A banner table (also called a crosstab or cross-tabulation) is the core deliverable of quantitative MR. Rows represent survey questions or response codes. Columns represent demographic or attitudinal subgroups (the "banner"). Each cell shows what percentage of that column's respondents gave that response.
Column Percentage
The fundamental cell statistic is the column percentage:
Column % = (count of respondents in cell / base of banner point) × 100
The denominator is always the banner point base — the total number of respondents (weighted or unweighted) who fall into that column, regardless of how they answered the stub question. This is what makes column % comparable across banner points of different sizes.
Weighting
When a weight variable is specified, each respondent's contribution is multiplied by their weight value before summing. Both the numerator (count in cell) and the denominator (banner point base) are weighted sums:
Weighted count = Σ weight_i for all i where banner mask AND stub code match Weighted base = Σ weight_i for all i where banner mask matches
Weights must be positive. Zero or negative weights are treated as 1 (unweighted) to prevent distorted bases.
Significance Letters (A/B/C Testing)
Each banner column is assigned a letter (A, B, C…). When a cell's proportion is significantly higher than another column, that column's letter appears as a superscript in the cell. Significantly lower values appear as lowercase letters.
The test used is the two-proportion z-test:
p̂_pooled = (x₁ + x₂) / (n₁ + n₂) z = (p̂₁ - p̂₂) / √(p̂_pooled × (1 - p̂_pooled) × (1/n₁ + 1/n₂)) p-value from standard normal distribution (two-tailed)
Multiple Comparison Corrections
With k banner points there are k(k-1)/2 pairwise comparisons. Running all comparisons at α=0.05 inflates the family-wise error rate. Two corrections are available:
- Bonferroni: Divide α by the total number of comparisons. Conservative but simple.
α_adjusted = α / [k(k-1)/2] - Holm (step-down): Sort p-values ascending, apply sequentially stricter thresholds. More powerful than Bonferroni while still controlling family-wise error rate.
Multi-coded Variables
For "select all that apply" questions, a single respondent can select multiple codes. Column percentages across all codes will sum to more than 100% — this is correct and expected. The denominator remains the banner base (all respondents in that column), not the number of selections.
Krosstabs warns when it detects potential multi-coded variables so you can verify before interpreting results.
Banner Point Base
The base row shows the total respondents (or weighted total) in each column. Bases below 30 are flagged — proportions computed from small samples are unstable and significance tests may be unreliable. Industry practice is to mark small bases with an asterisk (*) or suppress them entirely.
Interpretation Guide
- Read columns, not rows. Each column is its own 100% universe.
- Sig letters tell you which columns differ, not the direction of the market preference.
- A cell with no sig letters is not necessarily unimportant — it may simply not differ from other columns in a statistically meaningful way given the sample size.
- Always check the base. Significance results from very small bases (n<30) should be treated with caution.