Google Sheets add-on

Statbench

The statistics tools from Excel's Analysis ToolPak, working in Google Sheets™. Pick a tool, choose your data, and Statbench writes the results to a new sheet in the same tables, labels and order that Excel and XLMiner produce, so your output matches the course handout.

The ten core tools, free

Descriptive Statistics, Histogram, t-Tests (paired, equal and unequal variances), ANOVA: Single Factor, Regression, Correlation, Covariance, Moving Average, Sampling, and Rank and Percentile.

Familiar dialogs

Input range, grouped by columns or rows, labels in first row, alpha, confidence level, output range or new sheet: the same options as the desktop ToolPak, in the same words.

Pro: what-if and write-ups

Goal Seek, Data Tables and a Solver for linear and integer models, plus the F-test, z-test, two-factor ANOVA, exponential smoothing and plain-English and APA-style write-ups.

Getting started

Open a spreadsheet and choose Extensions → Statbench → Open Statbench.
Pick a tool from the list. Select your data in the sheet and press the selection button next to the range box, or type a range such as A1:C25.
Tick Labels in first row if your columns have headings, set any options, and press Run analysis.
The results appear in a new sheet named after the tool, or at the cell you chose under Output range.

Each tool remembers its settings for the spreadsheet. Extensions → Statbench → Run the last analysis again repeats the last one after you change the data.

The tools

Descriptive Statistics: mean, standard error, median, mode, standard deviation, sample variance, kurtosis, skewness, range, minimum, maximum, sum and count, plus the confidence level for the mean and the kth largest and smallest values.
Histogram: counts for each bin (a value goes in the first bin it's less than or equal to) with "More" for anything above the last bin. Leave the bin range blank for evenly spaced bins. Optional Pareto order, cumulative percentages and a chart.
t-Tests: paired two sample for means, two-sample assuming equal variances, and two-sample assuming unequal variances (Welch's test). Like Excel, the unequal-variances test rounds the degrees of freedom to a whole number by default; tick Exact df for the unrounded test.
ANOVA: Single Factor: the SUMMARY table and the ANOVA table with the P-value and F crit.
Regression: regression statistics, the ANOVA table and the coefficients with standard errors, t statistics, P-values and confidence limits. Optional residuals, standardised residuals and normal probability output, and a constant of zero.
Correlation and Covariance: matrices for every pair of columns. Covariance uses the population formula, as Excel's ToolPak does.
Moving Average: each average sits beside the last value it covers, with optional standard errors.
Sampling: every nth value, or a random sample with replacement. Enter a seed to get the same sample again.
Rank and Percentile: each value's rank (ties share the best rank) and percentile, largest first.

Pro

Goal Seek: changes one cell until a formula reaches the value you want. Press Undo to put the cell back.
Data Table: recalculates your model for each value in a list (or each pair of values from two lists) and writes the results as a table.
Solver: finds the values of your variable cells that maximise or minimise an objective, or make it equal a value, subject to constraints (≤, ≥, =, whole number, 0 or 1). Statbench checks the model is linear first; non-linear models aren't supported.
F-Test Two-Sample for Variances, z-Test: Two Sample for Means, Anova: Two-Factor with and without replication, and Exponential Smoothing.
Write-ups: a plain-English sentence and an APA-style line, such as t(46) = −4.66, p < .001, under t-tests, ANOVA, regression, correlation, F and z results.

Pro is £3.99 a month, £19 a year, or £39 once. A class or team licence for 5 or more people costs £12 per person a year; everyone with the buyer's school or work email address joins automatically.

Accuracy

Statbench's results are checked against reference values from SciPy and NumPy, and its regression reproduces the certified coefficients and standard errors of the NIST Statistical Reference Datasets' Longley problem to at least seven significant figures.

Privacy

Statbench opens only the spreadsheet you use it in. Your data is processed by Google Apps Script and never sent to Onibaku or anyone else. See the privacy policy.

Google Sheets is a trademark of Google LLC. Microsoft and Excel are trademarks of the Microsoft group of companies. XLMiner is a trademark of Frontline Systems, Inc. Statbench isn't affiliated with or endorsed by Google, Microsoft or Frontline Systems.