Departement Statistiek en Aktuariële Wetenskap Department of Statistics and Actuarial Science

Nuusbrief / Newsletter

September/October 2015

Teken aan! Die gewilde klasdag vir alumni van die Fak ulte it Ek onomie se e n Be stuurswetenskappe word op Donderdagmiddag, 10 Desember 2015, van 13: 00 tot 16: 00 by die CGW Schumanngebou op Stellenbosch gehou. A finger lunch will be followed by a variety of interesting (parallel) presentations by experts in the Faculty. Thereafter Dr Monique Nsanzabaganwa, Deputy Governor of the National Bank of Rwanda and former Rwandan Minister of State in charge of economic planning, will deliver the keynote address. Monique het in 2002 die eerste swart student van buite Suid-Afrika geword om haar meestersgraad in Ekonomie cum laude aan die Universiteit Stellenbosch te verwerf. In 2012 he t sy ook haar P hD -graad in Ekononomie aan US voltooi. Na afloop van haar Magisterstudies het sy 'n pos aanvaar as dosent by haar alma mater, die Nasionale Universiteit van Rwanda. Die volgende jaar is sy aangestel as minister verantwoordelik vir ekonomiese beplanning in haar land se ministerie van finansies.

In this issue Departement kry opgegradeerde kantore in 70 ste bestaansjaar Voorgraadse pryswenners van 2014 Voor- en nagraadse studente van 2015 Focus on alumni: Greetings from Stephen Reid in the USA Students receive prestigious bursaries and grants My route to Statistics by Trudie Sandrock Data Science in the spotlight Actuarial Science research contributes to financial industry Students participate in global research challenge Excellence in retirement fund industry awarded Personeel woon internasionale konferensies by Department ‘ecstatic’ about first prize in SASA competition Diary: Upcoming seminars Garret Slattery helps win international blind golf tournament The modern day statistics toolbox by Paul Mostert Papers co-authored by staff

Departement kry opgegradeerde kantore in 70ste bestaansjaar Hierdie is die eerste nuusbrief van die Departement Statistiek en Aktuariële Wetenskap en val saam met die opbou na die departement se 70ste verjaarsdag in 2016, sowel as die verhuising na die departement se nuut opgegradeerde fasiliteite.

Studente in die nuwe ontvangsarea van die departement

Dosente vergader in een van die wegbreekareas

Personeel verkeer gesellig in die moderne nuwe personeelkamer.

Geskiedenis word gemaak met die eerste lesing in die nuwe voorgraadse lesinglokaal met 110 sitplekke. Op die foto is die Finansiële Risikobestuur 242klas met die dosente, prof. WJ Conradie en dr. JD van Heerden in die agtergrond. In die voorgrond is die tolk.

waarin dosente, studente-dosente, nagraadse studente en besoekers gehuisves kan word, ‘n goed ingerigte administratiewe en ontvangsarea, drie volledig elektronies toegeruste nagraadse lokale en een middelslag lokaal met 110 sitplekke (vir voorgraadse modules), twee wegbreekareas, asook ‘n kleiner rekenaargebruikersarea wat slegs tot die beskikking van die nagraadse studente in die departement is. Dit het nie lank geneem vir die studente om laasgenoemde lokaal tot STARGA te doop nie – om te rym met FHARGA.

Alhoewel Statistiek as vak aan die US reeds in die 1920’s aan landboustudente doseer is en die eerste Statistiekkursus in 1933 vir sak e stude nte aange bie d is, he t die Departement Statistiek formeel tot stand gekom in 1946 met die aanstelling van Statistiek as vakdissipline is tans een van die prof. Faantjie Pretorius. Oor die afgelope 70 mees dinamiese dissiplines en statistiese jaar het die Departement gegroei tot die k undigheid is wêreldwyd in groot derde grootste in die Fakulteit Ekonomiese aanvraag; trouens, daar is regoor die en Bestuurs-wetenskappe. Sedert 1985 wê re ld ’n groot sk aarste aan goe d word Aktuariële Wetenskap op voor- en opgeleide statistici. In Suid-Afrika het dit tot nagraadse vlak aangebied en as gevolg van gevolg dat die Departement van Wetenskap die groei in die belangrikheid en status van en Tegnologie Statistiek as ’n “vulnerable opleiding in A k tuariële discipline” identifiseer het Wetenskap, is die en substansiële fondse via departement se naam in die NNS in die vorm van Hierdie is die eerste 1997 na die Departement nagraadse beurse, postvan Statistiek en doc toekennings en nuusbrief van die A k tuarië le We te nsk ap navorsingsondersteuning Departement Statistiek verander. Sedert 2000 beskikbaar gestel het vir word ook programme in die bou van statistieken Aktuariële Finansiële Risikobestuur k apasiteit (later meer Wetenskap en val saam aangebied. hieroor).

met die opbou na die Vanjaar beloop die totale Moderne tegnologie het aantal voorgraadse tot gevolg dat al hoe departement se 70ste inskrywings vir die meer data besk ik baar verjaarsdag in 2016. verskillende modules in w o r d (“ t h e b i g d a t a die departement sowat revolution”). Hierdie is ’n 5 500. Dit sluit studente in realiteit in die sak e-, die vier hoofvakke in die departement in, finansiële, beleggingsen naamlik Statistiek, Wiskundige Statistiek, versekeringswêrelde, asook in die A k tuarië le We te nsk ap e n Finansië le nywerheids- en landbou-industrieë – oral Risikobestuur, sowel as dienskursusse in die waar die departement se afgestudeerdes in Fakulteite van Ekonomiese en aanvraag is. Om in pas te bly met moderne Bestuurswetenskappe en Ingenieurswese. statistiek-metodologie en ’n geleentheid te Die departement bedien vanjaar ongeveer skep vir studente om gepaste opleiding te 120 nagraadse studente wat ingeskryf is vir ondergaan, begin die departement in die verskillende Honneurs-, Magister- en samewerking met die Departement van PhD -programme . Die aanv raag na Rekenaarwetenskap in 2016 met ’n nuwe nagraadse opleiding het die afgelope paar honneursprogram in Datawetenskap (meer jaar betekenisvol gegroei. Verder huisves hieroor later). die departement ook die Statistiese Statistiek gesproke, kry die departement sy Konsultasiesentrum wat navorsers aan die nuwe fasiliteite in ’n opwindende tyd. Die US ondersteun. departement staan voor interessante Dit is dus gepas dat die departement in sy 70ste jaar sy nuwe, opgegradeerde kantore in die Statistiek en Rekeningkundegebou kon betrek nadat dit tydelik ‘n jaar lank elders op kampus gehuisves was. Vir die eerste keer in die departement se bestaan word oor fasiliteite beskik wat aan al sy behoeftes voldoen. Dit sluit in voldoende kantore

uitdagings om gepaste en goeie opleiding in die dissiplines Statistiek, Aktuariële Wetenskap en Finansiële Risikobestuur volhoubaar aan te bied, asook om kwaliteit navorsing in hierdie velde te doen. Dit is voorwaar ’n lekker uitdaging wat nou in ’n esteties aangename werk somgewing aangepak kan word.

2

Voorgraadse pryswenners van 2014 Tydens die voorgraadse pryswennersfunksie in Oktober 2014 is die beste derdejaarstudente in die department aangewys: u

Tess Rodseth in Statistiek,

u

Marguerite Bezuidenhout in Finansiële Risikobestuur, en

u

Charl du Plessis in Wiskundige Statistiek en Aktuariële Wetenskap.

Hier volg meer inligting oor dié drie begaafde studente.

Tess Rodseth: Best third-year student in Statistics Where did you go to school? I attended Durban Girls’ College from Grade 1 to Matric.

