Spotlight series

My teaching philosophy is sharing

I come from a family of teachers and from a young age I aspired to follow my father’s steps, who would bring purpose, moral, and a touch of magic in every topic he was teaching. My path brought me again to LSE, that has a magic of its own, where I can share with my students, my experience, knowledge and indeed love for statistics, and by doing so, help them achieve their dreams and aspirations. The idea of sharing was also the drive to introducing the GTA mentoring scheme, where teaching staff can share experiences, tips and assist one another in improving and developing their skills.

Statistics and mathematics are not just numbers but logic

The most common advice I give to my students is this. Break the problem into small components, visualise how the different parts are connected and follow logical steps to navigate your way towards the final answer. If you can’t solve a difficult problem, solve a simpler one. Ideally, solve many simpler problems. Practice, practice and more practice. You will soon decipher the logic behind the formulae and before long you will have surpassed your target.

As a teacher, witnessing my students’ progress from novices to masters is my best reward.

Statistics is not only relevant but essential in life

During my first class, I challenge my students to find a topic where statistics is not used, or to think how many hours ago they came across a statistical result (inadvertently or not). Throughout my teaching, I try to show the connection between the syllabus material and real life statistical applications. The deeper one understands the former, the broader the scope of the latter. In a rapidly changing work space, statistics has a prominent place. Teaching is not simply showing students the solution to existing problems but mainly inspiring them to invent new problems and quest for innovative solutions.

LSE, is an exciting pool of diverse talents, backgrounds and dreams, and teachers and students together have the opportunity to take part in shaping a bright future.

Teaching is itself a learning experience

Every class teacher goes through a similar process, that includes preparing and presenting a topic in front of an audience. It is natural to feel nervous at first, but with every class the confidence increases. Below are a few tips to make this process easier and more efficient.

1. Go through the same thought process as the students, by solving the problem independently and using first principles.

2. Do a rehearsal of the key points and have a plan, it provides reassurance.

3. Spend a few minutes in the beginning summarising the relevant theory. This makes the connection between lecture and class and helps establish continuity.

4. Give references to real life applications of the topic covered.

5. Do not feel embarrassed when you make a mistake, everyone does, it’s part of the learning process, just thank the student who spotted it.

6. Explain the logic of a formula or a proof, and feel free to skip some of the detailed calculations. Break the question into a small number of logical steps and emphasise the logic behind the solution.

7. Alternate the questions you discuss between homework and unseen. This increases the student involvement and participation.

8. Finish the class by emphasising the importance of the material covered during the class.

9. You are strongly encouraged to attend a class of a more experienced class teacher or lecturer.

Teaching improves important interpersonal skills, like presentation, assertiveness, communication and listening that are extremely useful both inside and outside the classroom. 

It is with great sadness to inform you that David Bartholomew passed away on 16th October 2017. David was an Emeritus Professor of Statistics at LSE. In the recent interview below, he reflected on his research career with Irini Moustaki and Fiona Steele. He made important contributions to diverse areas in statistics, including his highly influential work on social measurement and latent variable modelling, and worked tirelessly to improve the standing of the quantitative approach to social science.

What were the first research topics you investigated and what was the motivation?

My first publications rose out of my PhD project while I was at University College London. After that I spent two years working for the National Coal Board in their Field Investigation Group and, although I had no time for formal research, I met a large number of problems which were clearly in need of some research attention. My first major research project was therefore at the University of Keele on what became known as the “ordered alternatives problem”. I had often met it in problems in the form of a one-way analysis of variance where one had prior information about the order of the means. There was a well-known solution to this problem in the case of a one-tailed test, but there appeared to be no way of dealing with several means.

I worked on the ordered alternatives problem for several years and have good reason to suppose that was instrumental in getting me invited to join Dennis Lindley in his new department at the University College of Wales, Aberystwyth. While there I came under the general influence of Dennis and of the question of how one could incorporate prior information into statistical inference procedures. I was especially interested in non-Bayesian methods of which the ordered alternative problem provided a good example. This work culminated in the book Statistical Inference under Order Restrictions (with R.E. Barlow, J.M. Bremner and H.D. Brunk). While at Keele I was involved in a number of collaborative projects, mainly with social scientists, and these led to publications.

