Downloaded from https://www.everyspec.com ADS-79D-HDBK σ 2 = variance µ = population mean = undamaged population mean = estimated sample mean ȳ i = sample mean of undamaged components ȳ 2 = sample mean of damaged components z = (standard normal) distribution Z= confidence Interval I.3 GENERAL GUIDANCE I.3.1 Background . To verify and validate CIs and His, an appropriate sample size of faulted and unfaulted components needs to be determined. In theory, more samples would be better. But in reality, sample size is always a limiting factor for validation problems. Many factors impact the selection of appropriate sample size. When choosing a sample size, the following issues must be considered:
a. What population parameters we want to estimate; b. Criticality due to lack of accurate information; c. Cost of sampling (importance of information); d. How much is already known (prior knowledge / experience); e. Spread (variability) of the population; f. Practicality: how hard is it to collect data; g. How precise we want the final estimates to be. Inevitably, there is a trade-off among sample size, cost and precision of the anticipated regression model. To develop a highly accurate predictive model, more samples are generally required which results in higher validation cost. To select an appropriate sample size, a probabilistic statement about what is expected of the sample is needed.
The targeted estimate from sampling must be determined first. Typically, several statistical characteristics are of interest, including mean value, standard deviation, population proportion, etc. Depending on an identified target estimate, different sampling requirements may be followed. For example, to establish a reasonable statistical estimate for average value, a small sample size (as low as 8 – 10) may be sufficient, while a larger sample size may be needed to ensure a good estimate of standard 147