Why did you decide to study Statistics and what do you like most about it? I never formally “decided” to study Statistics, it was something that sort of just happened. I have always been more of a “numbers” person than a “words” person, so I knew I wanted to go into something that had a mathematical foundation. I really enjoy how useful statistics is, and have begun to realise that throughout my BCom degree as it was incorporated into each and every one of my subjects.

What degree are you currently doing and why did you decide to study in this department? I am currently studying BCom Honours in Statistics. After finishing my undergraduate BCom in Statistics and Investments, I realised that there was so much more that I wanted to learn about statistics. In addition, I knew that an Honours in Statistics would prove to be extremely beneficial to my career.

What do you like most about the department and the postgraduate programme you are doing? I really enjoy how friendly, passionate and helpful the Statistics Department is regarding their students’ needs. In my opinion, the Statistics honours courses have successfully incorporated technology into the postgraduate programme, and I have learnt valuable computer skills that will be used frequently and are in high demand in the workplace.

Do you think Statistics is a useful subject that can bring change to our society at large? Why? I think that Statistics is a very useful subject, and one that does not receive the credit that it deserves. Not only is statistics used in most academic fields ranging from finance to medicine, but it is becoming a very important tool due to the rapid increase in technology and social media creating large amounts of data that need to be analysed.

What are your future career plans? I am currently undecided on future career plans, but I hope to go into the statistical side of the financial world. I am at present also considering studying further in Statistics.

What are some of your hobbies, other interests and achievements? Being from Durban, I enjoy going to the beach. I also enjoy playing netball and I have a passion for the bush and wildlife. Graduating from Stellenbosch University with a BCom Cum Laude was one of my greatest achievements to date.

3

Marguerite Bezuidenhout: Best third-year student in Financial Risk Management Where did you grow up? I grew up in Pretoria, in the area known as the Moot. It is known as a good community with friendly and loyal people. I had a really good upbringing and I will always call it home.

Where did you go to school? I went to Oos-Moot High School. It was the place that taught me everything I had to know before embarking on my studies here at Stellenbosch University. My high school had really strongminded and hardworking teachers and without them I would not be where I am today.

Why did you decide to study Mathematical Statistics /Financial Risk Management and what do you like most about it? I chose to study in these fields because I have a love for mathematics and I enjoy solving problems. I always knew I was going to work in an environment one day that requires problemsolving and strong analytical skills. There is also a high demand in the business world for individuals that have the specific skills that these courses can provide. There are risks all around us, and we have to make choices while taking into account all of the risks incurred to arrive at an optimal solution. Financial risk management sheds light on how to quantify risks in the financial world. The concept of how to quantify uncertainty was always a very interesting concept for me and therefore I chose to study in a field that required a statistical background. Statistics can be applied to different areas such as the financial industry, medical industry and technology. The benefits of having a statistical and mathematical background will, I believe, count in a person’s favour. I enjoyed everything about my degree. The department incorporates so many interesting modules that taught me everything I was always wondering about.

What degree are you currently doing and why did you decide to study in this department? I am currently studying towards my honours degree in Financial Risk Management. The financial industry has always intrigued and fascinated me. Financial risk management teaches you about how to price derivatives, how hedging strategies are implemented, how to take account for certain risks in investments and how to build appropriate models for pricing securities (to name a few concepts). Studying Financial Risk Management, you also gain knowledge in the fields of statistics and financial mathematics. This really helps an individual understand the more complex mathematics behind every-day concepts used in the financial industry. The financial industry is always changing and for this reason a person working in this field will constantly have to think of new ideas and models to adapt to changes. This is also one of the key reasons for my choice, because I believe that it will always be a demanding and very interesting field of work.

What do you like most about the department and the postgraduate programme you are doing? The department has a really strong structure in place, with an exceptionally talented group of lecturers that are very friendly

and loyal, with broad knowledge and experience in these fields. They took the time to think about what skills a postgraduate student might need in order to evolve their knowledge for the business world. My postgraduate studies have taught me so many new skills that will only make life easier in future. I love that it is a very personal level of teaching and that students are not too intimidated to ask questions or understand concepts better. It is a very demanding degree, thus a student will definitely learn how to prioritise and use time optimally. We are also introduced to a lot of computer software and how to code, which is very interesting and definitely a skill worth obtaining in the world we live in today.

Do you think Mathematical Statistics/Financial Risk Management are useful subjects that can bring change to our society at large? Why? Yes. I definitely believe that studying in one of these fields can bring forth change. These fields broaden your knowledge on various concepts that are used in everyday life. It also helps you learn to think in a different way, and teaches you to solve exciting problems by implementing new solutions. I believe that individuals with fresh young minds and the appropriate skills are always needed in this world, since it is constantly changing. Having obtained the appropriate skills, an individual can make better informed decisions if they properly understand all consequences.

What are your future career plans? I definitely see myself working in the financial sector, most probably in the banking industry. I am planning on moving to Johannesburg to start off my career closer to home. Hopefully I will become a successful financial risk manager of one of the top five South African banks. If I were to come across the opportunity to work abroad, I will most definitely consider the offer.

What are some of your hobbies, other interests and achievements? I love music; it helps me focus and relax, therefore I enjoy a good concert or going to a music festival. I am very outgoing and I love being outdoors and exploring. One of my favourite hobbies is photography, capturing the moments that makes life worthwhile. I am always taking nonstop photos of everything. I love a good laugh and I also love telling stories.

4

Charl du Plessis: Best third-year student in Mathematical Statistics and Actuarial Science Where did you grow up? I am originally from Bloemfontein but for the most part I grew up in Durbanville in the northern suburbs of Cape Town.

Where did you go to school? I matriculated at Stellenberg High School in Durbanville.

Why did you decide to study Mathematical Statistics/Actuarial Science and what do you like most about it? I have always wanted to use my analytical abilities to solve real-world problems. I believed that the background gained in Mathematical Statistics and Actuarial Science would give me the potential to add value in any quantitative career. The feature that I like most about it is that the learning material is intellectually stimulating and always challenges your thinking.

What degree are you currently doing and why did you decide to study in this departement? I am currently doing my honours degree in Actuarial Science. I believe that studying in this departement will equip me with the necessary fundamental knowledge and skills to be successful in the working world as a quantitative professional.

What do you like most about the departement and the postgraduate programme you are doing? The postgraduate courses offered by the departement provide comprehensive training that further develop our ability to solve real-world problems. The honours programme in Actuarial Science is challenging, but very enlightening. In particular, the programme develops strategic and broad thinking skills which, I believe, will be valuable in a practical working environment.

Do you think Mathematical Statistics and Actuarial Science are useful subjects that can bring change to our society at large? Why? Yes. These subjects challenge us to think critically and so develop our logical problem-solving skills. They further allow us to make better sense of data and improve our understanding of financial risks. This gives us the potential to tackle complex issues and develop innovative solutions to real-world problems.

What are your future career plans? I will endeavour to follow an innovative actuarial career. Initially I wish to gain a broad overview of developments in the risk industry and exposure to the range of opportunities offered. I have a particular interest in the development and programming of risk models.

What are some of your hobbies, other interests and achievements? I enjoy reading both fiction and non-fiction books, computer programming, fitness (including running and endurance training), watching movies and certain sports, gaming, and sightseeing.

5

Voor- en nagraadse studente van 2015

Third-year Actuarial Science class. Three years ago, more than 60 first-years started the course. Dedication and hard work have brought these students to where they are.

Honneursstudente in Wiskundige Statistiek saam met een van hulle dosente, prof. Danie Uys (heel links agter)

The wait is almost over. The Mathematical Statistics 318 class is overjoyed as the long and challenging first semester draws to an end. Now all that’s left to do, is to ace the exams!

