The output header says n = 167. Your dataset has 214 rows. Somewhere between opening the file and reading the coefficient table, 47 nurses left the study, and nothing on the screen asked whether that was all right.
That gap is listwise deletion. Most statistics software treats it as the default: if a case is missing a value on any variable in the model, the entire case drops out. Sometimes that is the right call. More often it is a decision made by a setting rather than by the person whose name goes on the manuscript.
The Decision Nobody Voted On
Missing data handling is a claim about who counts in your study, and it belongs in the method section next to your sampling plan. Three questions settle it. How much is missing. Where it sits. Why it is gone.
The third question does the most work. If item 12 asks about household income and respondents over 55 skip it more often than younger ones, the missingness carries information about the very people you are trying to describe. Listwise deletion then removes a group in a patterned way, and the coefficient you report describes a sample you never designed. Power goes with it: 214 down to 167 costs roughly a fifth of the cases and widens every confidence interval in the table.
None of this makes listwise deletion wrong. With 3 percent missing, scattered at random across variables, it is fine and simple to defend. The failure mode is silence. Nobody looked, so nobody knows which situation the study is actually in.
What the Mac Changes
Multiple imputation is not one analysis. It is m copies of the dataset, each filled in with a different plausible set of values, each analyzed separately, with the results pooled by Rubin's rules so the extra uncertainty stays visible in the standard errors. Twenty copies of a 214-case dataset used to mean starting the job and going to find coffee.
Apple Silicon changed the arithmetic of that wait. A full imputation and pooling run on a dataset this size finishes in seconds on a current Mac, locally, with no server in the loop. The practical consequence is not speed for its own sake. It is that comparing your options gets cheap enough that you actually do it: fit the model on complete cases, fit it again on imputed data, and see whether your conclusion depends on which one you picked. That comparison is the strongest sentence you can put in a limitations paragraph.
Seeing the Sample Before You Lose It
Quanta, our statistics app for the Mac, puts the missing values in front of you before the model runs. Load the CSV and the variable list shows how many values each variable is missing and what share of the sample that represents. Ask for a regression and a line reports how many complete cases the model would use, so the drop from 214 to 167 is something you read in advance rather than something you notice afterward in an output header.
From there the choice is explicit. Complete cases, pairwise, or multiple imputation with the number of imputations you set and pooling handled by Rubin's rules. Whatever you pick, the report carries it: the method, the number of imputations, and the n your estimates rest on, in an APA 7 table you can paste into the manuscript. The engine is written in Swift for Apple Silicon and checked against NIST reference datasets, and the whole run stays on your machine, which matters when the rows are patients or students.
The Part That Stays Yours
Quanta does not think for you. It removes the translation between your question and your evidence, and that is the only thing it removes. Software can show you that income is missing for 31 percent of respondents over 55. No software can tell you whether those nurses skipped the item because of how it was worded, because of what the answer might reveal to an employer, or because it sat on page four of a survey they were tired of. That reading takes what you know about your population, and it is what makes the method section defensible.
Imputation deserves an honest label too. It does not recover information that was never collected. What it does is carry the uncertainty of the missing values forward into the estimates instead of pretending the absent cases never existed. Better than a silent deletion, and still not the same as having the data.
What you get back is a minute at the start of the analysis instead of a hedge at the end. You open the file, you see 47 cases at risk and where they cluster, you choose a method on purpose, and the report records the choice. When a chair asks on Thursday why n dropped, the answer is already in the output. Six months later, when reviewer 2 asks the same question, you are not reopening the file to reconstruct what the software did in August.
The missing data was always a decision. Now it is one you made.
Quanta is available for the Mac at relicheck.com and on the Mac App Store, with a free 30-day trial.