How to Master One of the Hardest Parts of the ASQ CSSGB Exam: Analyze Phase

Ask Certified Six Sigma Green Belt (CSSGB) candidates which part of the exam worries them most, and the answer is usually some version of “the statistics.” Many of those questions live in the Analyze Phase. It’s where you have to pick the right hypothesis test, compute a test statistic by hand with a basic calculator, read a p-value correctly, and know what a correlation coefficient does and doesn’t prove.

What makes this section hard isn’t just the math. It’s the decision-making around the math: choosing between a z-test and a t-test, deciding whether a result is one-tailed or two-tailed, and phrasing your conclusion the way statisticians do.

This guide breaks down what the Analyze Phase covers, explains the core concepts with the formulas you need, walks through worked examples, and gives you a focused plan to turn this section into a strength.

What Does the Analyze Phase Cover on the CSSGB Exam?

Under the 2022 CSSGB Body of Knowledge, the Analyze Phase is Section IV and carries 18 of the 100 scored questions. It includes three main areas:

Exploratory data analysis: Multi-vari studies, correlation, and simple linear regression.
Hypothesis testing: The basics (hypotheses, significance, errors, power, and p-values) and tests for means, variances, and proportions.
Additional analysis methods: Gap analysis and root cause analysis.

Because Analyze builds directly on the Measure Phase’s probability, distributions, and descriptive statistics, weaknesses there show up here too.

What Core Concepts Do You Need to Know?

Practical Problem to Statistical Problem

In Analyze, you translate a practical question (“Did the new supplier change our fill weights?”) into a statistical one. The null hypothesis (H0) states that there’s no difference or no effect, such as μ = 500 mL. The alternative hypothesis (Ha) is what you’re looking for evidence of: μ ≠ 500 (two-tailed), μ > 500, or μ < 500 (one-tailed). The equals sign always belongs in the null.

Errors, Significance, and Power

Type I error (α): Rejecting a true null hypothesis, a “false alarm.” The significance level α, often 0.05, is the probability of this error you’re willing to accept.
Type II error (β): Failing to reject a false null hypothesis, a “missed signal.”
Power (1 – β): The probability of detecting a real difference. Power increases with larger sample sizes, larger true differences, less variation, and a larger α.

The p-Value and the Decision Rule

The p-value is the probability of getting a result at least as extreme as yours if the null hypothesis were true. The rule is simple: if p < α, reject H0. If p ≥ α, you fail to reject H0. You never “accept” the null. You only conclude there isn’t enough evidence to reject it.

Tests for Means

One-sample z-test: Use when the population standard deviation (σ) is known. z = (x̄ – μ0) / (σ / √n).
One-sample t-test: Use when σ is unknown, and you estimate it with the sample standard deviation (s). t = (x̄ – μ0) / (s / √n), with n – 1 degrees of freedom.
Two-sample t-test: Compares the means of two independent groups. Use the pooled version if variances can be assumed equal, and the unpooled version if not.
Paired t-test: Use when the same units are measured twice, such as before and after a change. You test whether the mean of the differences equals zero.

Tests for Variances

Chi-square test for one variance: χ² = (n – 1)s² / σ0², with n – 1 degrees of freedom.
F-test for two variances: F = s1² / s2², usually with the larger variance on top, using n1 – 1 and n2 – 1 degrees of freedom. Variance tests matter in Six Sigma because reducing variation is often the whole goal.

Tests for Proportions

For a one-sample proportion test, z = (p̂ – p0) / √(p0(1 – p0) / n), where p̂ is the sample proportion and p0 is the hypothesized proportion. This normal approximation works when np0 and n(1 – p0) are both large enough, commonly at least 5.

Correlation and Regression

The correlation coefficient (r) ranges from -1 to +1 and measures the strength and direction of a linear relationship. Simple linear regression fits a line, ŷ = b0 + b1x, where b1 is the slope. The coefficient of determination, R², equals r² in simple linear regression and tells you the proportion of variation in y explained by x. For example, an r of 0.8 means R² = 0.64, so 64% of the variation is explained.

Multi-Vari Studies and Root Cause Analysis

A multi-vari study is a graphical way to see whether variation is mostly positional (within a part or across fixtures), cyclical (part to part over a short run), or temporal (over time). It narrows the search before you run formal tests. Root cause analysis tools, such as the 5 Whys, cause-and-effect (fishbone) diagrams, and Pareto charts, help you move from symptoms to verified causes. Gap analysis compares current performance with the desired state.

What Are the Most Common Analyze Phase Traps?

