@brainpilot/skills 0.0.6

package/skills/02_Cross-Domain_Foundation/cogsci-visualization/references/plot-recipes.md

@@ -0,0 +1,709 @@
+# Visualization Recipes for Cognitive Science
+
+Concrete code recipes for common cognitive science figures. Each recipe specifies: when to use, tool (Python/R), key parameters, APA formatting tips, and common mistakes.
+
+---
+
+## Recipe 1: RT Distribution -- Raincloud Plot
+
+### When to Use
+- Visualizing reaction time distributions across conditions
+- Any continuous dependent variable where distribution shape matters
+- Replacing bar charts in manuscripts (Allen et al., 2019; Weissgerber et al., 2015)
+
+### R Code (ggrain)
+
+```r
+library(ggplot2)
+library(ggrain)
+
+p <- ggplot(d, aes(x = condition, y = RT, fill = condition, color = condition)) +
+ geom_rain(
+ rain.side = "l", # Density on left
+ point.args = list(size = 1, alpha = 0.3),
+ boxplot.args = list(width = 0.1, outlier.shape = NA),
+ violin.args = list(alpha = 0.5)
+ ) +
+ scale_fill_manual(values = c("#E69F00", "#56B4E9", "#009E73")) + # Okabe-Ito
+ scale_color_manual(values = c("#E69F00", "#56B4E9", "#009E73")) +
+ labs(x = "Condition", y = "Reaction time (ms)") +
+ theme_classic(base_size = 12) +
+ theme(
+ legend.position = "none",
+ axis.text = element_text(size = 10),
+ axis.title = element_text(size = 12)
+ )
+
+ggsave("rt_raincloud.tiff", p, width = 3.3, height = 3.3, dpi = 300, units = "in")
+```
+
+### Python Code (PtitPrince)
+
+```python
+import ptitprince as pt
+import matplotlib.pyplot as plt
+import matplotlib
+matplotlib.rcParams['font.family'] = 'Arial'
+matplotlib.rcParams['font.size'] = 10
+
+fig, ax = plt.subplots(figsize=(3.3, 3.3), dpi=300)
+
+pt.RainCloud(
+ x="condition", y="RT", data=d, ax=ax,
+ palette=["#E69F00", "#56B4E9", "#009E73"], # Okabe-Ito (Okabe & Ito, 2002)
+ bw=0.2, # Bandwidth for density estimation
+ width_viol=0.6,
+ point_size=2,
+ alpha=0.5,
+ orient="v",
+ move=0.2
+)
+
+ax.set_xlabel("Condition")
+ax.set_ylabel("Reaction time (ms)")
+ax.spines[['top', 'right']].set_visible(False)
+
+fig.tight_layout()
+fig.savefig("rt_raincloud.tiff", dpi=300, format='tiff')
+```
+
+### Key Parameters
+- **Bandwidth (bw)**: Controls smoothness of density. Too narrow = noisy; too wide = oversmoothed. Default (Silverman's rule) is usually acceptable.
+- **Point alpha**: **0.2-0.4** for large N (> 50 per condition); **0.5-0.8** for small N (< 30)
+- **Colors**: Use Okabe-Ito palette for colorblind safety (Okabe & Ito, 2002)
+
+### APA Formatting Tips
+- Width: **3.3 in** (single column) or **6.9 in** (double column)
+- Font: Arial or Helvetica, **10-12 pt** in final printed size
+- Caption: "Raincloud plot showing RT distributions by condition. Points represent individual trials/participants. Box plots show median and IQR. Density estimates show distribution shape."
+
+### Common Mistakes
+- Not jittering points sufficiently (creates a solid line instead of visible scatter)
+- Using bar charts instead because "that's what my advisor always uses"
+- Not specifying what individual points represent (trials? participant means?)
