statistical power analysis

Statistical power analysis

BMT Oceanica is actively engaged in statistical research and recently won a BMT Innovations scholarship to develop specialist power and cost-benefit analysis tools.

Much of BMT Oceanica’s environmental monitoring work is focussed on detecting environmental changes arising from anthropogenic impacts.  We continue to actively research into the design of robust environmental monitoring and management programs which assist both proponents and regulators.  We recognise that best practice detection of environmental changes typically involve complex experimental designs to minimise the risks associated with:

  • false detection of an impact when the change is due to chance alone (Type I error); or 
  • failure to detect an impact if one exists (Type II error).  

These risks are complimentary and inherent in all statistical analyses.  At BMT Oceanica we seek to develop programs that are acceptable to both the proponent and the regulator and which find a balance between these two errors.  This balance may be found through the application of statistical cost-benefit analysis.  However, such analyses are not straight forward, particularly when applied to complex environmental survey designs which are often typical for EIA programs.

 

Statistical cost-benefit analysis

To ensure BMT Oceanica remains at the forefront of monitoring program design we have developed a suite of tools which balance the cost of a monitoring program against its statistical sensitivity.  This is achieved through a determination of the probability of Type I and II error for complex survey designs together with a review of the optimum field costs and statistical power for these designs.

 

Power for percentiles

BMT Oceanica also has a strong understanding of the non-parametric statistical approach advocated by National Water Quality Management Strategy.  This approach, which involves comparison of percentile statistics, is also subject to Type I and II errors.  We are also actively examining the optimum program design which balances sampling effort and statistical rigour.