咨询：基于ANOVA分析的五组有序受试群体样本量确定

Hey there! Let's walk through how to figure out how many participants you need per academic group for your skill level ANOVA. This balances statistical rigor with practical sense, so let's break it down step by step:

Key Parameters to Lock In First #

Before crunching numbers, you need to define these critical values to ground your calculation:

• Significance Level (α): The standard choice is 0.05—this is the probability of incorrectly concluding there’s a group difference when there actually isn’t one.
• Statistical Power: Aim for 0.8 or 0.9 (0.8 is the common baseline). This is the probability you’ll correctly detect a real group difference if it exists.
• Effect Size: This measures how large the group differences are likely to be. For ANOVA, we use Cohen's f:
  • Small effect: f = 0.1 (corresponds to η² ≈ 0.01)
  • Medium effect: f = 0.25 (η² ≈ 0.06)
  • Large effect: f = 0.4 (η² ≈ 0.14)
    If you have pilot data or similar studies, use those to estimate a realistic effect size. If not, start with a medium effect—it’s a safe, reasonable guess when you don’t have prior data.
• Number of Groups (k): You have 5 groups, so this is fixed at 5.
• Variance Estimate: If you have pre-test scores, use their standard deviation (SD) to estimate variance. Since your scores range 0-30, a reasonable guess for SD is 5-8 if you don’t have existing data.

The Easiest Tool for Calculations: G*Power #

G*Power is a free, widely used tool for power and sample size math—no coding required. Here’s how to use it for your case:

1. Open G*Power and select Tests > Means > ANOVA: Fixed effects, omnibus, one-way from the menu.
2. Plug in your parameters:
   • Number of groups (k): 5
   • α err prob: 0.05
   • Power (1-β err prob): 0.8 (or 0.9 if you want higher confidence)
   • Effect size f: Pick your chosen value (e.g., 0.25 for medium effect)
3. Click Calculate to get the total sample size, then divide by 5 to get per-group numbers.

Example Breakdowns #

• Medium effect (f=0.25), α=0.05, power=0.8: Total sample size ≈ 120 → 24 participants per group
• Large effect (f=0.4), same α and power: Total sample size ≈ 50 → 10 participants per group
• Small effect (f=0.1): You’d need ~600 total participants (120 per group)—this is rarely feasible, so if you suspect a small effect, think carefully about whether your study can justify that scale.

Practical Adjustments to Consider #

• Dropout Rate: If you expect some participants to drop out (no-shows, incomplete tests), add 10-20% to your calculated sample size. For example, if you need 24 per group, aim for 27-29 to account for attrition.
• Post Hoc Tests: If you plan to run pairwise comparisons (e.g., "do PhD students score higher than freshmen?"), your power for those tests will be lower than the omnibus ANOVA. Boost your sample size by 10-15% to compensate.
• Unequal Group Sizes: If you can’t get equal numbers (e.g., fewer honor students available), ANOVA can handle unequal sizes—but you’ll need a slightly larger total sample to maintain the same power. G*Power lets you input unequal group ratios if needed.

Final Quick Tips #

• If you can run a small pilot study first, use its data to refine your effect size estimate—it’ll make your sample size calculation way more accurate.
• Always document your assumptions (effect size, power, α) in your study plan or write-up so others can follow and replicate your reasoning.

内容的提问来源于stack exchange，提问作者ISOmetric