Honneursstudente in Aktuariële Wetenskap saam met een van hulle dosente, mnr. Rob Clover (agtste van links, heel agter)

Derdejaarstudente in Finansiële Risikobestuur saam met hul dosent, prof. Willie Conradie.

‘n Groep nagraadse studente in Finansiële Risikobestuur saam met een van hulle dosente, dr. JD van Heerden (heel regs agter)

Derdejaar- Statistiekstudente saam met hul tweede semester dosente, dr. Morné Lamont en prof. Tertius de Wet (heel agter).

Honneursstudente in Statistiek

6

Focus On Alumni Greetings from Stephen Reid I suppose that the harebrained scheme came to nascency one Monday morning somewhere in the autumn of 2009. Professor Steel was in the front of the class, teaching a handful of us some statistical learning theory. We were covering chapter 10 of the Elements of Statistical Learning – a particularly colourful chapter in the book. The season’s last birds were chirping in the oaks outside and the crisp morning air carried the varied mumbles of the students passing between classes downstairs. We were roughly halfway through one of those epically attritional lectures so favoured for postgrad tuition. The wooden desks were, as ever, creaky and reassuringly uncomfortable. Despite the general contentedness of my time in Stellenbosch, I remember those heady days not for a low grade, but unignorable unease. My thoughts kept returning to what my mother had told me one day, many, many years before. Having listened to her son’s particularly vehement railing against the unthinking betrayal of a young crush, she uttered a seemingly innocuous epigram. In reference to the other young lothario in question – the one who had so engaged my young crush’s affections – she merely said, “Don’t worry, Son. He is a large fish in a small pond.” Of course the image is evocative and instantly brought me solace. For surely such a delusional individual deserves pity, not envy. Moments later, I realised, to my ongoing chagrin, that I would forever fear such complacency. The goal became always, always, to be striving for that larger pond. Perhaps it was that life in Stellenbosch was becoming too comfortable, despite those wooden class benches. The education I was receiving was truly top class. My pedagogues – both in the Economics and Statistics & Actuarial Science departments – had, many of them, spent time at some of the leading institutions in their fields. Some were champions of local industry; others eminently well regarded academics; others still, local doyens of their professions; some were just talented teachers. They taught me from the same textbooks taught in the halls of the truly lofty academic institutions, but did so by imbuing it with their own understanding and experience. They were thorough and proceeded methodically. I try to replicate their approach here at Stanford. Even here – especially here, where days often pass too frenetically – their teaching ethic is greatly appreciated. I tell my students that I am merely relaying the way I was taught. continues on next page

Stephen Reid In 2005, Stephen enrolled as a first-year BCom (Actuarial Science) student at SU. In 2007 he obtained his BCom degree Cum Laude with distinctions in all his modules over all three years. (His lowest mark was 83 %.) He received prizes for the best third-year student in Mathematical Statistics and Actuarial Science in 2007. He went on to obtain his BCom Honours degree in Actuarial Science Cum Laude the next year. Due to his excellent study record, he was awarded the Chancellor’s medal for the most deserving Matie who graduated in 2008. In 2009 and 2010 respectively he obtained BCom Honours in Mathe matical Statistics and Economics C um Laude , and MC om in Mathematical Statistics, also Cum Laude. Stephen may certainly be regarded as one of the best students in the history of the SU Department of Statistics and Actuarial Science.

Stephen received prizes for the best third-year student in Mathematical Statistics and Actuarial Science in 2007 from the then dean of the Faculty of Economic and Management Sciences, Prof Johan de Villiers.

In 2011 Stephen enrolled as a PhD student in the Department of Statistics at Stanford University, California, USA, one of the top ivy-league Statistics Departments in the world. His supervisor for his PhD study is well-known Prof Robert Tibshirani.

Stephen with his supervisor, Prof Robert Tibshirani, in front of the Statistics Department at Stanford University.

7

Of course Stellenbosch was itself beautiful and had numerous other distractions. Life was good there. My future was bright. Perhaps then it was not Stellenbosch that was lacking, but rather that the lure of Stanford, as promised in those glossy pages, was simply too great. I could not explain exactly why, but the idea of making pictures like the ones found on those pages, sounded like “what I wanted to do with my life”. Quite why it needed to be at Stanford, I do not know. Maybe because the book was written there. Maybe because it just seemed like a really big pond. Either way, the harebrained idea had been implanted. With confidence and knowledge instilled by my influences at Stellenbosch, I was to take on this new challenge. Ultimately, I suppose, it was a product of a young man’s ambition and the power of a mother’s unintended wisdom. I arrived at Stanford in September 2011 and was instantly overwhelmed. It is an exciting and intensely scary place. The students are bright; the faculty gifted and world-renowned. One has to be on top form every day just to keep up with the pace. Most days are more challenging than I would like. Still, I do not think I’ll be happier elsewhere. Collaboration between departments (world-class ones like computer science, statistics, electrical engineering, biology and medicine) is entrenched, probably more so than back home. It seems a little easier to push the very cutting edge of knowledge if others are there, pushing from their own sides. Eventually I found my way to Rob Tibshirani’s office and he agreed to be my advisor. We share a similar philosophy that stats is ultimately there for real world application. He values good practical ideas over the grinding, squiggly theory. That comes later. Someone else usually does that. His preeminence as a researcher ensures a steady stream of datasets from various sources. Every now and then he kicks some of them to me. I try to hold my own and come up with something useful. He has not complained yet. His support is stoic and unwavering. I owe him indescribably much. Those happy days in Stellenbosch are sorely missed. The winter rains (especially here in drought-ridden California), De Akker (with its sticky floors, smoky pall and undeniable character), the warmth of the people back home: I find those lacking here. Silicon Valley is fast-paced. The highways are wider. There is more internet bandwidth. Fortunes are made and lost. Sometimes the human touch is lost in the gold rush. This summer, I start work at a data science consulting startup. I do not know exactly what they are going to do, but it sounds something like what got me so excited that day, in a Stellenbosch autumn, some years ago. Maybe they will even give me some equity. The prospect is rather exciting. Watch this space. Perhaps soon I’ll be hiring statisticians for the next big thing in data science consulting. I favour South African ones. Initially, I had no idea what I was doing at Stellenbosch. Not really. I was 19 years old. Like many, I stumbled into my course out of high school on the lure of it being for those who are “good with numbers”. Professor Slattery tried his best to disabuse us of our illusions early on – tried to tell us that it was about so much more than just numbers – but, you know, by then we were committed. Despite my ignorance, I learned much from my time there. I daresay I grew up a little bit. Hardly any courses or experiences were truly wasteful. The rest have woven themselves into a rich tapestry of experience, knowledge, critical thinking and resourcefulness under pressure. It has prepared me for the challenges I face here on the far side of the world, pursuing my succession of harebrained ideas. Thank you all for that. Greetings, from the far side of the world, Stephen Reid MCom (Mathematical Statistics) 2010

8

Students Receive Prestigious Bursaries And Grants SASA-NRF Crisis in Academic Statistics grants In 2014, the national Department of Science and Technology (DST) classified Statistics as a “vulnerable discipline” in South Africa due to the scarcity of wellqualified professionals and academics in this field. As a result, R6 m was allocated to building capacity in statistics. Administered by the NRF and the South African Statistical Association, the fund is used for full-time bursaries to Masters and PhD students, as well as grants to postdoctoral students and postgraduate

supervisors. (The purpose of postgraduate supervisors is to assist academic departments in appointing additional supervisors.) Doctoral students receive R150 000 per year for a three-year period, and Masters students receive R100 000 per year for two years. The only criterion to obtain a bursary is academic merit. Our department is very proud of the following students who have obtained bursaries in 2015: Masters students in Mathematical Statistics Chané Orsmond, Dr Margaret de Villiers, and Sven Buitendag; PhD student Mr Francois Kamper; and Dr Trudie Sandrock who has received a postdoctoral grant.

