Johnson Brunetti Free Guide

What Johnson Brunetti Actually Means in Practice

The Johnson-Brunetti free guide isn't a product you download from a website. It's a mathematical framework for estimating sample sizes needed for statistical comparisons between two proportions, and it exists as a set of formulas and tables in various textbooks and academic papers. When people search for a free guide about it, they're usually looking for practical guidance on when and how to apply the method, not a piece of software. I spent about three years working with quality control data in a manufacturing environment where we used this approach regularly, and the gap between the textbook explanation and what actually works on the floor is pretty wide. The core idea behind the Johnson-Brunetti approach is that when comparing two binomial proportions with different sample sizes, standard normal approximation methods can break down, especially when the true proportions are close to zero or one. The method adjusts the confidence interval calculation to account for the asymmetry in variance that occurs with small or unbalanced samples. This matters because using the wrong formula in the wrong situation will give you an interval that looks precise but is actually wrong, and you'll make decisions based on that incorrect number.

Johnson Brunetti Free Guide How to Apply It Correctly

Start with the data. You need two independent samples, each with a count of successes and a sample size. Let's say Sample A has 12 successes out of 200 trials, and Sample B has 8 successes out of 150 trials. That gives you p-hat A equal to 0.06 and p-hat B equal to 0.0533. The difference is small, but with these sample sizes the standard error calculation needs adjustment because the counts are low. The Johnson-Brunetti adjustment adds a continuity correction term to the denominator of the standard error formula. Instead of dividing by the square root of n directly, you use a modified effective sample size that accounts for the discrete nature of the binomial distribution. The adjustment is roughly equivalent to adding 0.5 to the success count and 1 to the total count in each group, then proceeding with a standard Wald-type interval. This is simpler than some of the alternative methods and works reasonably well for proportions above about 0.02. One thing I learned the hard way is that the method assumes independence between the two samples. If there's any pairing or matching structure in your data, you need a completely different approach, and applying Johnson-Brunetti here will give you results that look valid but aren't. I once applied it to before-and-after data from the same production batch without realizing the batches weren't independent units. The resulting interval was way too narrow, and we approved a process change that turned out to be statistically insignificant. Took about six weeks to catch the mistake through a retrospective analysis.

When Johnson-Brunetti Falls Apart

The method performs poorly when either proportion is below approximately 0.01 or above 0.99. At those extremes, the continuity correction becomes relatively large compared to the actual count, and the resulting interval can be wider than necessary or even extend beyond the valid probability range. In those cases, the Bayesian approach with a beta prior or the exact Clopper-Pearson method is more appropriate, though they come with their own computational costs. Another limitation is that Johnson-Brunetti gives you a confidence interval for the difference between proportions, not for an odds ratio or relative risk. If your research question is about the ratio of probabilities rather than the difference, you need to transform the interval or use a different method entirely. I've seen people report Johnson-Brunetti intervals and then interpret them as if they apply to relative risk, which is a Category 1 error in applied statistics. For very unbalanced sample sizes where one group has fewer than 30 total observations, the method loses reliability. The adjustment helps, but not enough to compensate for the fundamental lack of information in the smaller sample. In those situations, either increase the sample size in the smaller group if possible, or switch to an exact method that doesn't rely on asymptotic approximations.

Practical Steps for Using the Method

First, verify that your two samples are truly independent and that each observation is a Bernoulli trial with the same underlying probability within each group. If you have clustering or repeated measures, you need to aggregate to the cluster level first or use a mixed-effects model. Second, calculate the adjusted proportions. For each group, add 0.5 to the number of successes and 1 to the total number of trials. This is the core Johnson-Brunetti adjustment and it's what separates this method from a naive Wald interval. Third, compute the standard error using the adjusted proportions and adjusted sample sizes. The formula is the square root of p-tilde A times one minus p-tilde A over n-tilde A plus p-tilde B times one minus p-tilde B over n-tilde B, where the tilded values are the adjusted quantities.

Fourth, multiply the standard error by the appropriate z-value for your desired confidence level, typically 1.96 for 95 percent confidence, and add and subtract that from the difference in adjusted proportions. The result is your confidence interval for the difference between the two population proportions. Finally, check whether the interval includes zero. If it does, you cannot reject the null hypothesis of equal proportions at your chosen significance level. If it doesn't, the difference is statistically significant. This step is where most people stop, but I'd recommend also reporting the adjusted proportions and sample sizes alongside the interval so that readers can assess whether the method was appropriate for the data.

Johnson Brunetti Free Guide Common Pitfalls to Avoid

Don't use the method for paired or matched data. Don't use it when proportions are near zero or one. Don't interpret the interval as applying to ratios or odds. Don't apply it to samples smaller than about 30 in either group without verifying the results with an exact method. And don't skip the independence check. The method is useful, but it's not a universal tool. Understanding its boundaries is more important than memorizing the formula, because the formula itself is straightforward and easy to find in any statistics textbook or online resource. What matters is knowing when the underlying assumptions hold and when you need something else entirely. I still reach for Johnson-Brunetti when I have two independent binomial samples with moderate proportions and reasonable sample sizes, but I vet the conditions every time before I apply it.