Using z when you should use t. If the question gives you a sample standard deviation and doesn’t state σ, you almost always need the t distribution, especially with small samples.
Mixing up one-tailed and two-tailed tests. “Is it different?” is two-tailed. “Has it improved?” or “Did it increase?” is one-tailed. This changes the critical value and can flip your decision.
“Accepting” the null. Answer choices that say you’ve proven no difference are wrong. The correct phrasing is “fail to reject.”
Treating paired data as independent. Before-and-after measurements on the same units call for a paired test, which removes unit-to-unit variation.
Confusing correlation with causation. A strong r shows association, not proof that x causes y. Confirming cause usually requires a designed experiment in Improve.
Confusing statistical and practical significance. With a huge sample, a tiny difference can be statistically significant but not worth acting on.
Forgetting that r only captures linear relationships. A strong curved relationship can produce an r near zero. Always picture the scatter plot.

How Do You Reason Through Sample Analyze Phase Questions?

Scenario 1: Is the Filling Process Off Target?

Consider a question where a filling line targets 500 mL. A Green Belt samples 25 bottles and finds a mean of 502 mL with a sample standard deviation of 5 mL. At α = 0.05, is there evidence that the process mean differs from the target?

Start with the setup. H0: μ = 500 and Ha: μ ≠ 500, so this is two-tailed. σ is unknown, so use a one-sample t-test. t = (502 – 500) / (5 / √25) = 2 / 1 = 2.0. With 24 degrees of freedom, the two-tailed critical value at α = 0.05 is about 2.064. Because 2.0 is less than 2.064, you fail to reject H0. There isn’t enough evidence that the mean differs from 500 mL.

Notice the trap. If you used the z critical value of 1.96, you’d wrongly reject the null. That’s exactly the kind of distractor the exam likes to include. A sharp candidate might also note that a larger sample would increase the test’s power to detect a 2 mL shift if it’s real.

Scenario 2: What Does a Strong Correlation Tell You?

Now imagine a team finds a correlation of r = 0.85 between oven temperature and the number of surface defects per batch. The manager asks whether the team should immediately lower the oven temperature.

First, R² = 0.85² ≈ 0.72, so about 72% of the variation in defects is associated with temperature. That’s a strong lead, but it isn’t proof of cause. Another factor, such as humidity that rises along with oven temperature, could be driving both.

The best answer typically recommends confirming the relationship, for example with a designed experiment that changes temperature while controlling other factors, before rolling out a process change. That reasoning connects Analyze directly to Improve.

How Does the Analyze Phase Connect to Other Sections?

Define Phase: The problem statement and CTQs from the charter determine what you’re testing and what difference matters in practice.
Measure Phase: Hypothesis tests rely on distributions, the central limit theorem, and descriptive statistics. They also rely on trustworthy data, which is why MSA comes first. Capability indices such as Cp and Cpk often frame the size of the gap that Analyze must explain.
Improve Phase: Analyze identifies likely causes, and DOE confirms them and finds the best settings.
Control Phase: Control charts provide continuous hypothesis testing. An out-of-control signal is like rejecting the null hypothesis that the process is stable, and the Type I and Type II error trade-offs apply there too.

A Focused Study Plan for the Analyze Phase

Review Measure first. Make sure you’re comfortable with the normal, t, chi-square, and F distributions and with standard error (s / √n) before tackling tests.
Build a test-selection chart. Make a one-page decision tree that asks: means, variances, or proportions? One sample or two? Paired or independent? σ known or unknown? Put it in your bound notes for exam day.
Drill by hand. Work 20 to 30 test-statistic problems with a basic calculator and printed tables, since you won’t have statistical software on the exam.
Write your conclusions in words. For each problem, state the decision (“reject” or “fail to reject”) and what it means for the process.
Use targeted practice. In Pocket Prep, use Build Your Own Quiz to isolate the Analyze Phase, read every explanation (even for questions you get right), and revisit errors a few days later with the Missed Questions quiz.
Mix it back in. Once your scores stabilize, return to mixed Timed Quiz sessions so you can spot Analyze-style reasoning in Measure and Control questions too.

Start Preparing for the ASQ CSSGB Exam With Pocket Prep

Hypothesis tests get much less intimidating after you’ve worked through enough realistic problems. Pocket Prep’s ASQ CSSGB practice questions give you 1,000 exam-style questions across every BoK section, each with a detailed explanation that shows the setup and the math. Pair Build Your Own Quiz with Level Up to build from fundamentals to exam-level difficulty. With steady practice, the Analyze Phase can become one of your strongest sections.