+
+---
+
+## Recipe 2: ERP Waveform Plot
+
+### When to Use
+- Comparing ERP waveforms across conditions at a specific electrode or ROI
+- Showing temporal dynamics of brain responses
+
+### Python Code (MNE-Python)
+
+```python
+import mne
+import matplotlib.pyplot as plt
+import matplotlib
+matplotlib.rcParams['font.family'] = 'Arial'
+matplotlib.rcParams['font.size'] = 10
+
+fig, ax = plt.subplots(figsize=(4.5, 3.0), dpi=300)
+
+# Plot evoked responses for each condition
+colors = {"congruent": "#0072B2", "incongruent": "#D55E00"} # Okabe-Ito subset
+linestyles = {"congruent": "-", "incongruent": "--"}
+
+for cond, evoked in evoked_dict.items():
+ times = evoked.times * 1000 # Convert to ms
+ data = evoked.data[channel_idx, :] * 1e6 # Convert to uV
+
+ ax.plot(times, data,
+ color=colors[cond], linestyle=linestyles[cond],
+ linewidth=1.5, label=cond)
+
+# Convention: negative up (traditional ERP; Luck, 2014)
+ax.invert_yaxis()
+
+# Mark stimulus onset
+ax.axvline(x=0, color='black', linestyle=':', linewidth=0.5)
+ax.axhline(y=0, color='black', linestyle='-', linewidth=0.5)
+
+# Shade component window (e.g., N400: 300-500 ms; Kutas & Federmeier, 2011)
+ax.axvspan(300, 500, alpha=0.1, color='gray', label='N400 window')
+
+# Shade baseline
+ax.axvspan(-200, 0, alpha=0.05, color='gray')
+
+ax.set_xlabel("Time (ms)")
+ax.set_ylabel("Amplitude (uV)")
+ax.set_xlim([-200, 800])
+ax.legend(loc='lower right', fontsize=8, frameon=False)
+ax.spines[['top', 'right']].set_visible(False)
+
+fig.tight_layout()
+fig.savefig("erp_waveform.tiff", dpi=300, format='tiff')
+```
+
+### R Code (ggplot2)
+
+```r
+library(ggplot2)
+
+p <- ggplot(erp_data, aes(x = time_ms, y = amplitude_uv, color = condition,
+ linetype = condition)) +
+ geom_line(linewidth = 0.8) +
+ geom_vline(xintercept = 0, linetype = "dotted", linewidth = 0.5) +
+ geom_hline(yintercept = 0, linewidth = 0.5) +
+ annotate("rect", xmin = 300, xmax = 500, ymin = -Inf, ymax = Inf,
+ alpha = 0.1, fill = "gray") +
+ scale_y_reverse() + # Negative up (Luck, 2014)
+ scale_color_manual(values = c("#0072B2", "#D55E00")) + # Okabe-Ito
+ labs(x = "Time (ms)", y = expression(paste("Amplitude (", mu, "V)"))) +
+ theme_classic(base_size = 10) +
+ theme(legend.position = c(0.8, 0.2), legend.background = element_blank())
+
+ggsave("erp_waveform.tiff", p, width = 4.5, height = 3.0, dpi = 300, units = "in")
+```
+
+### Key Parameters
+- **Line width**: **1.0-1.5 pt** for waveforms (Luck, 2014)
+- **Time range**: Typically **-200 to 800 ms** (adjust for paradigm)
+- **Y-axis direction**: Invert for negative-up convention
+- **Component shading**: Use named time windows from literature (e.g., N400: **300-500 ms**; Kutas & Federmeier, 2011)
+
+### APA Formatting Tips
+- Caption: "Grand-averaged ERP waveforms at electrode [Cz/Pz/etc.] for [condition names]. Negative is plotted upward. Shaded region indicates the [component] time window ([start]-[end] ms). N = [number]."
