Figure 4
The shape of normalized likelihood functions under a number of different approximations for different input parameters indicates that the use of a point estimate for negative intensities or those with high noise values results in significant deviations from the ideal likelihood function. The difference between a t-based noise model and a normal noise model is small, but significantly affects the tail behavior of the likelihood function. Amplitude and standard deviation estimates for both the French–Wilson and Sivia approaches are given in the figure. |