Comparison of Selected Methods of the Confidence Intervals for Long-Term Noise Indicators

The basic percentile model (per model) has been compared with interval estimation algorithms of the classical (parametric) statistics assuming that the standard deviation of the population is either known ( Nσk model) or unknown ( Nσu model). A theoretical basis of the proposed model is presented as well as an example of calculation process which makes possible determining confidence intervals of the expected value of long-term noise indicators LDEN and LN. The statistical analysis was carried out for 95% interval widths obtained by using each of these models. The inference of their usefulness to the type A standard uncertainty evaluation of long-term noise indicators was performed on the basis of results of non-parametric statistical tests at significance level α = 0.05. The simulation experiment showed that the width differences of confidence intervals are statistically significant for all three models. The per algorithm is characterised by a high degree of coverage of the measured (actual) value. This method can be successfully applied to a small random sample not having any asymptotic properties. The confidence intervals of the per model are asymmetric with respect to the point estimate in contrast to other described models. Moreover, the confidence intervals obtained using the Nσk model have a constant width. The data used to illustrate the proposed solutions and carry out the analysis were results of continuous monitoring of traffic noise recorded in 2008 in one of the main arteries of Krakow in Poland.