How did your interest in factor analysis begin?

The origin of my interest in factor analysis is very interesting. As a PhD student I had attended lectures at the LSE given by Maurice Kendall in an extremely clear fashion using a geometric approach. Many years after I began to wonder what one would do if variables were not continuous but categorical, and in particular binary. Eventually I decided that the appropriate course to follow was that used in most of statistics: that is, one first proposed the probability model for the experimental/observed situation which one faced, and then produced tests by the usual methods. This led eventually to the general overall model which would cover all cases that I knew of, and this included the case of binary variables as a special case.

In many ways this is a field in which no one offers you any thanks. Psychologists believe that the subject of factor analysis was settled many years ago in a manner which is unsatisfactory to statisticians. Statisticians on the other hand tend to look down on factor analysis as a rather poor relation of principal components analysis. Anyone who attempts to cover the two within the same statistical framework is therefore destined to be ignored by both groups. My impression is that this has happened in my case.

How did you manage to combine and maintain so many strands of research?

I am not aware that I have ever worked simultaneously on two major topics. They have followed one another sequentially in roughly the order of the publications. However, I have often been asked to write particular articles or chapters in books and these have typically been listed in the time when the request came or the deadline to which I was working. So there was no serious conflict between different research projects.

How has your teaching influenced your research, and vice versa?

One thing I rather regret is that I have not been able to teach courses closely related to my research contributions. The latter have obviously affected the slant which I gave to teaching, but teaching was being dictated by the needs of the moment in the place where it happened.

Looking back on your career, which of your contributions to statistical methodology do you consider the most important?

I would, without hesitation, say that the modern approach to factor analysis heads the list. Looking back it is possible to discern lines of development which were not obvious at the time. I think the framework into which they all fit is that of introducing probability reasoning into the more arid branches of descriptive social statistics (beginning with the book Stochastic Models for Social Processes). My intention throughout all this was to make social statistics a branch of Statistics which would bear comparison with biological statistics and other major divisions, at least. Until recently this was certainly not the case and I hope in the future that intention will be more clearly obvious.

David's former LSE colleagues Celia Swan and Martin Knott have written an obituary which will be published in the Journal of the Royal Statistical Society. 

Henry Wynn is an Emeritus Professor of Statistics at the LSE. Please have a look at the YouTube video of his interview. 

Howell Tong has been an Emeritus Professor of Statistics at the London School of Economics since 1st October 2009. Please find an interview with Howell conducted by Kung-Sik Chan and Qiwei Yao.

How would you describe your research problem to general audience? 

In this information age featured with big data, various forms of observation are recorded over time at many locations or for many individuals. At each given time point, the recorded data can be a long vector, a curve, a space, or a network. My major research interest is to analyse those complex time series for better understanding the underlying dynamics and for forecasting accurately the future values.

How did your research influence the field? 

I try to transform a complex problem into a simpler one or several simple ones. One approach is to find a latent and low-dimensional structure in a high-dimensional problem. This has been proved effective in analysing high-dimensional time series, high-dimensional and high-frequency financial returns, spatial-temporal data, and curve time series.

What new results have you established? Did it change your work? 

The newly proposed time series principal component analysis (TS-PCA) should have wide impact in analysing multiple time series, especially when the dimension of time series is high. It transforms a multiple time seires into several low dimensional uncorrelated series (across all time lags), leading to better influence and better future prediction. I expect TS-PCA to be used as a standard pre-step in analysing multiple time series.

How does your work is used by outside the academic sector? 

I enjoy working with industry. In addition to the satisfaction of solving reallife problems,  it also leads to interesting, challenging and non-standard research questions. I have written several research papers through working together with EDF and Barclays Bank on their business problems. Each of them was great fun!

How would you describe your research in your own words?

