The effect of experimental uncertainty on the ability to constrain the value of the random loss fraction.

<p>Fits to simulated data with Gaussian random noise of varying standard error of the mean (s.e.m.) added to simulate experimental uncertainty. (A)–(E) Simulated model parameters: <i>μ</i> = 100 hr, <i>σ</i> = 25 hr, with (A) <i>f</i> = 0.2, (B) <i>f</i> = 0.8. For each value of <i>f</i> and each value of the s.e.m. of the noise, the fit was repeated 1000 times with independently-generated noise – the box-and-whisker plots indicate median, interquartile range, 2.5 and 97.5 percentiles, and outliers are plotted as individual dots. (C) Interquartile range as a function of random loss fraction for the various levels of noise added to the simulated data. For a given value of the s.e.m. of the noise, higher values of the random loss fraction are generally better constrained than lower (non-zero) values.