+- Always state polarity convention, electrode, and number of participants
+
+### Common Mistakes
+- Not stating polarity convention (negative up vs. down)
+- Using different y-axis scales across panels without noting it
+- Not showing stimulus onset marker
+- Plotting only group averages without any indication of variability
+
+---
+
+## Recipe 3: ERP Topographic Map
+
+### When to Use
+- Showing scalp distribution of ERP effect at a specific time window
+- Validating that the observed effect has the expected topography for a given component
+
+### Python Code (MNE-Python)
+
+```python
+import mne
+import matplotlib.pyplot as plt
+
+fig, axes = plt.subplots(1, 3, figsize=(6.9, 2.5), dpi=300)
+
+# Time windows for topographic maps
+time_windows = {
+ "N1 (80-120 ms)": (0.080, 0.120),
+ "N400 (300-500 ms)": (0.300, 0.500),
+ "P600 (500-800 ms)": (0.500, 0.800)
+}
+
+for ax, (label, (tmin, tmax)) in zip(axes, time_windows.items()):
+ # Compute mean amplitude in time window
+ evoked_diff = mne.combine_evoked(
+ [evoked_cond_a, evoked_cond_b], weights=[1, -1]
+ )
+
+ evoked_diff.plot_topomap(
+ times=[(tmin + tmax) / 2],
+ average=(tmax - tmin),
+ axes=ax,
+ colorbar=False,
+ cmap='RdBu_r', # Diverging colormap (Crameri et al., 2020)
+ vlim=(-3, 3), # Symmetric limits in uV
+ show=False
+ )
+ ax.set_title(label, fontsize=9)
+
+# Add shared colorbar
+sm = plt.cm.ScalarMappable(cmap='RdBu_r', norm=plt.Normalize(-3, 3))
+sm.set_array([])
+cbar = fig.colorbar(sm, ax=axes.tolist(), shrink=0.8)
+cbar.set_label('Amplitude (uV)', fontsize=9)
+
+fig.tight_layout()
+fig.savefig("erp_topomap.tiff", dpi=300, format='tiff')
+```
+
+### Key Parameters
+- **Colormap**: `RdBu_r` (diverging, blue-white-red, colorblind-safe; Crameri et al., 2020)
+- **Limits (vlim)**: Symmetric around zero; set to capture the range of the data
+- **Time windows**: Use established component windows from literature (see `../cogsci-statistics/references/common-analyses.md`)
+- **Sensor positions**: Show electrode markers on the map
+
+### APA Formatting Tips
+- Caption: "Topographic distribution of the [effect name] (condition A minus condition B) at [time windows]. Scale in microvolts. Electrode positions marked as dots."
+- Use consistent color scales across all topographic maps in the same figure
+
+### Common Mistakes
+- Using non-diverging colormap for difference data (obscures the zero-crossing)
+- Different color scales across panels within the same figure
+- Not specifying whether the map shows absolute values or condition differences
+- Omitting the color bar
+
+---
+
+## Recipe 4: fMRI Activation Map
+
+### When to Use
+- Showing brain regions activated by a contrast of interest
+- Whole-brain statistical results display
+
+### Python Code (nilearn)
+
+```python
+from nilearn import plotting
+import matplotlib.pyplot as plt
+
+# Option 1: Glass brain (whole-brain overview)
+fig = plt.figure(figsize=(6.9, 2.5), dpi=300)
+
+display = plotting.plot_glass_brain(
+ stat_map_img,
+ threshold=3.1, # z = 3.1 ~ p < 0.001 (Eklund et al., 2016)
+ colorbar=True,
+ cmap='inferno', # Perceptually uniform (Crameri et al., 2020)
+ plot_abs=False, # Show both positive and negative
+ display_mode='lyrz', # Left, top, right, lateral views
+ figure=fig,
+ title='Contrast: Task > Rest'
+)
+
+fig.savefig("fmri_glass_brain.tiff", dpi=300, format='tiff')
+
+# Option 2: Slice montage with anatomical underlay
+fig, axes = plt.subplots(1, 1, figsize=(6.9, 2.0), dpi=300)
+
+display = plotting.plot_stat_map(
+ stat_map_img,
+ bg_img=mni_template, # MNI152 T1 1mm template
+ threshold=3.1, # z = 3.1 (Eklund et al., 2016)
