Towards multimodal emotion recognition in e-learning environments

This paper presents a framework (FILTWAM (Framework for Improving Learning Through Webcams And Microphones)) for real-time emotion recognition in e-learning by using webcams. FILTWAM offers timely and relevant feedback based upon learner's facial expressions and verbalizations. FILTWAM's facial expression software module has been developed and tested in a proof-of-concept study. The main goal of this study was to validate the use of webcam data for a real-time and adequate interpretation of facial expressions into extracted emotional states. The software was calibrated with 10 test persons. They received the same computer-based tasks in which each of them were requested 100 times to mimic specific facial expressions. All sessions were recorded on video. For the validation of the face emotion recognition software, two experts annotated and rated participants’ recorded behaviours. Expert findings were contrasted with the software results and showed an overall value of kappa of 0.77. An overall accuracy of our software based on the requested emotions and the recognized emotions is 72%. Whereas existing software only allows not-real time, discontinuous and obtrusive facial detection, our software allows to continuously and unobtrusively monitor learners' behaviours and converts these behaviours directly into emotional states. This paves the way for enhancing the quality and efficacy of e-learning by including the learner's emotional states.     

Approaches for Specifying the Level-1 Error Structure When Synthesizing Single-Case Data

Multilevel modeling has been utilized for combining single-case experimental design (SCED) data assuming simple level-1 error structures. The purpose of this study is to compare various multilevel analysis approaches for handling potential complexity in the level-1 error structure within SCED data, including approaches assuming simple and complex error structures (heterogeneous, autocorrelation, and both) and those using fit indices to select between alternative error structures. A Monte Carlo study was conducted to empirically validate the suggested multilevel modeling approaches. Results indicate that each approach leads to fixed effect estimates with little to no bias and that inferences for fixed effects were frequently accurate, particularly when a simple homogeneous level-1 error structure or a first-order autoregressive structure was assumed and the inferences were based on the Kenward-Roger method. Practical implications and recommendations are discussed.     

Optimal cognitive distance and absorptive capacity

In this paper we test the relation between cognitive distance and innovation performance of firms engaged in technology-based alliances. The key finding is that the hypothesis of an inverted U-shaped effect of cognitive distance on innovation performance of firms is confirmed. Moreover, as expected, we found that the positive effect for firms is much higher when engaging in more radical, exploratory alliances than in more exploitative alliances. The effect of cumulative R&D turns out to be mixed. It appears to increase absorptive capacity, as expected, but there is clear evidence that it also reduces the effect of cognitive distance on novelty value, making it increasingly difficult to find additional novelty.     

The relationship between student-generated learning issues and self-study in problem-based learning

A major assumption of problem-based learning (PBL) is that learning issues, generated by students while discussing a problem, are used as guides for self-directed learning activities. This assumption, though basic to PBL, has never been tested. At the University of Limburg, the Netherlands, two procedures have been developed that reflect the extent to which students are able to identify important learning issues given a particular problem, and whether subsequent, independent, learning corresponds with these learning issues. The focus of the present article will be on the relationship between the two. We have explored to what extent student-generated learning issues are a major factor influencing the nature of students' self-study, or whether other factors may be involved in decisions on what to study and how much time to spend on topics selected. First, the production of learning issues was studied and represented as the percentage of overlap between learning issues raised by students and pre-set faculty objectives for each problem. The second procedure consisted of the administration of a Topic Checklist (TOC) which purports to measure students' actual self-directed learning activities. The TOC consists of a list of topics specifying the intended course content. Students were asked to indicate on a five-point Likert scale how much time they had spent studying each topic and to what degree they had mastered that topic. Third, learning issues and TOC topics were compared directly in a qualitative sense. Comparisons between the procedures revealed that a low proportion of variance of TOC scores could be predicted from the percentage of faculty objectives identified for each problem and the direct match between learning issues and TOC scores. It is concluded that scrutinizing student-generated learning issues and topics covered during self-study may provide information about what content is covered by students in tutorial groups. The discrepancy between the results of the measurements suggests, however, that learning issues produced during group discussion are not the sole source on which students base self-study decisions. Several other factors may be involved, such as tutor guidance, content already covered in previous units, issues raised during sessions with resource persons, and the nature of the learning resources available. Therefore, the relationship between learning issues and content covered during self-study is not as straightforward as is suggested.     

