Changelog

Microsoft sign-in asks for an email when it has none. The header shows names instead of placeholders. Empty states are centred. Buttons and inputs are consistent across the app.

• Microsoft sign-in asks for an email when the account has none
• The header shows your name
• Empty lists sit in the centre of the page
• Buttons and inputs use the same controls

A significance letter sits on the column with the higher percentage, or the higher mean. The letter still names the other column. Upper and lower case still follow how small the p value is. The table builder and the analysis forms use the same fields as the rest of the app.

• On a 1 to 5 satisfaction table, the letter for the higher group sits on that group
• Suggested questions use a category for a crosstab, or they say to bin a continuous column such as Age

A 1 to 5, 1 to 7, or 1 to 10 scale shows whole numbers and offers Top 2 box and Bottom 2 box. A nominal variable such as gender does not. A column that mixes text and numbers is saved as text, with a warning, instead of failing the upload.

• Suggested questions skip id columns such as RespID
• Opening a report uses the saved run when the tab is gone
• If that dataset was deleted, the message says so

The code screen says a code can take a minute and shows the address it was sent to. You can resend after 30 seconds. Microsoft sits next to Google.

• With datasets on the home page, upload and new dataset sit by the title
• With no datasets, the templates and upload box stay
• Deleting a dataset asks once, then shows progress and the result

R cells load the open dataset on the server, with the column types from the file. A long sheet no longer gets pasted into the cell. Every print comes back in order. A failed cell shows the R error, and the full text is under Details.

• Shift+Enter and Cmd/Ctrl+Enter run the text in the editor
• Two prints on one line both come back
• Internal columns such as _row_uid stay out of df

A banner net stays on that table. Saving a net onto the variable is a separate step. A 1 to 5 scale stored as decimals counts the same codes you see in the builder, including Top 2 box. Re-running a table from an Excel file keeps the R check, and a comparison in the summary has to match the cells.

• Q1 by Group with Top 2 box counts the people on codes 1 to 5
• Re-run of an xlsx table stays verified
• A false "highest" sentence is dropped
• Inline tables in an answer render as tables

Automatic ARIMA picks its model the same way as R's auto.arima, choosing differencing first and then comparing models by AICc. Reports show both AIC and AICc, and seasonal terms are skipped when there are fewer than three full seasons of data.

Decision trees now default to a max depth of 5 and a minimum leaf size of 5, so they no longer overfit out of the box. Random forests use the same leaf size. Clear the depth field if you want an unrestricted tree.

The docs now cover survival, time series, survey techniques and machine learning.

We compared Tensr's output against R across the core analyses and fixed every mismatch we found. Each fix now has a test that checks it against R on the same data.

• Tukey post-hoc tables fill in standard errors and show very small p-values as < .001 instead of 0
• Mixed ANOVA between-subjects sums of squares now match R and SPSS, and Mauchly's test is hidden when there are only two levels
• Descriptives show the statistics you select
• Negative binomial regression estimates dispersion rather than fixing it at 1, so standard errors and p-values are correct
• Cohen's kappa is correct (it could previously show 1), and weighted kappa now reports SE and p
• Lilliefors p-values match R's nortest
• The Moses test runs again and reports a p-value
• Effect-size labels use one rule at cutoffs and cite their source
• Written summaries only mention your own variables

Column letters now run on the weighted banner, not a separate test menu. When a client asks where a letter came from, the cell still points at the respondent rows.

• 95% and 90% column proportions
• Low-base cells stay visible and skip the letter
• Letters export into Excel and PowerPoint with the grid

Excel export was writing raw codes in the banner stub. It now writes the value labels you see in the grid, so the book matches the screen.

• Multi-response sets keep their set label
• Weighted bases round the same way as the grid