Media 24-beurse

PSG graduate programme

Magisterstudente Arnu Pretorius (Wiskundige Statistiek) en Monika d u To i t ( S t a t i s t i e k ) h e t ' n besondere prestasie behaal deur prestigeryke Media 24-beurse te verwerf. Magisterstudente oor alle dissiplines aan US kom vir hierdie beurse in aanmerking; derhalwe is die kompetisie besonder straf.

Initiated at the end of 2014, PSG's graduate programme offers a bursary towards completing a student's honours year, followed by an internship opportunity. One of the first recipients of this bursary is Duayne le Roux, an honours student in Mathematical Statistics. His participation in the PSG graduate programme has already allowed him to gain invaluable experience under the mentorship of Greg Hopkins, PSG Chief Information Officer. A ccording to Duayne, “The experience has been very helpful. It focuses on you as future working professional and specifically on placing you, your talents and abilities in the area of the company where you will be most successful. While the exposure requires effort, plenty of reading and considered thought, it comes with the satisfaction that your efforts have real implications.”

NRF- en Media 24beurshouers

Voor (v.l.n.r.) is dr. Margaret de Villiers, Charné Orsmond en Monika du Toit en agter is Sven Buite ndag (link s) e n A rnu Pretorius (regs).

Faantjie en Lettie Pretoriusbeurs vir voorgraadse studente Die opbrengs van 'n aansienlike skenking deur die erfgename van prof. Faantjie Pretorius, die eerste professor in Statistiek aan US, word jaarliks gebruik om 'n beurs aan ‘n verdienstelike voorgraadse student in die departement toe te ken. Die eerste ontvanger van die beurs (in 2014) was Liezel Möller, toe ‘n statistiekstudent in haar derde jaar. Liezel is tans besig met haar Honneursgraad in Statistiek. Vanjaar deel twee BCom (Aktuariële We te nsk ap) stude nte , Briance Ma t h e b u l a e n C h r i s v a n de r Westhuizen, die beurs.

Die Faantjie en Lettie Pretoriusbeurshouers is (v.l.n.r.) Briance Mathebula, Chris van der Westhuizen en Liezel Möller.

9

My route to Statistics by Trudie Sandrock, recipient of a NRF postdoctoral grant I am currently doing postdoctoral research in Statistics at S t e l l e n b o s c h University. However, I have to confe ss: I never intended to become a statistician! I left school not really knowing what statistics was, and with my mind set on becoming an actuary. After a BCom (Maths and Stats) at NMMU, I started work as an actuarial assistant but was sufficiently intrigued by statistics to take the “scenic route” to becoming an actuary. I therefore enrolled for an Honours degree in Statistics at Unisa before starting any actuarial exams. Soon I was hooked by Statistics and all thoughts of an actuarial career flew out the window. For large parts, my career path did not involve much formal statistics. I worked in data analysis roles at Sanlam, Investec and a financial services company in the United Kingdom. This meant that I got to dabble in diverse fields, including Geographical Information Systems (GIS) and market research, e.g. client satisfaction surveys and developing investor confidence indices. In terms of more formal statistical applications, I looked at datamining problems such as the prediction of life policy lapses and helping to target direct marketing campaigns and improve response rates. For a number of years I was also a full-time stayat-home mom, and for days on end my statistical undertakings were confined to estimating the probability that all three of my children would sleep through on any given night (spoiler alert: this probability was – and still is – miniscule). My current field of research is multi-label classification, and I was led in this direction by my desire to combine my love of music with my love of statistics. For my PhD dissertation, I specifically looked at the problem of polyphonic musical instrument recognition. Being able to combine two such diverse passions is what I love most about statistics, and John Tukey summed it up perfectly when he said, “The best thing about being a statistician is that you get to play in everyone else's backyard.”

My advice to students considering a corporate career in statistics is: Don't expect to be doing fancy statistical calculations all the time. If you pursue a career in data science, you might end up doing formal statistics about 20% of the time, while the rest of the time you might function as a glorified data “panel beater”, trying to beat your data into submission. Embrace that, rather than becoming frustrated; you learn a lot through the process. Along the same vein, work with actual “dirty” datasets every chance you get. In the real world, datasets are seldom as cle an as the be nchmark datasets you typically work with at university. Also, you won't get pointers regarding which statistical technique to use when. Find datasets on the internet (Kaggle is a great source) and play with the data; you will gain invaluable experience. Hone your presentation skills, both written and oral. Practise this as much as you can; what will set you apart from other people with the same qualification as you, is an understanding of business and crucially, the ability to break down complex information and figures in a meaningful way and to present it in such a way that people with little or no background in statistics can understand it and act on it.

10

Data Science in the spotlight As the amount of data in the world continue s to grow, the de mand f or individuals who are skilled in analysing and using large-scale data is not only high but on the increase. A report by global management consulting firm McKinsey states that “…by 2018, the United States alone could face a shortage of 140 000 to 190 000 people with deep analytical skills as well as 1.5 million managers and analysts with the know-how

to use the analysis of big data to make effective decisions. Leaders in every sector will have to grapple with the implications of big data, not just a few data-oriented managers. The increasing volume and detail of information captured by enterprises, the rise of multimedia, social media, and the 'Internet of Things' will fuel exponential growth in the data for the foreseeable future.” The demand for Data Science expertise is also evident in South Africa. A Google search currently yields 162 Data Science job opportunities locally.`

New honours programme focussing on Data Science The Statistics and Actuarial Science Department will introduce Data Science as a new focus to the Mathematical Statistics honours programme from February 2016. This will be done in collaboration with the Computer Science Division of the Department of Mathematical Sciences. The new Data Science Honours programme is intended to equip students with the necessary programming and analytical skills to use large-scale data to solve challenging problems. It will have a dual purpose: to teach students the knowledge and skills to be able to use complex models to extract valuable insights from big data, and to equip students to efficiently deal with the technological, computational and design challenges associated with big data management and computations. Due to the importance of the second purpose, Computer Science up to at least second year level will be an admission requirement. Students will also have to include specific modules from the Computer Science honours programme as part of their degree.

Industry involvement

Programme structure

Meetings with industry representatives have conveyed a pressing need for post-graduate students with the knowledge and skills that the Data Science programme is aiming at. Industry involvement will be an important aspect of the programme, since hands-on experience in industry is essential to prepare students for practical demands. At this stage, industry involvement includes student bursaries, student internships, Data Scie nce mark e ting initiative s, and involvement in student projects.

Candidates will have to enrol for a BCom Honours or a BSc Honours in Mathematical Statistics (focus area Data Science). The formal requirement for admission to the programme will be Mathematical Statistics as one of the third-year subjects, and Computer Science up to at least second-year level. During the honours year, students will study a selection of modules presented in the Department of Statistics, as well as selected modules in Computer Science.

11

Modules from which students can select in the D e partme nt of Statistics include D ata Mining, Stochastic Simulation, Statistical Learning Theory, Bayesian Statistics and Time Series Analysis. On the Computer Science side, students will be able to select from modules such as Machine Learning, Advanced Algorithms, Databases and Computer Vision. Students who wish to be considered for admission to the programme should apply through the usual channels

for admission to Honours in Mathematical Statistics. As for all honours programmes, the Data Science programme will include a compulsory research project, which will contribute 25% of the total credits. Project topics will typically be identified through company liaisons, which will offer students the opportunity to gain both soft skills and experience in solving realworld problems.

Post-graduate students meet with industry During the first semester, a group of postgraduate students from the Department of Statistics and Actuarial Science met with industry a number of times. Some of these

meetings culminated in internship offerings. We thank the students involved for their interest in the new Data Science programme, and post some of their comments below.

