tool usage

1# Good: Specific and detailed
2claude "Load the CSV file data/experiment.csv, run a two-way ANOVA with
3factors 'treatment' and 'time', check assumptions, and create an interaction plot"
4 
5# Less effective: Vague
6claude "analyze my data"

1# First pass
2claude "create a basic scatter plot of x vs y"
3 
4# Refine
5claude "add a regression line, confidence interval, and make it publication-ready"
6 
7# Further refine
8claude "change the color scheme to colorblind-friendly and add axis labels"

1# Project: Effects of Sleep on Cognitive Performance
2 
3## Data Structure
4- data/raw/ - Original experiment data (do not modify)
5- data/processed/ - Cleaned datasets
6- analysis/ - R and Python scripts
7- figures/ - Output visualizations
8 
9## Analysis Standards
10- Use alpha = 0.05 for significance
11- Report effect sizes (Cohen's d or eta-squared)
12- All figures should be 300 DPI for publication
13 
14## Key Variables
15- dv: reaction_time (ms)
16- iv: sleep_condition (control, deprived, extended)
17- covariates: age, gender, caffeine_intake

1# Save your Claude Code sessions
2claude --save-conversation analysis_session_2024.md
3 
4# Or use git to track all changes
5git add -A && git commit -m "Analysis updates via Claude Code"