Stats & Plotting Guide — Choosing the Right Graph and Statistical Test

1

What the Stats Suite does

The Stats Suite is AutoBlot Studio's built-in plotting and analysis tool. You can paste or import a data table, choose from a wide range of graph types, and run the appropriate statistical test — all in the browser, with results and figures you can export directly into a manuscript or presentation.

It's designed to mirror the workflow of dedicated statistics software (the kind labs typically pay for separately), but it stays connected to the rest of AutoBlot Studio: quantification results from the Analyzer can be sent straight into a stats session, and finished graphs can be saved to your dashboard or shared with your group.

Key principle

Choosing a graph and a statistical test are two separate decisions. The graph is how you show your data; the test is how you support a claim about it. Both should match your experimental design — not just look good.

2

Getting your data in

You can build a dataset in the Stats Suite in three ways:

Method	When to use it
Paste from a spreadsheet	Copy columns straight from Excel, Sheets, or Prism and paste into the data table — column headers become group labels automatically.
Send results from the Analyzer	Quantification values from a saved blot analysis can be pushed directly into a new stats session, so you don't have to retype anything.
Import a survival table	For Kaplan-Meier analysis, provide a Time column and an Event column (1 = event occurred, 0 = censored) for each group.

Tip — table shape matters

Whether your table is a simple list of groups, a grouped/two-factor table, a contingency table, or a survival table changes which graphs and tests are available. The Stats Suite detects the table type automatically and adjusts its recommendations.

3

Choosing the right graph type

The Stats Suite includes a wide range of chart types, grouped here by what they're best at showing:

Graph type	Best for
Bar / grouped bar / stacked bar	Comparing means across a small number of groups or sub-groups. Stacked bars work well for compositional data (parts of a whole).
Box & whisker / violin	Showing the full distribution — median, quartiles, spread, and (for violins) shape — rather than just a mean. Better than a bar chart when your data is skewed or you want to flag outliers.
Dot plot / before-after	Showing every individual data point. A before-after (paired slope) plot is ideal for repeated-measures or matched-sample designs, since it visually links each subject's two values.
Scatter / line / area	Continuous relationships — one variable plotted against another (e.g. concentration vs. response, or a time course). Lines connect ordered observations; areas emphasise cumulative or stacked trends.
Dose-response (4PL)	Fitting a four-parameter logistic curve to concentration-response data and extracting an IC₅₀ / EC₅₀.
Kaplan-Meier survival	Time-to-event data with censoring — survival curves with at-risk tables and log-rank comparison between groups.
Heatmap / correlation matrix	Visualising many pairwise relationships or a large grid of values at once (e.g. expression across many conditions, or correlations between many variables).
Forest plot / CI plot	Comparing effect sizes and their confidence intervals across multiple comparisons or studies at a glance.
Volcano plot	Screening many comparisons at once — typically effect size vs. statistical significance (e.g. omics-style data).
QQ plot / ECDF / Bland-Altman / ROC / PCA	Specialist diagnostic plots: checking normality (QQ), comparing distributions (ECDF), assessing agreement between two methods (Bland-Altman), evaluating a classifier (ROC), or reducing many variables to their main axes of variation (PCA).

Bar charts can hide the truth

A bar chart only shows the mean. Two datasets with identical means can have completely different spreads — one tightly clustered, one wildly variable — and a bar chart makes them look the same. Where possible, choose a graph that shows the underlying distribution (box, violin, dot plot) or overlay individual points on your bars.

4

Letting AutoBlot recommend a test for you

Once your data is in, the Stats Suite analyses its shape — number of groups, sample size, whether groups are paired, and whether the data looks normally distributed — and suggests an appropriate statistical test. You don't have to memorise a decision tree; the recommendation box explains why a particular test is suggested for your specific table.

Here's the logic behind the recommendations, so you understand what's happening under the hood:

Your data	Recommended test	Why
2 groups, normally distributed	Unpaired t-test (Welch's)	Compares two independent group means without assuming equal variances.
2 paired/matched groups	Paired t-test	More powerful than an unpaired test when each measurement has a natural partner (e.g. before/after in the same subject).
2 groups, small n or non-normal	Mann-Whitney U (unpaired) or Wilcoxon signed-rank (paired)	Non-parametric alternatives that compare ranks rather than means — robust when you can't assume a normal distribution.
3+ groups, normally distributed	One-way ANOVA + post-hoc	Tests whether any group differs, then a post-hoc test (e.g. Tukey's) identifies which pairs differ — without inflating your false-positive rate the way repeated t-tests would.
3+ groups, non-normal or small n	Kruskal-Wallis + post-hoc	The non-parametric counterpart to one-way ANOVA.
Repeated measures across 3+ conditions	Repeated-measures ANOVA or Friedman test	Accounts for the same subjects being measured multiple times, which increases statistical power and avoids treating dependent observations as independent.
Two factors (e.g. treatment × time)	Two-way ANOVA	Tests both main effects and whether the two factors interact.
Continuous x vs. continuous y	Pearson (linear, normal) or Spearman (monotonic, non-normal) correlation, or linear regression	Quantifies the strength and direction of a relationship, or fits a line to predict one variable from another.
Categorical / contingency table	Chi-square or Fisher's exact test (2×2, small samples)	Tests whether two categorical variables are associated.
Time-to-event with censoring	Kaplan-Meier + log-rank test	Estimates survival curves and compares them between groups, correctly handling subjects who didn't experience the event during the study.
Single group	Normality test (Shapiro-Wilk) first	Establishes whether your data meets the assumptions other tests rely on, before you compare it to anything else.

