> ## Documentation Index
> Fetch the complete documentation index at: https://support.deepfield-ai.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Interpret significance
> Explain crosstab colors and letters, check the statistical base, and report differences with appropriate limits.
Significance markers identify differences that would be unlikely under the test's no-difference assumption. They describe statistical evidence in the current table; they do not measure business importance or establish causation.
## Before you start
Open the table’s legend and identify its raw column bases, confidence levels, and tests.
## Distinguish colors from letters
* **Significant cells** colors compare each column percentage with the overall percentage on the same row. Green is higher and red is lower at the displayed level.
* **Column proportions test** letters compare columns directly. B in column A means A is significantly higher than B.
Uppercase letters and stronger colors use the primary confidence level. Lowercase letters and lighter colors use the optional secondary level. Open the crosstab **Legend** to confirm the active settings.
## Interpret a 95% result
At a 95% confidence level, the result indicates that a difference at least this large would occur in fewer than 5% of repeated samples under the no-difference assumption used by the test.
It does not mean:
* The result is proven with 95% certainty.
* The difference is large enough to matter to the decision.
* One variable caused the other.
* An unmarked difference is evidence that the groups are identical.
Report the observed size and the statistical result separately. For example: “Purchase intent was 14 points higher in column A than column B; the column-proportion test marked the difference at 95%.”
## Check the unweighted base
Crosstab tests use real respondent counts even when weighted values are displayed.
| Base | Interpretation in the table |
| ----------- | ------------------------------------------------ |
| 100 or more | Eligible for testing without a small-base marker |
| 30–99 | Tested and displayed in parentheses as a caution |
| Below 30 | Displayed with `**` and not tested |
If all columns are below 30, no significance comparison is performed. A table without markers can therefore mean either that tests found no marked differences or that the bases were not eligible. Read the note beneath the table.
## Account for repeated comparisons
A large crosstab can run many cell comparisons. At a 95% threshold, isolated markers can occur by chance across a large set of tests. Give more weight to comparisons specified in the analysis plan and to coherent patterns across related rows than to one unexpected marker found after scanning the table.
The plan folder applies a selection of its own: each plan crosstab uses coloured significant cells and letters at a 95% primary level, and keeps only the comparisons in which a column with an unweighted base of 30 or more has a significant cell. A kept comparison can rest on a single marked cell, so read the whole table before reporting it. A planned comparison that was not kept is either untested because its bases were small or not significant; it is not evidence that the groups are identical.
## Write the result
Include:
* The compared groups and measured answer.
* The percentage or mean in each group.
* The point difference where useful.
* The unweighted bases.
* The confidence level and test used.
* A directional qualifier when the result is not marked or not tested.
Avoid replacing a missing test result with a claim of no difference. State that the observed difference was not significant at the selected level, or that the base was below the testing threshold.
## Check the result
The written conclusion names the compared groups, values, raw bases, test, and confidence level. Distinguish an untested small base from a tested difference that was not significant.
## Related
* [Build a crosstab](/analysis/crosstabs)
* [Choose a sample size](/recruiting/choosing-a-sample-size)
* [Create and apply weighting](/analysis/weighting)
* [Decide whether to weight](/analysis/weighting-methodology)