Bayesian hypothesis testing-use in interpretation of measurements.

Area of Science:

• Statistics
• Scientific Computing

Background:

• Traditional hypothesis testing methods can be limited in interpreting data for detection.
• Bayesian approaches offer an alternative framework for statistical inference.

Purpose of the Study:

• To demonstrate the equivalence of Bayesian hypothesis testing to evaluating the posterior distribution for true positive amounts.
• To introduce a novel prior definition for hypothesis testing.
• To provide a Bayesian interpretation of hypothesis testing based on posterior probabilities.

Main Methods:

• Redefining the prior distribution as a mixture of the original prior and a delta-function at 0.
• Calculating the posterior probability of the modeling hypothesis relative to the null hypothesis.
• Applying the method to real numerical examples, including internal dosimetry data.

Main Results:

• Bayesian hypothesis testing provides a qualitative interpretation of data as "detected" or not.
• The probability of the non-null hypothesis was analyzed across 4,000 internal dosimetry cases.
• The proposed Bayesian method was compared with classical statistical approaches.

Conclusions:

• Bayesian hypothesis testing, with a redefined prior, offers a robust framework for data interpretation.
• This approach provides a clear measure of evidence for or against a null hypothesis.
• The method is applicable to real-world datasets and comparable to existing statistical techniques.