Outlier handling

The Stats Suite can optionally exclude outliers using Grubbs' test (α = 0.05) before running your analysis — useful for catching a single mis-pipetted well or corrupted measurement without manually hunting through your table.

5

Reading your results

Running a test produces a results panel with the test statistic, degrees of freedom, effect size, and exact p-value — plus, where relevant, post-hoc comparisons between specific group pairs (e.g. Tukey's after an ANOVA, or Dunn's after a Kruskal-Wallis).

What you'll typically see Test name · Statistic (e.g. t, F, H, U, χ², r) · Degrees of freedom · Effect size (e.g. Cohen's d, η²) · Exact p-value

Significant pairwise comparisons can be displayed directly on your graph as significance brackets and asterisks, so your figure communicates the result without needing a separate table.

Report the exact p-value and the effect size

"p < 0.05" or "ns" alone tells a reader very little. Reviewers increasingly expect the exact p-value and an effect size — a tiny, statistically significant difference may not be biologically meaningful, while a large effect that narrowly misses significance might still be worth reporting and following up.

6

Overlays, grouping, and comparing datasets

For bar, dot, scatter, line, box, and violin charts, you can overlay multiple datasets on the same axes — useful for comparing replicates, conditions, or experiments side by side without juggling separate figures.

Grouped tables (two-factor designs, such as treatment × genotype) are detected automatically and unlock grouped-bar layouts and two-way ANOVA, so the structure of your experiment is reflected in both the figure and the analysis.

Tip — keep biological replicates as the unit of analysis

When overlaying or grouping datasets, make sure each point you're comparing represents an independent biological replicate, not a technical replicate (e.g. triplicate wells from the same prep). Pooling technical replicates as if they were independent observations artificially inflates your sample size and can produce a falsely significant result.

7

Exporting a publication-ready figure

Once your graph looks right, you can save it to your dashboard, share it with your lab group, or export it directly for use in a manuscript, poster, or presentation — keeping the figure and the underlying data and statistics together.

A few things worth checking before you export:

Check	Why it matters
Axis labels and units are present	A figure should be interpretable without needing to read the methods section.
Individual data points are visible (n < ~10 per group)	Most journals now require or strongly prefer showing individual points rather than mean ± SEM alone — it lets readers judge consistency, not just the average.
Significance annotations match the test you actually ran	Brackets and asterisks should reflect the specific comparison and correction method used — not a generic "significant" marker.
Error bars are defined in your figure legend	State explicitly whether bars show SD, SEM, or a confidence interval — these look identical but mean very different things.

8

Common mistakes and how to avoid them

Mistake	How to avoid it
Running multiple t-tests across 3+ groups	Use one-way ANOVA (or Kruskal-Wallis) with a post-hoc correction — repeated t-tests inflate your false-positive rate.
Treating paired data as unpaired	If each measurement has a natural partner (same subject, same blot, before/after), use a paired test — it's more powerful and answers the right question.
Choosing a parametric test without checking normality	Use the normality recommendation first, especially with small n — if your data isn't normally distributed, switch to a non-parametric test (Mann-Whitney, Wilcoxon, Kruskal-Wallis).
Showing only mean ± error bars	Overlay individual points or switch to a dot/box/violin plot — a bar chart alone can hide bimodal data, outliers, or inconsistent replicates.
Pooling technical replicates as independent n	Each independent experiment (biological replicate) should contribute one data point to your statistical comparison — not every well or lane.
Ignoring censoring in time-to-event data	Use Kaplan-Meier rather than comparing raw "time to event" means — it correctly accounts for subjects who didn't experience the event during the observation window.
Reporting "significant"/"ns" with no numbers	Always report the exact p-value, the test used, and an effect size alongside any significance claim.

Stats & Plotting in AutoBlot Studio

What the Stats Suite does

Getting your data in

Choosing the right graph type

Letting AutoBlot recommend a test for you

Reading your results

Overlays, grouping, and comparing datasets

Exporting a publication-ready figure

Common mistakes and how to avoid them

Try the Stats Suite yourself