+ display_mode='z', # Axial slices
+ cut_coords=[-20, -5, 10, 25, 40, 55], # Show multiple levels
+ colorbar=True,
+ cmap='cold_hot', # Diverging for activation/deactivation
+ annotate=True, # Show coordinates
+ figure=fig
+)
+
+fig.savefig("fmri_slices.tiff", dpi=300, format='tiff')
+
+# Option 3: Surface projection
+fig, axes = plt.subplots(2, 2, figsize=(6.9, 5.0), dpi=300,
+ subplot_kw={'projection': '3d'})
+
+for ax, hemi, view in zip(axes.flat,
+ ['left', 'right', 'left', 'right'],
+ ['lateral', 'lateral', 'medial', 'medial']):
+ plotting.plot_surf_stat_map(
+ surf_mesh=f'fsaverage5_{hemi}',
+ stat_map=stat_map_surface,
+ hemi=hemi,
+ view=view,
+ threshold=3.1,
+ colorbar=True,
+ cmap='inferno',
+ axes=ax
+ )
+
+fig.savefig("fmri_surface.tiff", dpi=300, format='tiff')
+```
+
+### Key Parameters
+- **Threshold**: z = **3.1** corresponds to approximately p < 0.001 uncorrected (Eklund et al., 2016 -- minimum for cluster-forming threshold)
+- **Colormap**: `inferno` or `hot` for sequential; `cold_hot` or `RdBu_r` for diverging (Crameri et al., 2020)
+- **Background**: MNI152 T1 **1 mm** template for anatomical detail
+- **Display mode**: `'z'` for axial, `'x'` for sagittal, `'y'` for coronal, `'ortho'` for all three
+
+### APA Formatting Tips
+- Caption: "Statistical map showing regions with significantly greater activation for [contrast]. Voxel-level threshold: z > 3.1 (p < 0.001 uncorrected); cluster-level correction: p < 0.05 FWE. Coordinates in MNI space. Color bar indicates z-score."
+- Always report: threshold, correction method, coordinate space, template
+
+### Common Mistakes
+- Using jet/rainbow colormap (Borland & Taylor, 2007)
+- Not reporting the threshold in the figure or caption
+- Showing only a single slice that happens to contain the peak voxel
+- Using different thresholds in the figure vs. the statistics reported in the text
+
+---
+
+## Recipe 5: Behavioral Accuracy -- Dot Plot with Within-Subject CI
+
+### When to Use
+- Displaying condition means for accuracy or any proportion in a within-subjects design
+- When standard (between-subject) CIs would be misleading for within-subject comparisons
+
+### R Code
+
+```r
+library(ggplot2)
+library(Rmisc) # for summarySEwithin
+
+# Within-subject summary statistics (Morey, 2008)
+d_summary <- summarySEwithin(d, measurevar = "accuracy",
+ withinvars = "condition",
+ idvar = "subject")
+
+p <- ggplot() +
+ # Individual subject data (connected for within-subjects)
+ geom_line(data = d, aes(x = condition, y = accuracy, group = subject),
+ alpha = 0.2, color = "gray60") +
+ geom_point(data = d, aes(x = condition, y = accuracy),
+ alpha = 0.3, size = 1, color = "gray40",
+ position = position_jitter(width = 0.05)) +
+ # Group means with within-subject CI (Morey, 2008)
+ geom_pointrange(data = d_summary,
+ aes(x = condition, y = accuracy,
+ ymin = accuracy - ci, ymax = accuracy + ci),
+ size = 0.8, linewidth = 1.0,
+ color = "black") +
+ labs(x = "Condition", y = "Accuracy (proportion correct)") +
+ scale_y_continuous(limits = c(0, 1)) +
+ theme_classic(base_size = 11) +
+ theme(axis.text = element_text(size = 10))
+
+ggsave("accuracy_dotplot.tiff", p, width = 3.3, height = 3.3, dpi = 300, units = "in")
+```
+
+### Python Code
+
+```python
+import numpy as np
+import pandas as pd
+import matplotlib.pyplot as plt
+matplotlib.rcParams['font.family'] = 'Arial'
+
+# Within-subject CI (Cousineau, 2005; Morey, 2008 correction)
+def within_subject_ci(data, dv, subject, condition, ci=0.95):
+ """Compute within-subject confidence intervals with Morey (2008) correction."""