I am deeply interested in inter-disciplinary work. For me, research makes sense if I am at the boundary between different disciplines. I have been lucky to work in a wide range of areas, such as economics, finance and more recently the environment. What I particularly appreciate in my research is to see what happens when these areas meet.

Recently, my interest has been about understanding "what is a model". Not necessarily from a statistician’s point of view but more generally. What does a model mean for different fields of research? How do they approach a problem and develop a model to solve it? In this sense, I'm a bit of an outlier, reaching researchers outside of my own field.

A problem encountered by people working on a trans-disciplinary project is that concepts have different meanings depending on your area of expertise. Other scientists will have a different understanding than yours; the vocabulary is not the same for a biologist, an economist or a physicist. Therefore, a lot of time is spent going back to that point, communication, and then you increase your knowledge about other disciplines.

When it works, it challenges the way you think and generates something rich and useful.

How do you export the statistical tools used for the financial market to environmental management?

In a way, looking at the things I have been doing in my research requires thinking about management of risk as well as the decision process. Looking at how to transfer risks from the insurance sector to the financial market and manage them in a better way. Regarding environmental issues, it is also very much about considering the risk you are taking to develop a policy that can lead to an improvement. That is why it goes beyond finance and insurance but also the environment. It is more general.

For a couple of years now, I have been working on a project precisely on this, where I am looking at the very definitions of model, risk and uncertainty in different areas. I have been interviewing researchers from diverse fields, such as statistics, cosmology, biology, chemistry, medical research or drug design. At some point it always comes down to the same idea: to consider a risk and take a decision. 

How do you see research?

For me, research is influenced by the way you are trained. A PhD student will follow a supervisor who might shape his vision of research. Mine had a very sociable approach, and through her I learned how to do research as a sort of nice human experience. Nowadays, a research project can easily start from a conversation with someone I get along with. This is what happened for the project on environmental management; I met this person, an environmental economist, who was very nice and we started this project together.

I don't dissociate the process of doing research from the human experience behind it, and I get a great deal of learning from each other. In a way, what really matters to me, more than the direct outputs, is the experience I get through it.

Professor Pauline Barrieu is the Head of Department of Statistics.  

How would you describe your current research?

A common theme is the statistical methodology of sample surveys and government statistics. I’ve always enjoyed engaging with real world challenges in these areas. One current example is work with the Department of Health on ways to collect sample data to obtain a more accurate statistical basis for the reimbursement of community pharmacies dispensing medicines in the NHS. A second is to look at ways in which the population census can go beyond traditional methods to make use of new data sources such as administrative records. But I’ve also always enjoyed developing the underlying statistical theory and methods associated with such challenges, for example in the area of data privacy.

How has your work in data privacy developed?

It originated decades ago helping a group of social scientists make the case to government to release individual-level data from the GB census. The key obstacle to such release was whether this might breach confidentiality undertakings to people who completed their census forms. My role was to develop a systematic statistical way of assessing the risk of such a breach. Since then, whilst continuing to engage in practice, I’ve developed a theoretical basis for disclosure risk assessment.

More recently, there have been major developments in data privacy in computer science, with much work going on in places like Microsoft and Google, and there has been a need for methods to be integrated within the overarching field of data science. Last autumn I co-organised a six month programme at the Isaac Newton Institute for Mathematical Sciences in Cambridge, to bring together scholars from statistics, computer science and other disciplines to share ideas and develop a more unified theoretical basis for the field. Bridging disciplinary differences has also been an aim of the Journal of Privacy and Confidentiality, founded eight years ago and for which I served as one of the original editors.

And how about the census?

I was asked by the Royal Statistical Society to lead an independent review of the methodological options for the 2011 Census in England and Wales. This led me to look more into new developments in census methodology. Since there is usually only one census per country, international comparisons are essential to understand such developments. I have recently been looking more into this area and am currently preparing a high-profile lecture on international developments in census taking, to present in Washington DC later this year.

What are the big challenges in your field?

