Table for publication (estimates, precision, confidentiality)

In brief

Under "New analysis" → "Table for publication" you get a table as official statistics publish it: weighted estimates or proportions, each cell with its precision, small cells suppressed, with a legend, footnotes and source.

How it works

1. Set the case weight in the project (survey weight). Without a weight, the table shows plain counts.
2. Choose rows and, if wanted, columns; content: estimated number, proportion in % (of the row, the column, the total), mean or sum of a number.
3. Precision: if the file has replicate weights (for example w_rep01 to w_rep50), choose them and enter type and factor - both are in the description of the data. Otherwise Statable only computes standard errors if people were drawn individually.
4. Confidentiality: suppress (cells below the minimum count, plus enough others that no suppressed number can be worked out from the totals) or round.
5. Enter source and footnote, then export (Excel, Word, PDF) or add to the report.

Symbols

• "x": suppressed (too few cases or confidentiality).
• "( )": low reliability (few cases or a large coefficient of variation).
• "-": nothing there. ".": not available.

Important

• The limits (minimum count, limits for "x" and "( )") are set by the organisation. The defaults are examples.
• The check log (reasons for suppression, counts, protection intervals) is only for checking. It is not exported.
• Confidentiality protects a single table. If several tables from the same data are published, suppressed numbers may be worked out through another table; the responsible body checks this.

For specialists

Estimation with R survey (svytotal, svymean; Taylor linearisation or svrepdesign with mse = TRUE), degrees of freedom from degf. Secondary suppression with the hypercube method for two dimensions with margins; every suppression pattern is checked by linear optimisation (protection interval, suppressed counts at least 1).