Robustness and Precision: How Data Quality May Influence Key Dependent Variables in Infant Eye‐Tracker Analyses

In recent years, eye‐tracking has become a popular method for drawing conclusions about infant cognition. Relatively little attention has been paid, however, to methodological issues associated with infant eye‐tracking. Here, we consider the possibility that systematic differences in the quality of raw eye‐tracking data obtained from different populations and individuals might create the impression of differences in gaze behavior, without this actually being the case. First, we show that lower quality eye‐tracking data are obtained from populations who are younger and populations who are more fidgety and that data quality declines during the testing session. Second, we assess how these differences in data quality might influence key dependent variables in eye‐tracking analyses. We show that lower precision data can appear to suggest a reduced likelihood to look at the eyes in a face relative to the mouth. We also show that less robust tracking may manifest as slower reaction time latencies (e.g., time to first fixation). Finally, we show that less robust data can manifest as shorter first look/visit duration. We argue that data quality should be reported in all analyses of infant eye‐tracking data and/or that steps should be taken to control for data quality before performing final analyses.     

Structure Learning of Contextual Markov Networks using Marginal Pseudo‐likelihood

Markov networks are popular models for discrete multivariate systems where the dependence structure of the variables is specified by an undirected graph. To allow for more expressive dependence structures, several generalizations of Markov networks have been proposed. Here, we consider the class of contextual Markov networks which takes into account possible context‐specific independences among pairs of variables. Structure learning of contextual Markov networks is very challenging due to the extremely large number of possible structures. One of the main challenges has been to design a score, by which a structure can be assessed in terms of model fit related to complexity, without assuming chordality. Here, we introduce the marginal pseudo‐likelihood as an analytically tractable criterion for general contextual Markov networks. Our criterion is shown to yield a consistent structure estimator. Experiments demonstrate the favourable properties of our method in terms of predictive accuracy of the inferred models.     

Experimental designs for binary data in switching measurements on superconducting Josephson junctions

Summary.  We study the optimal design of switching measurements of small Josephsonjunction circuits which operate in the macroscopic quantum tunnelling regime. In the experiment, sequences of current pulses are applied to the Josephson junction sample, while the voltage over the structure is monitored. The appearance of a voltage pulse to a single applied current pulse, being governed by the laws of quantum mechanics, is purely random. Starting from the D -optimality criterion we derive the optimal design for the estimation of the unknown parameters of the underlying Gumbel-type distribution. As a practical method for the measurements, we propose a sequential design that combines heuristic search for initial estimates and maximum likelihood estimation. The design presented has immediate applications in the area of superconducting electronics, implying faster data acquisition. The experimental results presented confirm the usefulness of the method.     

Estimation of True Salmonella Prevalence Jointly in Cattle Herd and Animal Populations Using Bayesian Hierarchical Modeling

The Finnish salmonella control program (FSCP) for beef production is based on both randomized and selective testing of herds and animals. Sampling of individual animals in abattoirs is randomized. Herds are selectively tested for salmonella on the basis of clinical symptoms and/or other factors. Risk assessment of FSCP is inherently difficult due to the complexity of the complete data set, especially if the detailed labeling of the testing types is lost. However, for a risk assessment of the whole production chain, methods for exploiting all available data should be considered. For this purpose, a hierarchical Bayesian model of true salmonella prevalence was constructed to combine information at different levels of aggregation: herds in geographical regions and individual animals arriving for slaughter. The conditional (municipality specific) probability of selection of a herd for testing was modeled given the underlying true infection status of the herd and information about the general sampling activity in the specific region. The model also accounted for the overall sensitivity of the sampling methods, both at the herd and at the animal level. In 1999, the 95% posterior probability intervals of true salmonella prevalence in the herd population, in individual cattle, and in slaughter animal populations were [0.54%, 1.4%] (mode 0.8%), [0.15%, 0.39%] (mode 0.2%), and [0.12%, 0.36%] (mode 0.2%), respectively. The results will be further exploited in other risk assessments focusing on the subsequent parts of the beef production chain, and in evaluation of the FSCP.     