+ # Remove between-subject variance
+ subj_means = data.groupby(subject)[dv].transform('mean')
+ grand_mean = data[dv].mean()
+ data_norm = data.copy()
+ data_norm[dv] = data[dv] - subj_means + grand_mean
+
+ # Compute CI on normalized data
+ k = data[condition].nunique()
+ correction = np.sqrt(k / (k - 1)) # Morey (2008) correction factor
+
+ summary = data_norm.groupby(condition)[dv].agg(['mean', 'sem']).reset_index()
+ from scipy import stats
+ t_crit = stats.t.ppf((1 + ci) / 2, df=data[subject].nunique() - 1)
+ summary['ci'] = summary['sem'] * t_crit * correction
+ summary['mean_orig'] = data.groupby(condition)[dv].mean().values
+ return summary
+
+summary = within_subject_ci(d, 'accuracy', 'subject', 'condition')
+
+fig, ax = plt.subplots(figsize=(3.3, 3.3), dpi=300)
+
+# Individual subjects
+for subj in d['subject'].unique():
+ subj_data = d[d['subject'] == subj]
+ ax.plot(subj_data['condition'], subj_data['accuracy'],
+ color='gray', alpha=0.15, linewidth=0.5)
+
+# Group means with within-subject CI
+ax.errorbar(summary['condition'], summary['mean_orig'],
+ yerr=summary['ci'], fmt='ko', markersize=6,
+ linewidth=1.5, capsize=4)
+
+ax.set_xlabel("Condition")
+ax.set_ylabel("Accuracy (proportion correct)")
+ax.set_ylim([0, 1])
+ax.spines[['top', 'right']].set_visible(False)
+
+fig.tight_layout()
+fig.savefig("accuracy_dotplot.tiff", dpi=300, format='tiff')
+```
+
+### Key Parameters
+- **Morey (2008) correction**: Multiply SE by `sqrt(k / (k-1))` where k = number of conditions
+- **Individual data alpha**: **0.15-0.3** (thin, semi-transparent lines connecting within-subject data)
+- **Y-axis range**: [0, 1] for proportions
+
+### APA Formatting Tips
+- Caption: "Mean accuracy by condition. Error bars indicate within-subject 95% CI (Morey, 2008). Gray lines connect individual participants' data points. N = [number]."
