Statistics, in the sheet the numbers already live in.

Regression in Google Sheets

Regression is the one most people arrive looking for, so it goes first. Name the column you are trying to explain and the columns you think move it, and the report comes back with the coefficient on each input, its standard error, its t statistic and its p-value, a confidence interval either side of it, Multiple R, R Square, Adjusted R Square and the ANOVA table with the F test underneath.

It is written into a tab of your own spreadsheet, not into a panel you have to screenshot. Beside the table, on a fresh report tab, there is a short reading: a sentence giving this run’s R Square and its Significance F, which says whether the model as a whole beats a flat line against the alpha you set, and then a line on how to use the rest, that each coefficient is the change in the outcome per one-unit change in that input with the others held still, and that its P-value says whether that input is pulling any weight. Which input carries the outcome is a call you make from that table. The tool sets the evidence out for it rather than naming a winner.

Two more tools sit either side of it. Correlation gives you the whole matrix of which pairs move together before you decide what belongs in the model, and Logistic Regression takes over when the outcome is a yes or a no rather than an amount. That one is on the machine learning page.

Which of them do I actually need?

Two questions settle it, and both are about the design of the study rather than the numbers in it: how many groups are you comparing, and is each row in one group tied to one particular row in the other. Answer those and the shortlist above is usually down to one test, sometimes two.

Which statistical test should I use? is the chooser, one row per situation, and then the part that matters more: three cases where two defensible tests run on one dataset disagree. Nine matched weekdays where a rank test returns 0.0090 and a paired t-test returns 0.000281 on the same rows. Sixteen page loads where one test says 0.0104 and the other says 0.7253. And two t-tests that return the identical statistic and differ only in the degrees of freedom they will admit to.

What Google Sheets already does on its own

Some of these tools are a formula away, and a page that did not say so would be selling rather than explaining. T.TEST and F.TEST each return a probability, CORREL gives you a correlation, LINEST fits a regression and reports the standard error on every coefficient, and FREQUENCY, QUARTILE and RAND get you most of the way to a histogram, a set of outlier fences and a column of random draws.

Where is data analysis in Google Sheets? is the whole list, one row per analysis, with the built-in function named wherever there is one. Eight of them come back from a single call, twelve you can assemble out of built-ins and your own arithmetic, and eighteen have nothing behind them at all, including every ANOVA and every test that works on ranks instead of on a bell curve. That column was checked against Google’s published function list on 6 September 2026, and it carries the date because Google can move it.

Which t-test, and is the formula enough?

Three of the tools above are t-tests, and Google Sheets already has one of its own as a formula. Both make you choose between paired, equal-variance and Welch, and neither looks at your data to check that the choice fits. On one real pair of columns that choice moves the p-value by a factor of about 400.

Which of the three t-tests you need works through it on six datasets: what the built-in formula returns for each type, why Welch changes the degrees of freedom rather than the statistic, and the three things the report does not contain.

Three groups or more, and what a significant F leaves you

Four of the tools above take three groups or more: the three ANOVAs and Kruskal-Wallis, which asks the same question of the rank order instead of the averages. All four answer one narrow thing, whether the groups differ at all, and none of them says which pair is responsible. Eighteen exam scores across three revision methods come back at F 24.15 with a p-value of 0.00002, which settles that the methods are not interchangeable and names no winner, because naming one is not what the test does.

ANOVA in Google Sheets prints that run in full, both blocks and every column, and then does the part the report leaves to you: three pairwise t-tests judged against 0.05 divided by 3, so 0.0167 rather than 0.05, and why the threshold has to move once you ask three questions instead of one. It also runs the two-factor case, where twelve loaves separate what the recipe did from what the oven did and from the 0.5 cm interaction between them, and the case nobody writes up, where F comes back 2.42 on three crews and the honest answer is more days rather than a better test.

Fuzzy matching, for two lists that nearly agree

Fuzzy Lookup is in the same panel and is the one people are most surprised to find. Give it two columns from two exports and it scores every candidate pair, so “Globex Inc.” and “Globex Incorporated” end up on the same row instead of in two different totals.

It writes the matched pairs and their scores into a tab, which means you can read the near-misses and fix them by hand rather than trusting a join you cannot see.

Every result explains itself

A test that only prints a p-value is half a tool. Every run here ends with a card called What this means: the figure that matters, what it implies, and the caveat that goes with it. On the free plan that sentence is written by the add-on from the numbers the tool just computed, in your file, with nothing sent anywhere.

Pro adds an AI reading of the same result, written by Google’s Gemini API from the shape of the result alone, with a See what was sent panel showing the exact payload. Ratios, counts and cell references go; cell values, labels and formulas never do.

Free or Pro

Try it in your own sheet

1. Open Sortia in Google Sheets and choose Start from a template.
2. Pick one of the models above, and it loads with the inputs filled in.
3. Change the assumptions to fit your situation and press Run.

Never used Google Sheets? Start here goes the whole way, in seven steps, and assumes nothing.

Other methods: Monte Carlo · Decision trees · Schedule risk · Critical chain · Optimization under uncertainty · Machine learning · Forecasting · Optimization · What-if analysis