Proposed New Method for Applying Confidentiality to Business Demography Employment Statistics
Statistics New Zealand proposes to implement a new method for applying confidentiality to Business Demography employment data, to coincide with the release of the 2004 statistics later this year. Statistics New Zealand invites feedback from users of Business Demography statistics on this proposal by 30 June 2004.
Business Demography provides an annual snapshot of changes in the number, type and location of businesses in New Zealand. Analyses can be undertaken using a range of variables, including geographic area, industry, institutional sector, business type and business size (employment level). Business demographic statistics are sourced from the Statistics New Zealand Business Frame. This register of businesses in New Zealand, is primarily used to select businesses for inclusion in economic surveys.
Statistics New Zealand is required by the Statistics Act to protect the confidentiality of individual businesses in statistics it publishes. In Business Demography this involves only releasing aggregated data - that is, observations for individual businesses are grouped with other observations. Aggregation is done for all variables that Business Demography releases statistics for. In addition, published totals are rounded.
Statistics New Zealand has recently reviewed the method for protecting the confidentiality of employment data in Business Demography and proposes to enhance the level of protection for those published cells with relatively few numbers of observations. The greater availability of fine level data and cross tabulations on the Statistics New Zealand website, and the demand for customised outputs, have made it necessary to strengthen protection to avoid the possibility of disclosure of individual data.
There are two options for improving protection. One is to not publish statistics for data cells where there are relatively few observations. This option has been discounted because it would result in the loss of a significant amount of published information. Also it is practically difficult to implement with such a large data source which has many possible cross-tabulations.
The other option being proposed involves the addition of random noise to business employment statistics. For published cells containing many observations there will be little impact as the noise tends to cancel itself out. This method has a greater impact on cells containing relatively fewer observations, changing the values in these cells and thus providing a greater level of confidentiality.
The table below demonstrates that the noise method generally offers greater confidentiality protection to cell values than the current method of rounding cell values - as indicated in the percentage difference column. It also demonstrates that as the number of observations in a cell reduces (as represented by the number of geographic units), the level of protection offered by the noise method increases. When there are large numbers of observations in a cell, individual observations in these cells are protected by being aggregated with many other observations. In cells with larger numbers of observations the random noise applied to individual observations tends to cancel itself out, thus having little impact on the published data. In cells with smaller numbers of observations the random noise has a greater impact on the published value and thus affords greater protection to the individual observations in each cell.
|Number of Geographic units
||Full time equivalent persons engaged|
||Percentage change from unrounded |
||Noise Applied and Rounded
||Noise Applied and Rounded|