+- Always specify that CIs are within-subject corrected
+
+### Common Mistakes
+- Using between-subject CIs for within-subject comparisons (overestimates variability; Loftus & Masson, 1994)
+- Not connecting individual data points in within-subject designs (connection shows paired structure)
+- Starting y-axis at 0.5 when accuracy is near ceiling (may obscure meaningful variation -- but state the break)
+
+---
+
+## Recipe 6: Group Comparison -- Cumming Estimation Plot (DABEST)
+
+### When to Use
+- Comparing two or more groups or conditions
+- Emphasizing effect size and uncertainty over p-values
+- "Estimation statistics" approach (Cumming, 2014; Ho et al., 2019)
+
+### Python Code (DABEST)
+
+```python
+import dabest
+
+# Gardner-Altman estimation plot (Ho et al., 2019)
+analysis = dabest.load(d, x="group", y="score",
+ idx=("control", "treatment"))
+
+fig = analysis.mean_diff.plot(
+ color_col="group",
+ custom_palette=["#56B4E9", "#E69F00"], # Okabe-Ito
+ raw_marker_size=4,
+ es_marker_size=8,
+ swarm_label="Score",
+ contrast_label="Mean difference",
+ fig_size=(4.5, 4.0),
+ dpi=300
+)
+
+fig.savefig("estimation_plot.tiff", dpi=300, format='tiff')
+```
+
+### R Code (dabestr)
+
+```r
+library(dabestr)
+
+result <- dabest(d, x = group, y = score,
+ idx = c("control", "treatment"),
+ paired = FALSE) %>%
+ mean_diff()
+
+plot(result,
+ color.column = group,
+ palette = c("control" = "#56B4E9", "treatment" = "#E69F00"),
+ rawplot.markersize = 2,
+ effsize.markersize = 4) +
+ theme_classic(base_size = 10)
+
+ggsave("estimation_plot.tiff", width = 4.5, height = 4.0, dpi = 300, units = "in")
+```
+
+### Key Parameters
+- **Bootstrap resamples**: **5000** (default in DABEST; Ho et al., 2019)
+- **Effect size**: Mean difference with 95% CI (default) or Cohen's d
+- **Individual data**: Shown as a swarm plot (left panel)
+
+### APA Formatting Tips
+- Caption: "Gardner-Altman estimation plot showing individual data points (left) and the mean difference with bootstrap 95% CI (right). N = [per group]. [Number] bootstrap resamples."
+- The estimation plot naturally communicates effect size and uncertainty
+
+### Common Mistakes
+- Not reporting the numerical effect size and CI in the text (the plot supplements but does not replace statistical reporting)
+- Using this for within-subject designs without the paired option
+
+---
+
+## Recipe 7: Time-Frequency Representation (TFR)
+
+### When to Use
+- Visualizing oscillatory power changes over time and frequency
+- Showing event-related spectral perturbation (ERSP)
+
+### Python Code (MNE-Python)
+
+```python
+import mne
+import matplotlib.pyplot as plt
+
+# Compute TFR using Morlet wavelets
+freqs = np.logspace(np.log10(4), np.log10(40), num=30) # 4-40 Hz log-spaced
+n_cycles = freqs / 2 # Frequency-dependent cycles (standard: freq / 2; Cohen, 2014)
+
+power = mne.time_frequency.tfr_morlet(
+ epochs, freqs=freqs, n_cycles=n_cycles,
+ return_itc=False, average=True
+)
+
+# Apply baseline correction (dB; Cohen, 2014)
+# Baseline: -500 to -200 ms (avoid edge artifacts)
+power.apply_baseline(baseline=(-0.5, -0.2), mode='logratio')
+
+fig, ax = plt.subplots(figsize=(4.5, 3.0), dpi=300)
+
+power.plot(
+ picks=[channel_idx],
+ axes=ax,
+ baseline=None, # Already applied
+ cmap='RdBu_r', # Diverging for dB change (Crameri et al., 2020)
+ vmin=-1.5, vmax=1.5, # Symmetric for diverging map
+ show=False
+)
+
+ax.set_xlabel("Time (ms)")
+ax.set_ylabel("Frequency (Hz)")
+ax.set_title("") # Remove auto-title; use figure caption instead
+
+fig.tight_layout()
+fig.savefig("tfr.tiff", dpi=300, format='tiff')
+```
+
+### Key Parameters
+- **Frequencies**: **4-40 Hz** (or task-specific range) on a log scale (Cohen, 2014)
+- **Wavelet cycles**: `n_cycles = freq / 2` provides balanced time-frequency resolution (Cohen, 2014)
+- **Baseline**: Apply before plotting; use **-500 to -200 ms** to avoid onset contamination (Cohen, 2014)
+- **Baseline mode**: `'logratio'` (decibel, dB) is standard for TFR (Cohen, 2014)
+- **Colormap**: `RdBu_r` (diverging, centered on zero; for baseline-corrected power)
+
+### APA Formatting Tips
+- Caption: "Time-frequency representation at electrode [name] showing event-related spectral power (dB relative to baseline [start]-[end] ms). Morlet wavelets with [n_cycles specification]. Warm colors indicate power increase; cool colors indicate power decrease."