“The new Data Science focus at SU comprises complementary courses designed to adequately prepare students for the complex data-driven problems they are likely to encounter in the real world. Combining mathematical and statistical knowledge with competence in computer programming, Data Science graduates will be able to make an impact on a world increasingly run by data.” Arnu Pretorius

“New data are being generated at extraordinary speeds in our world, and the effects thereof are seen in every aspect of our lives. Data science is an exciting, highly interdisciplinary field and is relevant now more than ever with this explosion of information. The new focus on Data Science at our university equips students with insight into our expanding data universe, and provides tools for creating value and enhancing understanding.” Chané Orsmond

“Meeting with industry this year was an invaluable experience: hearing about the opportunities available, as well as obtaining a general idea of the skills that are most sought after in the Data Science world. Data Science is a very useful skill to have, and there are many opportunities available. I strongly recommend any student that is interested in pursuing a post-graduate degree in Statistics to consider the Data Science focus.” Dr Margaret de Villiers

“I have attended three meetings with industry. It was a great opportunity to get in touch with the world outside the academic environment and to get to know what the industry requires. I believe that these opportunities are invaluable for students in order to make an informed decision about their future career path.” Monika du Toit

12

SAS Data Science and Advanced Analytics Summit Dr Surette Bierman of the department attended the SAS Data Science and Advanced Analytics Summit held at SAS International Headquarters in Cary, NorthCarolina in April 2015. The objective of the two-day summit was to bring together academia and industry to discuss best practices in offering Data Science programmes. A tte nde d by 76 de le gate s, it comprise s 14 presentations, with an even spread between

presentations by industry and Data Science programme representatives. During a break-away workshop topics such as industry projects and sponsorships, student applications, and programme content were discussed in smaller groups. The summit concluded with a panel discussion on the critical success factors for Data Science programmes. Overall, attendance of the summit was a very worthwhile experience.

Post-graduate students with two of their lecturers, Dr Bierman and Prof Steel, on one of their excursions to industry: (f.l.t.r.) Dr Surette Bierman, Arnu Pretorius, Jacques Smidt, Chané Orsmond, Adriaan Mouton, Sven Buitendag, Dr Margaret de Villiers and Prof Sarel Steel.

Actuarial Science research contributes to financial industry Actuarial Science staff members are interested in a wide range of research topics, often addressing problems faced by the financial industry. Over the last year, actuarial lecturers, together with some of their students, have done valuable research and presented results at industry conferences. Davy Corubolo, Rob Clover and Adam Plantinga presented a research paper on the modelling of catastrophes at the 2014 Actuarial Society Convention. Insights from this paper could assist actuaries working in industry to assess the impact on insurance companies of this complicated risk. The paper was very well received and awarded the Actuarial Society prizes for the Best paper on risk, as well as the Best paper by first time authors. At this conference Davy also co-authored a paper on the impact on the South African insurance industry of improved mortality experience as a result of the roll out of anti-retroviral treatment. Insights from this paper were widely quoted in the South African media. Natalie van Zyl was a co-author on a paper that was awarded a prize at the 2014 International Congress of Actuaries. The paper, The Evolution and Future of Social Security in Africa: An Actuarial Perspective, analysed the state of social security in Africa and identified lessons that could be applied in the development of social security schemes. Natalie also produced a paper, together with her husband, Daniel, on potential applications of the field of

behavioural economics to the retirement industry. The paper was presented at the 2014 Actuarial Society Convention and makes a contribution to the industry discussion on how to improve provision for retirement. Work done by Stephen Burgess and Richard van der Berg provided delegates attending the annual Actuarial Society Life Assurance Seminar with valuable results on the lapse experience in the life insurance industry.

Actuarial Science lecturers Mr Rob Clover, Ms Natalie Van Zyl and Mr Davy Corubolo.

13

Students participate in global research challenge The Chartered Financial Analyst (CFA) Institute's Research Challenge is a global challenge that offers students an opportunity to put their theoretical knowledge regarding financial analysis into practice. Approximately 800 universities from 55 countries participate in the challenge.

away with the top spot yet, the challenge offers an exceptional learning opportunity and an excellent chance to experience what life in the industry entails.

University teams research and analyse a publicly traded company and write a research report to provide a buy, sell or hold recommendation. Each team is supported by both an academic and an industry mentor. The reports are graded by industry professionals within each country, and the top teams are invited to take part in the specific country's finals, which entails presenting and defending the team's research, analysis and recommendation to academic and industry experts. The winning team of each country is invited to participate in the regional finals abroad, and the winner goes on to compete in the global finals. For the past four years, Stellenbosch University's postgraduate students in Financial Risk Management have participated in the Research Challenge within the Europe, Middle East and Africa region. The Maties Research Team has been invited three times to present at the finals of the South African leg of the competition. Although the team has not walked

The Maties Research Team was supported by academic mentors Mr Davy Corubolo (on the left), Senior Lecturer in Actuarial Science, and Dr JD van He e rde n, Se nior Le cture r in Financial Risk Management. Rootstock Investments, a focused investment management firm of Stellenbosch, is the team's industry mentor.

Excellence in retirement fund industry awarded Dr JD van Heerden, senior lecturer in Financial Risk Management, was one of the judges in the Imbasa Yegolide Awards hosted by the Council of Retirement Funds for South Africa, Batseta. The awards serve to recognise service providers who have given excellent service in the retirement fund industry, while highlighting the role of principal officers and trustees of retirement funds in South Africa. This year, service providers were considered for awards across 25 different categories. Dr Van Heerden acted as one of four judges in the following six categories

which were won by: Equities Manager of the Year: Coronation Fund Managers, Responsible Investment Manager of the Year: Futuregrowth Asset Management, Overall Investment/Asset Manager of the Year: Investec Asset Management, Absolute Return Manager of the Year: Argon Asset Management, Transition Manager of the Year: BNP Paribas Cadiz Securities,

Dr JD van Heerden Hedge Fund Provider of the Year: Old Mutual Investment Group.

14

Personeel woon internasionale konferensies by Na 'n veeleisende eerste semester was dit oudergewoonte die jaarlikse uittog van baie akademici in SuidAfrika na buitelandse bestemmings. Junie en Julie is gewoonlik die maande waar akademici vak ge spe sialise e rde inte rnasionale k onfe re nsie s bywoon. Dis hier waar internasionale bande versterk

Prof Niel le Roux het die konferensie van die International Federation of Classification Societies (IFCS) gedurende Julie in Bologna, Italië bygewoon. Hy het die lesing met die titel: Biplot-based visualizations in latent class modelling met mede-outeurs Zsuzsa Bakk en Jeroen Vermunt van die Universiteit van Tilburg, Nederland gelewer. Prof Le Roux woon ook die Correspondence Analysis and Related Methods (CARME) konferensie gedurende September in Napels, Italië by. Hier sal hy die lesing met die titel Fisher's Optimal Scores Revisited met medeouteurs John Gower van die Open University, Verenigde Koningkryk en Sugnet Lubbe van die Universiteit Kaapstad lewer. Gedurende November bied Prof Le Roux die volgende Magisters-kursus by die Katolieke Universiteit van Leuven in België aan: Uni-dimensional and Multidimensional Scaling. Prof Nelmarie Louw het die International Federation of Classification Societies (IFCS) konferensie gedurende Julie in Bologna, Italië bygewoon. Sy het die lesing met die titel Small sample multi-label discriminant analysis gelewer. Prof Paul Mostert het die 16th Applied Stochastic Models and Data Analysis (ASMDA) konferensie gedurende Julie in Athene, Griekeland, bygewoon. Hier het hy die lesing met die titel Objective Bayesian analysis for the generalized Rayleigh model under asymmetric loss gelewer. Dr JD van Heerden het die First European Academic Research Conference on Global Business, Economics,

word en ook vriende gemaak word. Die Departement Statistiek en Aktuariële Wetenskap is geen uitsondering nie. Verskeie dosente van die departement het internasionale konferensies bygewoon en lesings aangebied:

