On Systematic Reviews in Social Work: Observations from Teaching, Learning and Assessment of Law in Social Work Education

This paper draws on the experience of completing a systematicreview of teaching, learning and assessment of law in socialwork education. It reviews core elements of the process andquestions whether systematic reviews as currently conceivedfor social work education and practice can realize the claimsadvanced on their behalf. The paper considers questions of evidence,quality, knowledge, dissemination and research use, and offersobservations on the potential of systematic review to provideknowledge for policy makers, practitioners, researchers andacademic tutors.     

Evaluation of Laplace distribution-based ANOVA models applied to microarray data

In a microarray experiment, intensity measurements tend to vary due to various systematic and random effects, which enter at the different stages of the measurement process. Common test statistics do not take these effects into account. An alternative is to use, for example, ANOVA models. In many cases, we can, however, not make the assumption of normally distributed error terms. Purdom and Holmes [6 Purdom, E. and Holmes, S. P. 2005. Error distribution for gene expression data. Stat. Appl. Genet. Mol. Biol., 4(1) article 16 [Google Scholar]] have concluded that the distribution of microarray intensity measurements can often be better approximated by a Laplace distribution. In this paper, we consider the analysis of microarray data by using ANOVA models under the assumption of Laplace-distributed error terms. We explain the methodology and discuss problems related to fitting of this type of models. In addition to evaluating the models using several real-life microarray experiments, we conduct a simulation study to investigate different aspects of the models in detail. We find that, while the normal model is less sensitive to model misspecifications, the Laplace model has more power when the data are truly Laplace distributed. However, in the latter situation, neither of the models is able to control the false discovery rate at the pre-specified significance level. This problem is most likely related to sample size issues.