Detecting Differential Speededness in Multistage Testing

A potential undesirable effect of multistage testing is differential speededness, which happens if some of the test takers run out of time because they receive subtests with items that are more time intensive than others. This article shows how a probabilistic response-time model can be used for estimating differences in time intensities and speed between subtests and test takers and detecting differential speededness. An empirical data set for a multistage test in the computerized CPA Exam was used to demonstrate the procedures. Although the more difficult subtests appeared to have items that were more time intensive than the easier subtests, an analysis of the residual response times did not reveal any significant differential speededness because the time limit appeared to be appropriate. In a separate analysis, within each of the subtests, we found minor but consistent patterns of residual times that are believed to be due to a warm-up effect, that is, use of more time on the initial items than they actually need.     

The misconception of the regression-based discrepancy operationalization in the definition and research of learning disabilities

In this article, I argue that the regression-based discrepancy method used in the diagnosis of learning disabilities is invalid because it is inconsistent with the underlying underachievement concept of which it is intended to be the operationalization. I mathematically demonstrate that the regression-based discrepancy method largely reflects achievement-specific determinants, thereby defeating its own object of describing aptitude-achievement discrepancies. The implications for research examining the role of intelligence in learning disabilities are outlined.     

Quality of assessment of prior learning (APL) in university programmes: perceptions of candidates,tutors and assessors

Formal diplomas and certificates have been accepted as proof that students may receive exemption for parts of their educational programme. Nowadays, though, it is socially desirable that informal and non-formal learning experiences are also recognised. Assessment of prior learning (APL) addresses this issue. In APL, the candidate's knowledge, skills or competences required in informal and non-formal learning are measured against a standard to determine whether they match the learning objectives. Although APL is frequently used in workplaces and vocational education, it is practised less in universities, and research is lacking in this context. This study aims to evaluate the first APL procedure in an academic computer science programme, and an adjusted APL procedure in an educational science masters programme. This is done from the perspective of the APL candidates, tutors and assessors, using the theoretical framework by Baartman et al. (2006). The computer science participants comprised 23 candidates from a police software company, four tutors and four assessors. From educational science, nine candidates, two tutors and two assessors participated. The results show that the APL procedure in educational science is viewed significantly more positively than that in computer science; further, the computer science assessors differ considerably from the other participants in their perceptions relating to the quality criterion ‘cognitive complexity’. Explanations for the difference between the two programmes are discussed in this article and assessor and tutor training highly recommended.     

Improving the reading of bisyllabic words that involve context-sensitive spelling rules: focus on successes or on failures?

The effect of two training procedures on the improvement of reading accuracy in poor readers was examined in relation to their initial reading level. A randomized controlled trial was conducted with 60 poor readers. Poor readers were assigned to a control group that received no training, or one of two training conditions. One training concentrated on the words the children read correctly (successes), the other on the words they read incorrectly (failures). They repeatedly read bisyllabic Dutch words, half of the training words involving context-sensitive spelling rules (vowel degemination or consonant gemination). Some children repeatedly read their successes, others their failures. The training used a computerized flashcards format. The exposure duration of the words was varied to maintain the accuracy rate at a constant level. In general, children who received training improved their reading accuracy and reading speed of trained words, and reading accuracy of untrained words, more than the control group. Which training focus is superior, depends on the reading level of the child and the type of words used. For children with a low initial reading level, to improve reading accuracy of bisyllabic words that follow context-sensitive spelling rules, a training focus on failures was superior over a training focus on successes. For children with a high initial reading level, improvement of reading speed was largest in a training on successes. Evidently, the improvement of word reading skills depends both on the children’s level of reading competence and on the type of training.     

Multilevel design efficiency in educational effectiveness research

In educational effectiveness research, multilevel data analyses are often used because research units (most frequently, pupils or teachers) are studied that are nested in groups (schools and classes). This hierarchical data structure complicates designing the study because the structure has to be taken into account when approximating the accuracy of estimation and the power for statistical testing, which should be sufficient to reach meaningful conclusions. Accuracy and power, both referred to as efficiency, can be optimized by carefully choosing the number of units to sample at each of the levels, taking into account the available resources and costs of sampling at these levels. We complement the findings that are found in the literature with regard to designing multilevel studies and propose a simulation approach that can be used to help making study-specific decisions.