How do I show statistical significance?

Add asterisks (* p<0.05, ** p<0.01) next to each Δ value, and a footnote explaining the test (e.g., paired bootstrap, B=1000).

PromptsCharts

ChartsText → ImageLow creditEN

Ablation Study Horizontal Bar Chart

Sorted horizontal bars showing the contribution of each component to overall accuracy.

When to use this prompt

For ablation tables in ML / NLP / vision papers where you remove one component at a time.

The prompt

A horizontal bar chart showing ablation study results.

Y-axis labels (top to bottom, sorted by accuracy descending):
- "Full model" (94.2%)
- "without data augmentation" (91.3%)
- "without attention" (89.1%)
- "without residual connections" (87.5%)
- "without pre-training" (82.7%)

X-axis: Accuracy (%), range 80-100, gridlines every 2.

Each bar:
- Color: navy
- Value label at the right end of the bar
- A vertical dashed reference line at the Full model accuracy (94.2%) to make deltas visible

Annotations:
- A small "Δ vs Full" column on the right showing -2.9 / -5.1 / -6.7 / -11.5 with red text for losses.

Style: clean academic chart, no chart junk, sans-serif, white background, restrained palette. Suitable for ML conference results sections.

Variations

Per-task panels

Replicate the chart as 4 small panels, one per benchmark task (e.g., MNLI, SST-2, SQuAD, CoLA). Stack panels vertically with a shared x-axis label.

Tips

Sort bars by performance descending. Unsorted ablation charts are hard to scan.
Always include a Δ column. Readers want to see absolute drops, not just bar lengths.
Keep x-axis to a tight range (80-100 for accuracy). Wide ranges crush the differences.