What Is Statistical Power and Why Should You Care?

Statistical power is one of those concepts that gets mentioned in every methods class but rarely sinks in until you need it for your dissertation proposal. Let's make it stick.

The Plain-Language Definition

Statistical power is the probability that your study will detect an effect if one truly exists. If your study has 80% power, there's an 80% chance you'll find a significant result when there really is something to find — and a 20% chance you'll miss it.

Missing a real effect is called a Type II error (or a false negative). Power is your defense against it.

An Analogy That Helps

Think of statistical power like a metal detector. A powerful metal detector picks up coins buried deep underground. A weak one only finds coins near the surface. The coins are there either way — the question is whether your tool is sensitive enough to detect them.

In research terms:

• The "coin" is the real effect in your population
• The "metal detector" is your study design and sample size
• "Power" is how deep your detector can reach

What Affects Power?

Four factors determine your study's statistical power:

1. Sample Size

This is the biggest lever you can pull. More participants = more power. With a larger sample, your estimates are more precise and your tests are more sensitive. This is why determining the right sample size is so important.

2. Effect Size

Larger effects are easier to detect. A tutoring program that boosts scores by 20 points is much easier to find statistically than one that boosts scores by 2 points. See our effect size guide for more on measuring and interpreting effect sizes.

3. Significance Level (Alpha)

A more lenient alpha (say, 0.10 instead of 0.05) gives you more power because you're casting a wider net. But it also increases the risk of false positives. Most researchers stick with 0.05 as the standard threshold.

4. Variability in Your Data

Less variability (noise) in your data makes effects easier to detect. You can reduce variability by using reliable measurement instruments, standardizing procedures, and choosing homogeneous samples. This is why using instruments with high reliability (Cronbach's alpha) matters.

Why 80% Is the Standard

Jacob Cohen recommended 80% power (0.80) as the minimum acceptable level for most research. This means accepting a 20% chance of missing a real effect. Some fields require 90% or 95% power for high-stakes decisions, but 80% is the standard for most dissertations.

What Happens With Low Power

Underpowered studies are problematic in several ways:

• You'll likely miss real effects. If power is 40%, you'll fail to find significant results more often than not, even when the effect is real.
• Non-significant results are uninterpretable. With low power, you can't tell whether you got a non-significant result because there's no effect or because your study was too weak to detect it.
• Significant results from low-powered studies tend to overestimate effect sizes. This is called the "winner's curse" — the only effects that reach significance in underpowered studies are those that happened to be unusually large in your sample.

How to Calculate Power

You have two options:

A Priori Power Analysis (Before Data Collection)

This is what your proposal requires. You specify your expected effect size, alpha level, and desired power, then calculate the minimum sample size needed. Use G*Power — it's free and handles most common tests.

Post-Hoc Power Analysis (After Data Collection)

Also called "observed power," this uses your actual results to calculate power after the fact. It's controversial among statisticians because observed power is mathematically determined by the p-value — it doesn't add new information. Some committees still request it, but many methodologists advise against it.

Power in Your Dissertation

Your proposal should include an a priori power analysis that justifies your target sample size. In your results chapter, discuss power when interpreting non-significant findings. If a result isn't significant, consider whether low power might be the reason. Report effect sizes regardless of significance — they give your reader useful information even when p-values don't reach the threshold.

The bottom line: power isn't an abstract concept. It directly determines whether your months of data collection will yield interpretable results. Plan for it in your proposal, and you'll save yourself frustration later.