Finance and Social Sciences konferensie gedurende Junie in Milan, Italië bygewoon. Hy het die lesing met die titel Liquidity and the cross section of equity returns: The case for South Africa, met mede-outeur Paul van Rensburg van die Universiteit Kaapstad gelewer. Dr Van Heerden woon ook die 27th Business & Economics Society International (B&ESI) Conference gedurende Julie in Albufeira in Portugal by. Hier lewer hy die lesing met die titel The impact of liquidity on equity returns: The case for South Africa, met mede-outeur Astrid Reisinger. Prof Martin Kidd van die SSK het die International Federation of Classification Societies (IFCS) konferensie gedurende Julie in Bologna, Italië bygewoon. Hy het die lesing met die titel Feature selection and kernel specification for support vector machines using multiobjective genetic algorithms met mede-outeurs Martin Philip Kidd en dr Surette Bierman van Universiteit Stellenbosch gelewer. Prof Sarel Steel en sy post-doktorale genoot, dr Trudie Sandrock, woon die European Conference on Data Analysis (ECDA2015) in Colchester, Verenigde Koninkryk gedurende September by. Dr Sandrock sal 'n lesing getiteld Variable Selection in Multi-Label Classification using Probe Variables lewer. Mr Davy Corubolo will attend the 2015 South African Actuarial Society conference in Johannesburg during October. The title of the talk is The Sources of South African Equity Fund Performance, with co-presenter Ms Nicole Lester of Stellenbosch University.

Besoeke ontvang

Personeel en oud-studente van die departement wat die International Federation of Classification Societies (IFCS) konferensie gedurende Julie in Bologna, Italië bygewoon het is (v.l.n.r.) Pieter Schonees (oudstudent van die departement en besig met PhD -studies in Rotterdam), Johanné Nienkemper-Swanepoel (dosent in Biometrie aan US), prof Sugnet Lubbe van UK (oudstudent), asook proff Nelmarie Louw, Niel Le Roux en Martin Kidd.

Internasionale besoekers aan die departement sluit prof Jef Teugels, vergesel van sy vrou Rita, van die K atholieke Universiteit Leuven in België in. P r o f Te u g e l s , w a t d i e de p ar te me nt v anaf midde l Oktober tot middel Desember Prof Tertius de Wet besig om sal besoek, doen navorsing data van ‘n klok in te samel. saam met prof Tertius de Wet en dr Pieta van Deventer in die veld van Kampanometrie (Campanometry). Die projek handel oor die meet en statistiese analise van die fisiese en akoestiese data van klokke in die Wes-Kaap, asook die inventarisering daarvan op SUNDigital.

15

Department 'ecstatic' about first prize in SASA competition Two students in the Department of Statistics and Actuarial Science recently became the first students in the history of the Department to win the annual honours project competition of the South African Statistical Association (SASA). This competition is sponsored by SAS® and administered by SASA's Education Committee. Dr Margaret de Villiers and Ms Dalene Saaiman, under supervision of Prof Paul Mostert who presents the postgraduate module in survival analysis, outdid 17 other entries to win the prize with their project titled Influence of RightCensoring on Some Kernel-Smoothed Hazard Rates. De Villiers and Saaiman chose this topic because it aligns well with one of their supervisor's research interests. Survival analysis is a branch of statistics that deals with analysis of time duration until one or more events happen, for example the re-occurrence of cancer after treatment. An important parameter in survival analysis is the hazard rate, which is the risk of a subject experiencing the event at any given moment (read more about the topic below). Entries for the competition were received from the Universities of Cape Town, Free State, KwaZulu-Natal, Pretoria and Stellenbosch as well the Nelson Mandela Metropolitan University and Wits. Each University may enter its top three projects for this competition. In their letter of congratulations, the organisers commented on the excellent quality of the 18 projects they had received: "There was a good variety of topics as well as statistical techniques utilised. Ranking the top projects was a difficult task, in particular due to the high standard, but also due to the number of projects that had to be adjudicated." De Villiers and Saaiman said they were "blown away and

Dr Margaret de Villiers, Prof Paul Mostert and Ms Dalene Saaiman. extremely happy" about the award, while Prof Mostert said the department is "ecstatic" about the first prize. "It is really an honour for the department to be awarded first prize. The honours project competition has always been stiff and the winning department gets some well-deserved exposure within the statistics community of South Africa." The prize will officially be awarded to the students during the opening ceremony of the annual SASA conference in December 2015 in Pretoria. Part of the prize money is to present a talk at the conference, and winners' expenses will be paid by the sponsor of the competition, SAS®.

Dr Margaret de Villiers

Ms Dalene Saaiman

While completing an Honours, Masters and PhD in Botany at Stellenbosch University, Margaret de Villiers realised the fundamental role that Statistics plays in all applied scientific disciplines. She therefore took some Statistics courses to improve her understanding of the analyses that one carries out in biological research. She is now studying for her Masters in Mathematical Statistics, working on developing models for predicting crop yields accurately.

Having obtained a BSc Honours in Physiotherapy at the University of the Free State in 1998, Dalene Saaiman worked as a physiotherapist. In 2006 she started studying Statistics part-time at Kovsies, where she completed her first and second year. She then changed universities and did both her third year and an Honours degree at Stellenbosch University. She is currently working as a biostatistician and plan to complete her Masters in Statistics part-time.

More about the topic Survival analysis is a branch of statistics that deals with analysis of time duration until one or more events happen, e.g. the reoccurrence of cancer after treatment. An important parameter in survival analysis is the hazard rate, which is the risk of a subject experiencing the event at any given moment, on condition the subject has survived up to that point in time. Most survival studies and clinical trials occur over a fixed period of time due to financial constraints and other practical and ethical considerations. For example, a study testing the efficacy of a new medication in preventing the recurrence of a certain type of cancer would probably span several years. In the case of the patients involved in the study that redevelop cancer, the length of their remission can be recorded exactly. However, some of the

patients will either leave the study prematurely or still be in remission at the end of the study, and an exact time to the event (re-occurrence of cancer) will not be recorded for them. The recorded time-to-event for these latter patients is referred to as being right-censored. The presence of right-censored observations in a data set is problematic, since their omission from the analyses would result in an underestimate of parameters such as the average length of time to the event and hazard rate. This project involved investigating the influence of the presence of right-censored observations on estimates of the hazard rate. An extensive simulation study was carried out to investigate the influence of censoring using different kernel functions for the hazard rate. The study revealed amongst other results that increasing the proportion of right-censored observations in a data set makes the hazard rate estimates less stable but still gives the correct estimate on average.

16

Upcoming Seminars 2 October

Marena Manley (SU Department of Food Science) Non-destructive characterisation of cereal grains with near infrared (NIR) hyperspectral imaging and principal component analysis (PCA)

16 October

Stephan van der Westhuizen (Compuscan, Stellenbosch) Statistics at the Credit Bureau - a data analyst's viewpoint

17 November

Gregory Levitin (Israel Electric Corporation, Haifa, Israel) Merging game theory and risk analysis in optimal defense of complex stochastic systems

Venue:

Room 2048, Statistics and Accounting Building, c/o Victoria and Bosman Street, Stellenbosch

Time:

13: 00

Enquires:

Danie Uys, tel: 021 808 3879 e-mail: [email protected]

Garret Slattery helps win international blind golf tournament In June this year, the International Blind Golf Association (IBGA) staged its second international team match event along the lines of the famous Ryder Cup, with a North America team pitched against a world team. The event was held at the magnificent Villa d'Este Golf Club near Lake Como in northern Italy. Teams consisted of 12 players each, and were selected from the

top-ranked players in each of the three sight categories (B1, B2 and B3). The North America Team was made up of players from the USA and Canada, while players from elsewhere formed the World Team. The only South African in this team was Garrett Slattery, a member of the Stellenbosch Golf Club, and professor in Actuarial Science at SU.

