Overview

The Options tab allows you to customize how Prism performs multiple comparison corrections, displays results, and creates diagnostic graphs. These settings control the output format and presentation of your analysis results.

The Options tab is divided into several sections:

1.Multiple comparisons test - Choose how to correct for multiple testing

2.Multiple comparisons options - Fine-tune comparison settings

3.Graphing options - Choose which graphs to create

4.Output - Control formatting of results tables

Multiple Comparisons Test

This section determines how Prism adjusts P values when performing multiple comparisons to control for the increased risk of false positives.

Option 1: Correct for Multiple Comparisons Using Statistical Hypothesis Testing (Recommended)

What it does: Uses classical multiple comparison correction methods that control the family-wise error rate (FWER) - the probability of making at least one Type I error (false positive) across all comparisons.

When to use:

•This is the recommended default for most situations

•When you want to control the overall error rate across all comparisons

•Standard approach in most biological research

Available tests/correction methods:

Tukey

•Controls family-wise error rate

•Good statistical power

•Works well for balanced and slightly unbalanced designs

•Best for comparing all pairs of groups

•Most commonly used in biological research

Dunnett

•Designed specifically for comparing multiple treatments to one control

•More powerful than Tukey when you have a clear control group

•Use when you selected "Compare groups to control" in the Multiple Comparisons tab

Šídák

•Similar to Tukey but slightly more conservative

•Alternative to Tukey with similar properties

Holm-Šídák

•Step-down procedure

•Generally more powerful than Tukey

•Good general-purpose alternative

Bonferroni

•Very conservative (may miss real differences)

•Simple method: divides alpha by number of comparisons

•Use when you want to be extremely conservative

•Often too conservative for many biological applications

How to choose:

For the most part, you won't need to worry too much about which of these options to select as Prism will provide a sensible default depending on the type of comparisons you've specified (all pairwise comparisons, comparisons to a control, etc.). However, if you have a specific need to apply a different method, you're able to choose from any applicable methods for the specified comparisons simply by selecting your preferred method in the dropdown menu.

Option 2: Correct for Multiple Comparisons by Controlling the False Discovery Rate

What it does: Uses False Discovery Rate (FDR) methods to control the expected proportion of false positives among significant results, rather than controlling the family-wise error rate.

False Discovery Rate (FDR):

•The expected proportion of false positives among all significant results

•Less stringent than controlling family-wise error rate

•More powerful (better able to detect true differences)

•Accepts that some significant results may be false positives

When to use FDR methods:

•Exploratory studies with many comparisons

•When statistical power is a concern

•Genomics, proteomics, or other high-throughput studies

•When you're willing to accept a small proportion of false discoveries

•Discovery-phase research where false positives can be verified later

When NOT to use FDR methods:

•Confirmatory studies

•When false positives have serious consequences

•Small number of comparisons (family-wise control is appropriate)

•When your field expects traditional multiple comparison corrections

Three FDR Methods Available:

1.Two-stage step-up method of Benjamini, Krieger and Yekutieli (recommended)

•Default choice for FDR control

•An improved, more powerful version of the original FDR method that uses a two-stage procedure.

•More powerful than the original Benjamini-Hochberg method

•Adaptive procedure that estimates the proportion of true null hypotheses

•Generally the best FDR method to use

•Good balance of power and FDR control

2.Original FDR method of Benjamini and Hochberg

•The original and most widely known FDR method

•Conservative but reliable FDR control

•Well-established and widely cited

•Simpler procedure than the two-stage method

•Controls FDR under any dependency structure of tests

•May be used if your field specifically expects this method

3.Corrected method of Benjamini and Yekutieli (low power)

•Most conservative of the three FDR methods

•Controls FDR even when tests are positively correlated

•Lower statistical power than the other methods

•Provides valid FDR control under any correlation structure

Option 3: Don't Correct for Multiple Comparisons. Each Comparison Stands Alone.

What it does: Performs comparisons without any adjustment for multiple testing. Each comparison uses the unadjusted alpha level (default 0.05).

Test used: Fisher's LSD test (Least Significant Difference)

When this might be appropriate:

•You have very few comparisons (2-3)

•You have strong a priori hypotheses (planned comparisons)

•The overall ANOVA is significant and you're doing "protected" LSD

•You're doing preliminary exploratory analysis

When NOT to use:

•Many comparisons (inflated Type I error rate)

•Exploratory analysis with no specific hypotheses

•Publication-quality analysis (reviewers typically expect correction)

 

Multiple Comparisons Options

Swap Direction of Comparisons (A-B) vs. (B-A)

This option changes the direction of subtraction when computing mean differences for your multiple comparisons.

Report Multiplicity Adjusted P Value for Each Comparison

When checked, Prism reports the actual adjusted P value for each comparison in addition to the significance summary.

Family-Wise Alpha Threshold and Confidence Level

This value is also known as alpha, and controls the overall significance level for the family of comparisons as well as the confidence level for confidence intervals. The default (and most common) value is 0.05.

Graphing Options

This checkbox allows you to specify whether or not Prism should generate a graph displaying the confidence intervals for each individual comparison specified for the analysis.

Output

Choose how you want P values reported, and how many significant digits you need.