Various Types of Statistical Analysis Using Maximum Likelihood With Left-Censored Quantification Limits
Yukio Takahashi, President, BioStat Institute Co. Ltd.
A lot of measurement data contains limits below or above values that cannot be accurately measured. For such data, an average value or 95 percent confidence interval should be calculated. For the sake of convenience, would it not be better to calculate the average value with half of the lower limit of quantitation, considering it 0, etc.? And how should the upper limit of quantitation be handled? In such cases, the statistical median value and interquartile range function as the average value and standard deviation (SD). However, there are limits to the analyses that can be performed using various statistical models. Although regression analysis should be performed with the measurement limits included, how should this be done? Analysis using maximum likelihood, which includes censoring data for univariate distribution and bivariate relationship, is now possible using Life Distribution and Fit Life by X platforms in JMP. In these platforms, not only is right censoring performed, but left censoring and interval censoring are also performed simultaneously. These platforms are multifunctional and require sensitive settings, but the instruction manual does not include examples of data analysis that includes quantification limits. I will be presenting analysis methods and results with various data examples.