Mark R. Lehto

User acceptance of YouTube for procedural learning: An extension of the Technology Acceptance Model

The present study was framed using the Technology Acceptance Model (TAM) to identify determinants affecting behavioral intention to use YouTube. Most importantly, this research emphasizes the motives for using YouTube, which is notable given its extrinsic task goal of being used for procedural learning tasks. Our conceptual framework included two proximal antecedents of behavioral intention as proposed by the TAM - perceived usefulness and perceived ease of use. Additionally, the four first-order constructs of user satisfaction, content richness, vividness, and YouTube self-efficacy, as well as one second-order construct of content richness, were additionally incorporated into the framework by elaborating the theoretical structure. Sample data was collected from 432 respondents who were given the opportunity to engage in procedural learning through YouTube in a lab setting. The results derived from fitting the structural equation model on the sample indicated that behavioral intention was significantly influenced by both perceived usefulness and user satisfaction. Moreover, task-technology fit, content richness, vividness, and YouTube self-efficacy emerged as significant predictors of perceived usefulness. However, perceived ease of use was not significantly predictive of either perceived usefulness or behavioral intention. Our proposed model explained 43.8% of the variance in behavioral intention. Overall findings suggest that YouTube may augment its function as a common channel for procedural learning and instruction.

Models of accident causation and their application: Review and reappraisal

Abstract Numerous accident causation models have been developed, but little effort has been devoted toward systematically evaluating their potential strengths and methods of application. In this paper, 54 different accident causation models and 16 methods of application are reviewed. The accident causation models were divided into three groups: general models of the accident process, models of human error and unsafe behavior, and models of human injury mechanics. Models of human injury mechanics were particularly well-developed and required only rudimentary forms of task analysis to guide the development of safer working conditions. The general models of the accident process were applied extensively in the surveyed application methods. However, models of human error and unsafe behavior were applied only in a very rudimentary form. This in part appears to explain certain deficiencies in traditional hazard analysis.

Computerized coding of injury narrative data from the National Health Interview Survey

OBJECTIVE To investigate the accuracy of a computerized method for classifying injury narratives into external-cause-of-injury and poisoning (E-code) categories. METHODS This study used injury narratives and corresponding E-codes assigned by experts from the 1997 and 1998 US National Health Interview Survey (NHIS). A Fuzzy Bayesian model was used to assign injury descriptions to 13 E-code categories. Sensitivity, specificity and positive predictive value were measured by comparing the computer generated codes with E-code categories assigned by experts. RESULTS The computer program correctly classified 4695 (82.7%) of the 5677 injury narratives when multiple words were included as keywords in the model. The use of multiple-word predictors compared with using single words alone improved both the sensitivity and specificity of the computer generated codes. The program is capable of identifying and filtering out cases that would benefit most from manual coding. For example, the program could be used to code the narrative if the maximum probability of a category given the keywords in the narrative was at least 0.9. If the maximum probability was lower than 0.9 (which will be the case for approximately 33% of the narratives) the case would be filtered out for manual review. CONCLUSIONS A computer program based on Fuzzy Bayes logic is capable of accurately categorizing cause-of-injury codes from injury narratives. The capacity to filter out certain cases for manual coding improves the utility of this process.

Models of the warning process: important implications towards effectiveness

This paper presents a model-guided evaluation of research findings pertaining to warning signs and labels. Several results of significance are noted. First, the observed noticing, reading, and behavioral influence of warnings varied greatly between studies. Analyzing behavior from the perspective of information processing at different levels of performance provides a way of clarifying these mixed research results. Second, this analysis raises the question of how to best accommodate people at varying levels of performance. This seems to be one of the most difficult problems in warning design, in that the most effective warnings at each level are fundamentally different. Third, this analysis points out directions for future research. In particular, there is a strong need to focus future research on skill and rule-based behavior for a wide ranging set of products and use environments.

Risk-taking, warning labels, training, and regulation: Are they associated with the use of helmets by all-terrain vehicle riders?

Training programs and warning labels have been advocated as a means of convincing consumers to follow safety precautions when using all-terrain vehicles (ATVs), and are emphasized in the recent consent decree signed between the Consumer Product Safety Commission and ATV manufacturers. An alternative approach is to mandate safe behavior through regulations and law. This field study of ATV operator behavior conducted in six states in 1988 and 1989 provides initial insight into the effectiveness of these approaches. Logit analysis revealed that helmet use was significantly higher when required by law or when riders had 3 or more years of riding experience. Helmet use was significantly lower when riders rode frequently, when helmets were perceived as uncomfortable or not used on motorcycles, and when subjects had been moderately or seriously injured in ATV accidents. The presence of warning labels, reading of a manual, and participation in a training course were not significantly related to helmet use.