The North America team (left) and the World Team (right) on the putting green. Garrett is in the back row with the green shirt and the big white and green hat.

17

Villa d'Este is considered one of the most challenging and difficult courses in Europe, and has hosted several significant amateur competitions. The clubhouse looks out across the course to Lake Montorfano.

The Vision Cup The competition was played over three days. On the first two days players teamed up in pairs and played using the Pinehurst system. According to this system, both players in the pair tee off and then swap balls to play the second shot, after which the better ball is selected and played out on an alternate shot basis. The third day had all 12 players playing singles. The World Team took a 1 point lead into the second day. They stretched their lead to 3 points after the second day, with the 12 singles matches still to follow. Day 3 arrived with both teams still believing the trophy was up for grabs. A good start from some of the North American players in the early matches saw them start to close the 3 point gap. In the second last match off, Garrett faced strong opposition in the form of C anada's De re k Kibblewhite. After losing the first hole to Garrett, Derek reeled off a string of 7 consecutive pars to be 3 up after 8 (despite Garrett being well under his handicap at that stage). Garrett rallied with 4 pars in the next 6 holes, and with Derek going off the boil a bit, the game swung in Garrett's favour. A

close match ended on the 18th with a 1 up victory to Garrett. By that stage the World Team had also amassed enough points to success - fully defend the trophy they had won two years earlier. The final score of 14¼ to 9¼ didn't reflect how tight the competition actually was.

Prof Garrett Slattery

At golf events for the visually impaired, the role of the sighted guide is crucial. Garrett was fortunate enough to have been guided by Stellenbosch teaching professional, Erich Kliem. At each morning's session on the driving range, Erich saw see how Garrett was swinging and gave him simple swing thoughts for the day. Erich's ability to accurately align Garrett, together with his calm nature and ability to assist, was a strategy that greatly contributed to Garrett's success. The event was played in a wonderful spirit, and sets the stage for a return match in 2017 in the USA.

18

The modern day statistics toolbox by Paul Mostert The following article is a concise version of the presidential address by Prof Paul Mostert at the 2014 conference of the South African Statistical Association at Rhodes University. doing private tuition in London during this time. He wrote a number of papers while being alive, but the most famous paper was published after his death. The 250th anniversary of that paper Essay Towards Solving a Problem in the Doctrine of Chances was celebrated in 2014. It was published in the Philosophical Transactions of the Royal Society of London in 1764. The way people describe the paper at the time is that it is a technique that based the probability of an event to happen in a given circumstance, on a prior e stimate of its probability unde r the se circumstances. This paper was sent to the Royal Society by his friend, Richard Price. Price had found it among Bayes' papers after he died.

Prof Paul Mostert The past couple of years marked a few celebrations in the world of statistics. In 2013 was the year of statistics and was celebrated all over the world and in SA. Locally, in 2014 it marks also the centenary year of the first Statistics Act of South Africa. The indirect focus of the talk was about a person who had a huge impact on statistics today and it all starte d a me re 250 y e ars ago. Reverend Thomas Bayes was an Englishman and studied initially theology. His work contributed to the field of probability and statistics. H is ideas have created much controversy and debate among statisticians over the years. There has been speculation that he was taught by de Moivre, who was

Basically, suppose X is binomial with parameters (n,π). He conceded that an objective belief would be that each value of X will occur equally often. The only prior distribution on π consistent with this is the uniform distribution and along the way, he codified Bayes theorem. This is also regarded by some as the start of objective Bayesian theory. However, many regarded that the real inventor of objective Bayes was Simon Laplace. He virtually always utilised a constant prior density and clearly said why he did so. He established a version of the central limit theorem, showing that, for large amounts of data, the posterior distribution is asymptotically normal and the prior does not matter. He solved many applications, especially in physical sciences. He had numerous methodological developments, for example a version of the Fisher exact test. It was called probability theory until 1838. From 1838-1950, it was called inverse probability, so named by de Morgan. From 1950 onwards it was called Bayesian analysis. Boole, Venn and many others had been calling the use of a constant prior logically unsound since the answer depended on the choice of the parameter, so alternatives were the order of the day, like Fisher's developments of likelihood methods, Neyman's development of the

19

frequentist philosophy, which obviously appeal to many.

Bayesians , with varying views on the role of frequentist ideas; and

Then Harold Jeffreys revived the objective Bayesian viewpoint through his work, especially through his transcripts in the Theory of Probability of 1937, 1949 and 1963. The now famous Jeffreys prior yielded the same answer no matter what parameterisation was used. His priors yielded the 'accepted' procedures in all of the standard statistical situations. He began to subje ct fre que ntist philosophie s to critical examination, including his famous critique of pvalues: “A hypothesis, that may be true, may be rejected because it has not predicted observable results that have not occurred.”

pragmatists, who do not have strong view and pick and choose what seems to work for the problem at hand.

Do we still need to debate of being Bayesian or being frequentist? We have a lack of an agreed inferential basis for statistics and this makes life interesting for academics, but at the price of negative implications for the status of statistics in industry, science, and government. Brad Efron once labelled the 19th Century as generally Bayesian, the 20th Century as generally frequentist, and suggested that statistics in the 21st Century will require a combination of Bayesian and frequentist ideas. There has always been this great divide between Bayesians and frequentists. The literature is full of these ex amples and counter ex amples and arguments for and against both frequentist and Bayesian approaches. Roderick Little in 2011 has started this debate and as he stated there are no clear winners – there is no agreed inferential philosophy for how to do statistics at the end of the day. Different inferential views are no longer compared and argued. Bayesians still debating 'objective' versus 'subjective' approaches to Bayesian inference, but that seems like an argument between siblings in the Bayesian family and mostly ignored by frequentists. Whether or not the inferential debate has receded, it is no longer an academic issue. In the eighties, applications using the Bayesian approach were limited to small problems by the inability to compute the high dimensional integrations involved in multi-parameter models. Increased computational power and the development of Monte Carlo approaches in computing posterior distributions and marginals have turned this weakness of Bayesian statistics into a strength. Today, statisticians can roughly and very crudely be divided into three main camps: frequentists, who reject the Bayesian approach, or never learned much about it;

Let's over-simplistically define what is meant by a frequentist: one who bases inference for an unknown parameter θ on hypothesis tests or confidence intervals, derived from the distribution of statistics in repeated sampling and a Bayesian: one who bases inferences about θ on its posterior distribution, under some model for the data and prior distribution for unknown parameters. Included in the latter is 'subjective' Bayes, where proper priors a r e e l i ci te d, a n d ' o b j e cti v e ' B a y e s , w h e r e conventional 'reference priors' are used. As a very general comment, we can see asymptotic maximum likelihood inference as a form of largesample Bayes, with the interval for θ being interpreted as a posterior credibility interval rather than a confidence interval. This broad view of Bayes provides a large class of practical frequentist methods with a Bayesian interpretation. Pragmatists might well argue that good statisticians can get sensible answers under Bayes or frequentist paradigms; indeed maybe two philosophies are better than one, since they provide more tools for the statistician's toolbox. Many examples exist where the Bayesian and frequentist approaches lead to fundamentally different results. Two basic examples are given as they can easily be handled on a basic level of statistics. Example 1: Chi-square tests of independence in a 2 × 2 contingency table. Treatment

