Assumptions and notations. Assume that Rater 1 and Rater 2 make binary votes, positive (+) or negative (−), on the same set of subjects with a total sample size N (Figure 2). For each of the rating subjects, we impose a two-step decision process: the two raters may feel certain or uncertain about each rating subject in step I, and make correct decisions in probability 1 (for certain cases) or with informative guessing (for uncertain cases) in step II. As mentioned earlier, the rater’s decisions are conditionally independent in our setting, which includes the complete independence assumption (in other’s work) as a special case with correlation coefficients fixed at 0. To apply the two-step decision process to generate raters’ voting, we first generate the underground truth of each rating subject j, j = 1, ..., N . The underlying outcome prevalence, or the true proportion of “+” outcomes in the population of rating subjects, is denoted as θ, which is a fixed attribute independent of specific rater’s response. In our simulation process, we generate raters’ votes by their consistency with respect to the underlying subject truth (i.e., “True (T)” vs “False (F)” in Figure 2), which can be later converted into votes (i.e., “+” vs “−” ) and aggregated into a 2 × 2 table (e.g., Table 1). In step I of the two-step decision process, we use a rater-specific probability pi to denote the probability to encounter uncertainty for Rater i, i = 1, 2. In step II, correct decisions aligned with the underlying truth are always made in the certain cases; for the uncertain cases, a rater’s subject-specific decision is exposed to the risk of misclassification, with a probability mi to make a wrong decision opposite to the underlying truth. Small mi indicates high accuracy of guessing from Rater i, which also partially describes the rater’s characteristics (e.g., a professional radiologist could still make a good judgment whether the subject is actually a disease case even though the evidence may not be completely deterministic). In step I, we incorporate the underlying correlation of rater behaviors using two contin- uous latent variables (LU , LU ). By introducing correlated latent variables, the difficulty of 1j 2j directly specifying correlation between binary variables could be coped with [54]. Detailed justifications of these latent variables are provided in Web Appendix B. In brief, (LU , LU ) 1j 2j ij denotes the Rater 1 and 2’s bivariate latent variables that controls the raters’ binary uncer- tain...
Assumptions and notations. Xxxx boards are denoted as Mx. Each device on a xxxx (sensor, actuator, tag, led, etc.) has a unique naming notation Dxy – indicating Device y on Mx; in our notation, actuators and sensors have the same format. Each device Dxy has a unique HANDLE ID (HIDxy), one of whose attributes is its network IPv6 address. Note that in a legacy system, the network address would be a combination of gateway address and legacy technology address; in this implementation, it can be a unique IPv6 address. DTLS secures many features. It can provide an authentication of the sender, integrity of the packet, and confidentiality. In this scenario, any attempt by the application to operate on a device requires authorisation through a secret Device Security Token (DST). Any data sent from a device must include a secret Device Authentication token (DAT). The DAT is required in case the sensor has been spoofed by a rogue sensor allegedly at the well-known network address of the real sensor. These terms and their significance have been fully analysed in Section 0. We assume that the complex task of assuring that only authorised entities are initiating an operation is done through the HANDLE security infrastructure together with the application. On receipt of an operation request, the device need check only the correctness of the DST. Any information received by the application from a device need check only the correctness of the DAT to determine that the information is from the indicated device.