The effects of text structure and prior knowledge of the learner on computer-based learning

This study is an attempt to investigate the effects of document structure and knowledge level of the reader on reading comprehension, browsing, and perceived control. Four types of texts are distinguished, differing in structure (linear text, hierarchical hypertext, mixed hypertext, and generative text). All the materials were on a PC. In all conditions, participants were allowed 1h to read through the document. After completing the reading part of the experiment, they were asked to fill out the perceived control questionnaire followed by the reading comprehension test. As far as reading comprehension was concerned, knowledgeable participants had higher reading comprehension scores than non-knowledgeable participants only in the linear text. In addition, there were no significant differences in terms of the reading comprehension scores of the knowledgeable participants among the four topologies. However, the performance of non-knowledgeable participants differed with respect to the type of the topology. In particular, non-knowledgeable participants in the hierarchical and generative conditions performed better than those in the other two conditions. With respect to perceived control, the performance of knowledgeable and non-knowledgeable participants was equivalent in all four conditions. The results are discussed in terms of their implications for the computer-based learning.

Bayesian methods: a useful tool for classifying injury narratives into cause groups

To compare two Bayesian methods (Fuzzy and Naïve) for classifying injury narratives in large administrative databases into event cause groups, a dataset of 14 000 narratives was randomly extracted from claims filed with a worker’s compensation insurance provider. Two expert coders assigned one-digit and two-digit Bureau of Labor Statistics (BLS) Occupational Injury and Illness Classification event codes to each narrative. The narratives were separated into a training set of 11 000 cases and a prediction set of 3000 cases. The training set was used to develop two Bayesian classifiers that assigned BLS codes to narratives. Each model was then evaluated for the prediction set. Both models performed well and tended to predict one-digit BLS codes more accurately than two-digit codes. The overall sensitivity of the Fuzzy method was, respectively, 78% and 64% for one-digit and two-digit codes, specificity was 93% and 95%, and positive predictive value (PPV) was 78% and 65%. The Naïve method showed similar accuracy: a sensitivity of 80% and 70%, specificity of 96% and 97%, and PPV of 80% and 70%. For large administrative databases, Bayesian methods show significant promise as a means of classifying injury narratives into cause groups. Overall, Naïve Bayes provided slightly more accurate predictions than Fuzzy Bayes.

An experimental comparison of conservative versus optimal collision avoidance warning system thresholds

In the distributed signal detection theoretic (DSDT) model, the human operator and the warning mechanism are independent decision makers who work together as a team. The DSDT demonstrates that the optimal warning threshold, in general, differs from the signal detection theoretic (SDT) threshold, which assumes a single decision maker. This prediction was tested in an experiment where drivers received monetary rewards for making safe passing decisions on a driving simulator. The experiment focused on evaluating the quality of the decision making of the drivers, and not on perceptual issues. A collision avoidance system provided a warning when the probability of an inadequate overtaking gap exceeded a threshold. Three thresholds were tested. The control threshold resulted in no detections or false alarms. The DSDT threshold resulted in some misses but no false alarms. The SDT threshold resulted in no misses but frequent false alarms. As predicted, (1) drivers performed the best when the warning system used the DSDT threshold, and (2) use of the SDT threshold improved performance over the control threshold, even though four of the 10 drivers occasionally ignored the warning and made risky passing attempts in the SDT conditions, possibly because of earlier false alarms. These findings support the conclusion that the DSDT model is a useful, quantitative tool that should be used by warning designers.

Improving the effectiveness of warnings by increasing the appropriateness of their information content: some hypotheses about human compliance

Abstract The ways that warnings are administered vary greatly. A warning may come as a message broadcast on the radio about severe weather, as a flashing light in the cockpit of an airplane, or as an audible smoke alarm. Typically, warnings provide an auditory or visual signal to assist in the detection of an anticipated stimulus. However, warnings tend to operate in an all or none mode: either the warning is present, or it is not. Consequently, the information they provide is limited. If warnings are provided too often, their information content becomes even lower and frequent false alarms render them ineffective because of the “cry-wolf” effect. On the other hand, if warnings are not administered frequently enough, they result in too many potentially costly misses. In this conceptual paper, it is argued that the effectiveness of warnings might be significantly improved if warnings are made more “intelligent” by providing information about the likelihood of the occurrence of the stimulus. Several representative cases are discussed and analyzed in order to demonstrate the advantages of the proposed methods.

Parallel, self-organizing, hierarchical neural networks with competitive learning and safe rejection schemes

A new neural network learning algorithm with competitive learning and multiple safe rejection schemes is proposed in the context of parallel, self-organizing, hierarchical neural networks (PSHNN). After reference vectors are computed using competitive learning in a stage of PSHNN, safe rejection schemes are constructed for reference vectors. The purpose of safe rejection schemes is to reject the input vectors which are hard to classify. The next stage neural network is trained with the nonlinearly transformed values of only those training vectors that were rejected in the previous stage neural network Two different kinds of safe rejection schemes, RADPN and RAD, are developed and used together. Experimental results comparing the performance of the proposed algorithms with those of backpropagation and the PSHNN with the delta rule learning algorithm are discussed. The proposed learning network produced higher classification accuracy and much faster learning. The classification accuracies of two methods for learning the reference vectors were compared. When the reference vectors are computed separately for each class (Method II), higher classification accuracy was obtained as compared to the method in which the reference vectors are computed together for all the classes (Method I). This conclusion has to do with rejection of hard vectors, and is the opposite of what is normally expected. In addition, Method II has the advantage of parallelism by which the reference vectors for all the classes can be computed simultaneously. >