The University of Warwick Q-Step centre research seminar series brings together staff and students from throughout the university who are interested in social science research using quantitative methods. All are welcome.
Autumn Term 2018
30th October 1200-1300 (Room H2.03)
Title: The Linguistic Ideologies of Deep Abusive Language Classification
Michael Castellle (Centre for Interdisciplinary Methodologies, University of Warwick)
This paper brings together theories from sociolinguistics and linguistic anthropology, along with the methodology of neural network-based machine learning, to critically evaluate the so-called “language ideologies”—the set of beliefs and ways of speaking about language—in the practices of abusive language classification in contemporary natural language processing (NLP). This argument is made at both a conceptual and empirical level, as I review approaches to abusive language from different fields, and use two neural network methods to analyze three datasets developed for abusive language classification tasks (drawn from Wikipedia, Facebook, and Stack Overflow). By evaluating and comparing these experimental results, I argue for the importance of incorporating theories of pragmatics and metapragmatics into both the design of classification tasks as well as in machine learning architectures.
Spring Term 2018
31st January 1300-1400 (Room S1.69)
Title: Socio-economic inequality in residential exposure to air pollution and noise
Dr Thomas Verbeek (Department of Sociology, University of Warwick)
Together with a growing empirical evidence on the health effects of environmental impacts, the fair distribution of these impacts across society receives increasing attention. However, today only a limited number of empirical studies exists, with varying results and usually focusing on a single environmental impact. This work presents a spatial data analysis in the city of Ghent (Belgium), combining residential exposure to both air pollution and noise with socio-economic variables and housing variables. Results show that more deprived neighbourhoods, with lower incomes, more unemployment, and a higher percentage of foreign origin people, are more exposed to modelled air pollution, but not to modelled traffic noise. We also found that neighbourhoods with more rental houses, more house moves, a shorter length of residence and lower house prices, bear a higher average exposure to air pollution, and to a much lesser degree to noise. Although the residential exposure variables are based on models and multidimensional analyses might lead to further understanding, the results provide a good starting point for discussions about environmental justice and the need to intervene.
Autumn Term 2017
18th October 1500-1700 (Room R3.41)
Compensating or compounding effects: Can school attended alleviate the effects of family background on academic achievement?
Dr Jenny Chesters (Melbourne Graduate School of Education, University of Melbourne)
Social mobility via education is one of the cornerstones of the Australian democratic state, however, previous research finds evidence of a strong association between parental education and child’s educational achievement and attainment suggesting that social origin plays an integral role in the achievements of successive generations of Australians. Sociologists draw on a range of theoretical perspectives to explain this association including Bourdieu’s cultural and social capital theories. Using data collected by the Longitudinal Survey of Australian Youth 2009 (LSAY09) project, I examine the associations between student SES, school SES and two outcome variables: PISA score and university enrolment. The results indicate that low SES students attending high SES schools perform better on PISA tests than low SES students attending low SES schools. High SES students attending low SES schools perform less well than their high SES peers attending high SES schools. After controlling for PISA score, low SES students were less likely than their high SES peers to enrol at university.
This seminar is organised in collaboration with the Department of Sociology seminar series.
22nd November 1300-1400 (Room S0.09)
Reproducible science is easy and difficult: Species Distribution Modelling and the ZOON project
Greg McInerny (Centre for Interdisciplinary Methodologies)
Like many activities in quantitative science, the massive growth in Species Distribution Modelling (SDM) has been fuelled by the increasing availability of modelling software and online data. By interpolating species’ distributions in terms of environmental variables these models are used to understand the determinants of biodiversity patterns and predict their futures across a wide range of pure and applied areas of ecological and evolutionary science. In this talk I will present our work on the ‘ZOON’ R package that enables reproducible approaches to SDM, and that could help solve some of the key controversies surrounding this set of methodologies. By composing SDM workflows from modifiable modules, ZOON allows users to approach SDM as an open science that can be validated and make advances more rapidly. Whilst the software is ‘user friendly’ and makes open and reproducible science ‘easy’, it also makes research much more difficult. Progress requires that we disentangle the traditional analytical, theoretical and methodological concerns of science from the cultural, social and technological drivers that determine how we approach science. Understanding these ‘unscientific’ components of science, and using them to re-design science, could lead us to a more scientific (and reproducible) approach to computational science.
Spring Term 2017
7th February 1300-1400 (Room S0.13)
The Fifty American States in Space and Time: Applying Conditionally Autoregressive Models
Dr Jamie Monogan (Department of Political Science, University of Georgia)
This paper makes the argument that spatial conditionally autoregressive (CAR) models in a hierarchical Bayesian framework can be informative for understanding American state politics, or any similar population of border-defined observations. Whenever observations are referenced geographically, there is a good possibility that neighbors' error terms correlate. CAR models can account for this, and can easily be included into a wide range of modeling frameworks. This paper shows how to incorporate this kind of error structure into cross-sectional, panel and event history models with applied examples related to public policy liberalism, public opinion on social policy, and the adoption of education lotteries in the 50 American states. With Monte Carlo analyses, we illustrate that as spatial correlation rises, the CAR model offers efficiency gains relative to standard linear models.
Autumn Term 2016