The demand for statistical data is ever-growing. But primary data collection faces growing costs and a relentless increase in "non-response" from people declining to take part. This raises key challenges in how to combine alternative types of data sources, including primary sources, administrative records and ‘big data’. Data privacy challenges are also likely to grow, in particular the challenge of how to protect the privacy of data used for research purposes, without damaging the value of the data to researchers.

Professor Chris Skinner is a Professor in the Department of Statistics.  

How would you describe your research to a general audience?

My recent research has focused on methods for analysing longitudinal data from surveys that track individuals over time. There are many sources of longitudinal data, including large-scale birth cohort surveys and household panel surveys that follow individuals over many years, and studies that collect ‘high frequency’ data (e.g. daily or weekly) over a shorter time period. The most appropriate study design depends on the research question of interest, and each has its own challenges when it comes to data analysis. A particular issue with any longitudinal design is that data are clustered from having repeated observations on the same individual, which means that standard statistical methods need to be adapted. One methodological challenge that interests me is how to account for household or area effects on individual behaviour when individuals are clustered in households and areas, but household membership and area of residence change over time. Another active area of research is how to model the complex interrelationships between the various types of decisions that people make over time, for example decisions about if and when to have children and when to move house.

The methodological topics I work on are motivated by real-world applications. Most of my research has been in demography, but I have also worked in health, psychology and education, most often with social scientists who provide subject-matter expertise. For me, one of the main attractions of social statistics is the wide range of application areas that we can research and learn about, and the availability of rich datasets.

My most recent project (co-funded by the Economic and Social Research Council and the Engineering and Physical Sciences Research Council) involves developing methods for the analysis of longitudinal data on pairs of individuals (or ‘dyads’). Such data are commonly collected in social research, for example on pairs of members of the same family to assess agreement in perceptions of the quality of their relationship. We will be applying these methods in a study of exchanges of support between parents and their adult children, using UK data from the Understanding Society survey. The research team includes social statisticians from the Department of Statistics and colleagues from the Department of Social Policy at LSE and the Institute for Social and Economic Research at the University of Essex.

How has your work been used by other researchers?

My work on methods for analysing multilevel time-to-event data has been widely used by social scientists and researchers from other disciplines. These data are highly complex because individuals can experience an ‘event’ (e.g. birth or divorce) more than once, they may move between different ‘states’ (e.g. employment and unemployment) over time, and there may be different reasons for moving out of a state (e.g. reasons for leaving a job). These methods have found applications in a range of areas, including sociology (e.g. the evolution of social networks), business (e.g. decisions about the location of multinational enterprises), and veterinary medicine (e.g. transitions between foot conformation, lameness and footrot in lambs).

A key factor in encouraging other researchers to adopt a new approach is to explain it in an accessible way, with examples, and to show how it can be implemented in software. Capacity building has been an important part of my work and can take many forms, such as tutorial articles in applied statistics and social science journals that explain how to apply a particular method (and interpret the results), online training materials, and traditional face-to-face training workshops. For example, as part of the LEMMA project (funded through the ESRC National Centre for Research Methods) I led the development of an extensive online training course in multilevel modelling. The course now has over 20,000 registered users, with a high proportion of students (60%) and many from overseas (75%).

And outside the academic sector?

My research has been used primarily by academic social scientists, and so its impact outside the academic sector is largely indirect through the substantive findings of academics who apply the methods in their own research.

One example of research that has had a more direct impact is a recent study on whether intentions, beliefs and mood following acute coronary syndrome predicts patients’ subsequent attendance at outpatient cardiac rehabilitation (CR). The study, led by healthcare researchers from the Universities of Dundee and Aberdeen, involved structural equation modelling of real-time diary data on cardiac-related beliefs and mood collected using an electronic diary. The findings have influenced work on the redesign of CR through the 2020 vision for CR.

Another way that my research has been taken up outside the academic sector is through the provision of on-line training materials. A large number of users (15%) of users of the LEMMA online course are non-academics, while attendees at face-to-face tend to be academics. It seems that the flexibility of the online format is especially attractive to non-academics. 

Professor Fiona Steele is a Professor in the Department of Statistics.