Bayesian model learning based on a parallel MCMC strategy

We introduce a novel Markov chain Monte Carlo algorithm for estimation of posterior probabilities over discrete model spaces. Our learning approach is applicable to families of models for which the marginal likelihood can be analytically calculated, either exactly or approximately, given any fixed structure. It is argued that for certain model neighborhood structures, the ordinary reversible Metropolis-Hastings algorithm does not yield an appropriate solution to the estimation problem. Therefore, we develop an alternative, non-reversible algorithm which can avoid the scaling effect of the neighborhood. To efficiently explore a model space, a finite number of interacting parallel stochastic processes is utilized. Our interaction scheme enables exploration of several local neighborhoods of a model space simultaneously, while it prevents the absorption of any particular process to a relatively inferior state. We illustrate the advantages of our method by an application to a classification model. In particular, we use an extensive bacterial database and compare our results with results obtained by different methods for the same data.     

Perception of Emotions in the Hand Movement Quality of Finnish Sign Language

This study investigated whether people can decode emotion (happiness, neutrality, and anger) communicated via hand movements in Finnish sign language when these emotions are expressed in semantically neutral sentences. Twenty volunteer participants without any knowledge of sign language took part in the experiment. The results indicated that the subjects were able to reliably decode anger and neutrality from the quality of hand movements. For happy hand expressions, the responses of happiness and neutrality were confused. Thus, the study showed that emotion-related information can be encoded in the quality of hand movements during signing and that this information can be decoded without previous experience with this particular mode of communication.     

A Probabilistic Transmission Model of Salmonella in the Primary Broiler Production Chain

Annual data from the Finnish National Salmonella Control Programme were used to build up a probabilistic transmission model of salmonella in the primary broiler production chain. The data set consisted of information on grandparent, parent, and broiler flock populations. A probabilistic model was developed to describe the unknown true prevalences, vertical and horizontal transmissions, as well as the dynamical model of infections. By combining these with the observed data, the posterior probability distributions of the unknown parameters and variables could be derived. Predictive distributions were derived for the true number of infected broiler flocks under the adopted intervention scheme and these were compared with the predictions under no intervention. With the model, the effect of the intervention used in the programme, i.e., eliminating salmonella positive breeding flocks, could be quantitatively assessed. The 95% probability interval of the posterior predictive distribution for (broiler) flock prevalence under current (1999) situation was [1.3%-17.4%] (no intervention), and [0.9%-5.8%] (with intervention). In the scenario of one infected grandparent flock, these were [2.8%-43.1%] and [1.0%-5.9%], respectively. Computations were performed using WinBUGS and Matlab softwares.     

How Do Efforts to Enhance Career Preparation Affect Peer Groups?

The present study investigated how efforts to enhance career preparation affect peer groups. The participants were 710 9th graders who were randomized into control and intervention groups and assessed 3 times during 1 academic year: at baseline (T1, Fall), immediately after the career preparation intervention (T2, 3 months after baseline), and 5 months after the intervention (T3, Spring). The results showed, first, that maintenance of a stable clique was more likely when most clique members participated in the intervention. Second, it was found that the members of adolescents' cliques resembled each other in respect of the strength of the intervention effect. Finally, the results showed that the similarity of the clique members in career choice preparedness first decreased as a result of the short‐term impact of the intervention but then increased after the intervention was over, suggesting that peer groups were partly responsible for weakening the effect of the intervention. These results have important implications for understanding how peer groups moderate external influences on their members and how the stability of peer groups is affected by these influences.     

Likelihood-free inference via classification

Increasingly complex generative models are being used across disciplines as they allow for realistic characterization of data, but a common difficulty with them is the prohibitively large computational cost to evaluate the likelihood function and thus to perform likelihood-based statistical inference. A likelihood-free inference framework has emerged where the parameters are identified by finding values that yield simulated data resembling the observed data. While widely applicable, a major difficulty in this framework is how to measure the discrepancy between the simulated and observed data. Transforming the original problem into a problem of classifying the data into simulated versus observed, we find that classification accuracy can be used to assess the discrepancy. The complete arsenal of classification methods becomes thereby available for inference of intractable generative models. We validate our approach using theory and simulations for both point estimation and Bayesian inference, and demonstrate its use on real data by inferring an individual-based epidemiological model for bacterial infections in child care centers.