Crosstab

Each banner column gets a letter. A letter inside a cell says that this cell's percentage is significantly higher than the same cell in the column with that letter, at the confidence level you chose: 90, 95 or 99 percent. The test is a two-sample z-test on column proportions, the SPSS-style convention, with optional Bonferroni correction for the number of comparisons in a block.

Columns with a base below your threshold are flagged with an asterisk and excluded from testing, so a letter never appears on ten respondents. The footer explains the letters and marks, and the rule is printed, not implied. The full method, including the documented column-overlap limitation, is on the methodology page.

Weighted surveys need two things from a table: percentages on the weighted counts and tests on the effective base, not the weighted one. Map Sum of Weight as the value and Sum of Weight squared next to it, and the visual computes the Kish effective base per column, n_eff = (Σw)² / Σw², and runs every test on it. The footnote prints the base and, under weighting, the effective n next to it.

Unweighted tables work with a plain respondent count and no second measure; the letters then use the raw base.

Top box and bottom box net rows summarize a scale question in one line, one to three scale points each, and carry their own letters. Mean rows put the scale mean under the block with Welch-test letters between columns. An index view shows each cell against the Total column, with diverging shading and marks for cells above or below the rest of the sample.