+- Always report: baseline period, baseline correction method, wavelet parameters, frequency range
+
+### Common Mistakes
+- Not applying baseline correction (raw power values are not interpretable across frequencies)
+- Using a sequential colormap for baseline-corrected data (which has both positive and negative values)
+- Using a linear frequency axis when a log axis would better represent the data (low frequencies are overrepresented)
+- Using fixed n_cycles across all frequencies (poor time resolution at low frequencies or poor frequency resolution at high frequencies)
+
+---
+
+## Recipe 8: Correlation Matrix -- Hierarchically Clustered Heatmap
+
+### When to Use
+- Visualizing relationships among multiple variables (behavioral measures, neural measures, questionnaire subscales)
+- Identifying clusters of related variables
+
+### Python Code (seaborn)
+
+```python
+import seaborn as sns
+import matplotlib.pyplot as plt
+from scipy.cluster.hierarchy import linkage
+from scipy.spatial.distance import squareform
+
+# Compute correlation matrix
+corr_matrix = d[variable_list].corr()
+
+# Hierarchical clustering for ordering
+linkage_matrix = linkage(squareform(1 - corr_matrix.abs()), method='ward')
+
+fig, ax = plt.subplots(figsize=(5.0, 4.5), dpi=300)
+
+g = sns.clustermap(
+ corr_matrix,
+ row_linkage=linkage_matrix,
+ col_linkage=linkage_matrix,
+ cmap='RdBu_r', # Diverging (Crameri et al., 2020)
+ vmin=-1, vmax=1, # Full correlation range
+ center=0,
+ annot=True, # Show correlation values
+ fmt='.2f', # Two decimal places
+ linewidths=0.5,
+ figsize=(5.0, 4.5),
+ dendrogram_ratio=(0.15, 0.15),
+ cbar_kws={'label': "Pearson's r", 'shrink': 0.8}
+)
+
+g.ax_heatmap.set_xticklabels(g.ax_heatmap.get_xticklabels(),
+ fontsize=8, rotation=45, ha='right')
+g.ax_heatmap.set_yticklabels(g.ax_heatmap.get_yticklabels(),
+ fontsize=8, rotation=0)
+
+g.savefig("correlation_heatmap.tiff", dpi=300, format='tiff')
+```
+
+### R Code (corrplot)
+
+```r
+library(corrplot)
+
+# Compute correlation matrix
+cor_mat <- cor(d[, variables], use = "pairwise.complete.obs")
+
+# Hierarchically clustered heatmap
+tiff("correlation_heatmap.tiff", width = 5, height = 4.5, units = "in", res = 300)
+
+corrplot(cor_mat,
+ method = "color",
+ type = "lower",
+ order = "hclust", # Hierarchical clustering order
+ hclust.method = "ward.D2",
+ col = colorRampPalette(c("#2166AC", "white", "#B2182B"))(200), # RdBu
+ tl.col = "black",
+ tl.cex = 0.7,
+ number.cex = 0.6,
+ addCoef.col = "black", # Show correlation values
+ cl.cex = 0.7)
+
+dev.off()
+```
+
+### Key Parameters
+- **Colormap**: `RdBu_r` (diverging, centered on zero; Crameri et al., 2020)
+- **Range**: [-1, 1] for correlation matrices
+- **Clustering method**: Ward's method (`ward.D2`) for balanced clusters
+- **Annotation format**: Two decimal places for correlation values
+
+### APA Formatting Tips
+- Caption: "Hierarchically clustered correlation matrix of [variable description]. Color indicates Pearson's r. Variables are ordered by Ward's hierarchical clustering. N = [number]."
