Standard Uncertainty Determination of the Mean for Correlated Data Using Conditional Averaging

Abstract

The correlation of data contained in a series of signal sample values makes the estimation of the statistical characteristics describing such a random sample difficult. The positive correlation of data increases the arithmetic mean variance in relation to the series of uncorrelated results. If the normalized autocorrelation function of the positively correlated observations and their variance are known, then the effect of the correlation can be taken into consideration in the estimation process computationally. A significant hindrance to the assessment of the estimation process appears when the autocorrelation function is unknown. This study describes an application of the conditional averaging of the positively correlated data with the Gaussian distribution for the assessment of the correlation of an observation series, and the determination of the standard uncertainty of the arithmetic mean. The method presented here can be particularly useful for high values of correlation (when the value of the normalized autocorrelation function is higher than 0.5), and for the number of data higher than 50. In the paper the results of theoretical research are presented, as well as those of the selected experiments of the processing and analysis of physical signals.

Estimation of Conditional Expected Value for Exponentially Autocorrelated Data

“…New possibilities in this scope for linear and non-linear models in metrological applications appear thanks to the use of functional and numerical conditional characteristics, in particular those of the conditional expected value and the conditional variance [14,15]. The conditional expected value ensures the best estimate of interdependencies of stochastic signals in the mean square sense.…”

“…In a simplified model of averaging non-correlated M fragments of x(t), after exceeding the ( ) p x t x = level [14], the assessment of the relative standard uncertainty of the conditional value of arithmetic mean is:…”

“…Therefore, in studies and publications, other probabilistic characteristics describing the relationships of a stochastic nature are introduced and applied [14]. Methods of measurement and analysis of such characteristics are being developed intensively.…”

A study of uncertainty of expert measurement results in the quality management system

“…To assess and eliminate uncertainty related to the incorrect choice of the number of experts, it is important to consider the statements of the probability theory (namely, the expression of an error confidence level) [22,23] and represent the evaluation of the expert number C expert at the given confidence probability P within the range of values inherent in metrology -namely, from 0.9 to 0.99 with the error Δ. Using the expression for calculating a confidence interval, the formula for calculating C expert , which is a prototype of the number of observations, could be written as follows: 2 2…”

Measurement of BLDC Motor Efficiency for Commutation Angle Optimisation

“…Measured data are not random sampling data with normal distribution but autocorrelated time series. Calculation of confidence interval is not straightforward for such data [9], [10]. The number of samples required for the mean estimation was determined by means of the experimental method from the long time measurement.…”