Two-Group Comparison

The classic two-group experiment: one treatment arm, one control arm, a continuous outcome.

Licklider executes Welch's t-test by default for independent two-group designs (robust to unequal variances). Normality assessment, advisory rank-based alternatives, effect size, and publication-ready figure generation follow — with disclosures required by high-impact journals.

STEP 1 — The Scenario

Researcher Profile

A cell biologist studying the effect of a novel HDAC inhibitor on tumor cell proliferation. They have performed an in vitro experiment with 8 biological replicates per group and need to compare cell viability (% of control) between treated and untreated cells.

Research Question

"Is there a statistically significant difference in cell viability between HDAC inhibitor–treated cells and DMSO-vehicle controls at 48 hours?"

Dataset

• 2 groups: Treatment (n=8), Control (n=8)
• Outcome: Cell viability (%), continuous
• Design: Independent groups, biological replicates (not technical replicates)

STEP 2 — Licklider Analysis Path

1. Data Upload & Profiling

Upload the CSV. Licklider automatically detects two columns: one categorical (Group), one continuous (Viability). It reports: n=8 per group, no missing values, no obvious data entry errors.

2. Data Contract Declaration

Licklider prompts: "What does each row represent?" → Answer: one biological replicate (one experimental well from one independent experiment). The observation unit is locked as biological replicate. The design is flagged as independent groups.

3. Normality Assessment

With n=8, Licklider runs the Shapiro-Wilk test on each group. Results: both groups pass (p>0.05). Licklider notes: "With n=8, Shapiro-Wilk has low power. Normality cannot be definitively confirmed. Welch's t-test is recommended as it is robust to mild non-normality with small n."

4. Test Selection

Licklider executes Welch's t-test (default over Student's t, as it does not assume equal variances). The rationale is disclosed: independent samples, Welch default, unequal variance not assumed. Mann-Whitney U remains available as an advisory rank-based alternative.

5. Effect Size Calculation

Cohen's d and its 95% CI are calculated alongside the p-value. The point estimate, CI, and t-statistic are all reported.

6. Figure Generation

Licklider proposes a Box Plot with strip overlay (each individual data point visible). Significance annotation is placed on a bracket above the two groups. The error bar type (showing median + IQR, or mean ± SD) is explicitly declared.

STEP 3 — Guards & Disclosures Activated

Effect Size + CI + N Mandatory (3.1.2)

The significance star annotation cannot be exported without Cohen's d, 95% CI, and n. Licklider inserts these automatically into the figure legend and the Methods text.

Error Bar Type Enforcement (3.1.1)

Before rendering, Licklider asks: "Should the box show median + IQR, or should the bars show mean ± SD?" The choice is recorded and disclosed in the figure legend.

Biological vs. Technical Replicate Guard (4.5.2)

Because n=8 is declared as biological replicates, the guard confirms no technical replicate averaging occurred. If technical replicates existed, they would need to be averaged before statistical testing.

N Disclosure (1.2.1)

Input N = 16, Analysis N = 16 (no rows excluded). Both numbers are disclosed in the figure legend and the Methods text.

One-Sided / Two-Sided Disclosure (3.1.3)

The test is two-sided (default). If the researcher specifies a directional hypothesis, Licklider requires a written rationale before proceeding with a one-sided test.

STEP 4 — Export Package

The researcher exports a complete package containing:

• Figure: Box plot with strip overlay, significance bracket, mean ± SD error bars — 300 dpi TIFF/PDF, journal-compatible color palette
• Statistical Report: Welch's t-test: t(12.4)=3.18, p=0.008; Cohen's d=1.59 (95% CI [0.42, 2.71]); Group means ± SD; n per group
• Methods Text (auto-generated): "Cell viability data were analyzed using a two-sided Welch's t-test. Data are presented as box plots showing median and IQR with individual data points overlaid. Effect size is reported as Cohen's d (95% CI). n=8 biological replicates per group. All analyses were performed using Licklider (version X.X)."
• Figure Legend (auto-generated): Includes test name, p-value, Cohen's d, 95% CI, error bar type, n
• Preprocessing Audit Log: No transformations applied; no rows excluded