+- For many variables (> 15), consider showing only significant correlations or using a threshold
+
+### Common Mistakes
+- Not centering the colormap on zero (asymmetric colors misrepresent the sign of correlations)
+- Using an unclustered default ordering (misses the modular structure)
+- Annotating with too many decimal places (clutters the figure)
+- Not specifying the correlation type (Pearson vs. Spearman) in the caption
+
+---
+
+## General Tips Across All Recipes
+
+### Save Settings for Reproducibility
+
+```python
+# Python: Save figure parameters
+FIGURE_PARAMS = {
+ 'single_col_width': 3.3, # inches (typical journal single column)
+ 'double_col_width': 6.9, # inches (typical journal double column)
+ 'dpi': 300, # Minimum for print (APA 7th, 2020)
+ 'font_family': 'Arial',
+ 'font_size': 10, # Base font size for final figure
+ 'line_width': 1.0, # Default line width
+ 'palette': ["#E69F00", "#56B4E9", "#009E73",
+ "#F0E442", "#0072B2", "#D55E00",
+ "#CC79A7", "#000000"] # Okabe-Ito (Okabe & Ito, 2002)
+}
+```
+
+```r
+# R: Theme for consistent formatting
+theme_publication <- function(base_size = 10) {
+ theme_classic(base_size = base_size) +
+ theme(
+ text = element_text(family = "Arial"),
+ axis.text = element_text(size = base_size - 1),
+ axis.title = element_text(size = base_size),
+ legend.text = element_text(size = base_size - 1),
+ strip.text = element_text(size = base_size),
+ plot.title = element_text(size = base_size + 1, face = "bold")
+ )
+}
+```
+
+### Colorblind Testing Checklist
+
+Before submitting any figure:
+1. View in grayscale (does the information survive?)
+2. Run through a CVD simulator (deuteranopia and protanopia at minimum)
+3. Verify that color is not the ONLY encoding (use linetype, shape, labels as redundant cues)
+
+---
+
+## References
+
+- Allen, M., et al. (2019). Raincloud plots. *Wellcome Open Research*, 4, 63.
+- Borland, D., & Taylor, R. M. (2007). Rainbow color map (still) considered harmful. *IEEE CGA*, 27(2), 14-17.
+- Cohen, M. X. (2014). *Analyzing Neural Time Series Data*. MIT Press.
+- Cousineau, D. (2005). Confidence intervals in within-subject designs. *TQMP*, 1(1), 42-45.
+- Crameri, F., et al. (2020). The misuse of colour in science communication. *Nature Communications*, 11, 5444.
+- Cumming, G. (2014). The new statistics. *Psychological Science*, 25(1), 7-29.
+- Eklund, A., et al. (2016). Cluster failure. *PNAS*, 113(28), 7900-7905.
+- Ho, J., Tumkaya, T., Aryal, S., Choi, H., & Claridge-Chang, A. (2019). Moving beyond P values: Data analysis with estimation graphics. *Nature Methods*, 16, 565-566.
+- Kutas, M., & Federmeier, K. D. (2011). Thirty years and counting. *Annual Review of Psychology*, 62, 621-647.
+- Loftus, G. R., & Masson, M. E. J. (1994). Using confidence intervals in within-subject designs. *Psychonomic Bulletin & Review*, 1(4), 476-490.
+- Luck, S. J. (2014). *An Introduction to the Event-Related Potential Technique* (2nd ed.). MIT Press.
+- Morey, R. D. (2008). Confidence intervals from normalized data. *TQMP*, 4(2), 61-64.
+- Okabe, M., & Ito, K. (2002). Color universal design (CUD). J*Fly*.
+- Weissgerber, T. L., et al. (2015). Beyond bar and line graphs. *PLoS Biology*, 13(4), e1002128.