Positive

Negative

I

87

13

II

95

5

Consider a one-sided test H0 : πI = πII vs Ha : πI > πII for independent samples assigned two treatments, where πj is the positive rate for treatment j, j = I, II. Standard Pearson chi-squared test (P) Pearson test with Yates's continuity correction (Y) Fisher exact test (F) Bayesian solution, which computes the posterior probability that πI < πII, based on some choice of prior distribution for the positive rates (Jeffreys' reference prior):

20

p(πI, πII) µ πI

−1/2

πII

−1/2

(B),

while emphasising that other prior distributions yield different answers. One-sided p-value of 0.0241 (P), 0.0415 (F), 0.0419 (Y), and a posterior probability that P(πI < πII|data)= 0.0202 (B). The p-values for F and Y tend to be similar for the one-sided problem, and are known to be conservative when one margin of the 2×2 table is fixed; P is better calibrated when one margin is fixed, but is approximate. F is exact if both margins are fixed. So for the frequentist, the choice of P versus F or Y comes down to whether or not we condition on the second margin. So there is no clear frequentist answer for this most basic of problem. The Bayesian answer avoids this vagueness about conditioning on the second margin; indeed conditioning is never really an issue with the Bayesian approach, because posterior distributions condition on all the data. On the other hand, there is nothing unique about the Bayesian answer either, since the posterior probability depends on the choice of prior, and the theory of reference priors has its own problems. Example 2: Single population inference with a bounded variance. Consider an independent normally distributed sample with n = 9 observations, sample mean = 2 and standard deviation s = 1.25. If the population standard deviation (say σ) is unknown, the usual 95% interval for the population mean is: (1) and a Bayesian interprets it as a 95% posterior credibility or probability interval, based on Jeffreys' reference prior (RP) distribution p(μ, σ) ∝ 1/σ . Both approaches yield the same interval. Suppose now that we know that σ = 2.0, as in the case when σ is known like in a measuring instrument. The standard 95% interval (when σ = 2.0): (2) and is again the correct inference, meaning the wider interval reflecting the fact that the sample variance s is underestimating the true variance σ. Now, suppose the expert reports that σ > 2.0, because of additional unaccounted sources of variability to be known. Three 95% intervals for μ are now to be considered: (1), (2), or the Bayesian credibility interval with the reference prior modified to incorporate the constraint that σ > 2.0, namely: (3)

Interval (3) can be obtained by usual MCMC methods with open source software like WinBUGS®. Now interval (1) seems to be the optimal 95% frequentist confidence interval, given that it is has exact nominal coverage and σ is unknown, but it is counterintuitive for inference: you will have difficulty explaining how the information that σ > 2.0 leads to a narrower interval than interval (2), the standard frequentist interval when σ = 2.0. Interval (1) cannot be accepted as it ignores the lower bound on σ and interval (2) is the obvious asymptotic approximation, given that 2.0 is the maximum likelihood estimate of σ. However, for a sample size n = 9, the asymptotic assumption is clearly incorrect. These and many more examples exist to illustrate conflicting results in simple settings. Examples like these may undermine the credibility of statisticians in much more complex real-world settings. The ideal is to unify these approaches. Unification of these approaches was pushed as early as 1980 by George Box and in 1984 by Donald Rubin. Many people in the past have discussed the strengths and weaknesses of both frequentist and Bayesian inferences to identify a statistical process that can capture the strengths of these approaches. Without going into all the detail and motivations, I briefly list them. Firstly, the frequentist paradigm avoids the need for a prior distribution. There is a clear separation of the role of prior information in model formulation and the role of data in estimating parameters. However, these components are treated on a more equal footing in the Bayesian approach. The frequentist approach is flexible, in the sense that full modelling is not necessarily required, and inferences lack the formal structure of Bayes' theorem under a fully specified prior and likelihood. In a sense any method is frequentist, provided its frequentist properties can be studied. Frequentist theory is not prescriptive - in the sense that it assesses the properties of inference and is not so much about the inference system as such. Frequentist theory is incomplete - here we can think of the interpretation with confidence intervals. Frequentist theory is incoherent, in the sense that it may violate the likelihood principle. The likelihood principle plays an important role in the inferential debate since it is satisfied by Bayesian inference and violated by frequentist inference. A classic example is a Bernoulli experiment, where π is the probability of success and 1− π the probability of a failure. Consider two experiments: (I) binomial sampling, repeated n = 17 times and X = 4 of successes (random number of successes); and (II) negative binomial sampling, where the experiment is repeated until a predetermined number of (x = 4)

21

successes and N = 17 repetitions were needed (random sample size). Both lead to the same likelihood, namely L(π) µ π 4(1 − π)13. Hence, under the likelihood principle, these two experiment/data combinations should yield the same inferences. The maximum likelihood estimates are the same (4/17), but since the sampling spaces are different, the pvalues from the usual tests of H0 : π = 1/2 vs Ha : π ≠ 1/2 are different— 0.049 for (I) and 0.021 for (II). The Bayesian paradigm addresses many of the weaknesses of the frequentist approach. For a given Bayesian model and prior distribution, Bayes' theorem is the simple prescription that supplies the inference. It may be difficult to compute, and checks are needed to ensure that the posterior distribution is proper, but the solution is clear and avoids vagueness. There are, however, difficulties in implementing the Bayes approach. Perhaps the most common is that Bayes is viewed as too subjective for scientific inference, requiring a subjective definition of probability and the selection of a prior distribution – but maybe a matter of degree. Bayesian methods include methods based on noninformative priors that some classify as frequentist. There are, however, some difficulties with the practical implementation of the Bayesian approach that I find more compelling. Bayes requires and relies on full specification of a model (likelihood and prior). Bayes yields 'too many answers'. The frequentist paradigm does not provide enough exact answers; with Bayes, there is an embarrassment of choices, because once the likelihood is nailed down, every prior distribution leads to a different answer. Bayesian hypothesis testing has the logic of Bayes' theorem in its favour, but comparing models of different dimension is tricky, and sensitive to the choice of priors. A crude summary of these is: Bayesian statistics is strong for inference under an assumed model, but is relatively weak for the development and assessment of models. Frequentist statistics provides a useful tool for model development and assessment, but is a weak tool for inference under an assumed model. So, a natural compromise is to use frequentist methods for model development and assessment, and Bayesian methods for inference under a model. This capitalises on the strengths of both approaches. What are the implications of this hybrid Bayesian approach for the two examples? Example 1 (continued). The conjugate Bayesian inference adds Beta priors for the success rates in the two groups and computes the posterior probability that πI > πII. Proper prior distributions may be entertained in certain contexts; when there is little prior evidence about the success rates, the

choice of 'objective prior' has been debated, and Jeffreys' prior is one plausible conventional choice. This hybrid Bayesian method limits the vagueness in the reference set for frequentist assessments to model evaluation, rather than to model inference under a specified model. Example 2 (continued): As a simple example of posterior predictive checking, is to determine the posterior predictive distribution of the sample 2 variance s* . The posterior probability of observing a sample variance in future datasets of size 9 as low as 2 the observed sample variance of s = 1.5625 is less than 0.05, which is low but not exceptional, so the Bayesian model seems not unreasonable. If hybrid Bayesian methods to be implemented appropriate ly for the future , what are its implications for teaching and in practice of statistics? More emphasis should be on statistical modelling than on statistical methods. Formulating useful statistical models for real problems is not simple, and students need more instruction on how to fit models to complicated datasets. More attention is needed to assessments of model fit. This is where frequentist methods have an important role. Bayesian statistical methods need to be taught. Currently, Bayesian statistics is absent or optional in many honours and masters programmes and in general are our post-graduate students taught with very little exposure to the Bayesian ideas, beyond a few lectures in a theory that is dominated by frequentist ideas. A full Bayes course should be a required component of any honours or masters programme in statistics. If we think of the consumers of statistics (students from our service courses), Bayesian statistics is not a part of the content, so most of them think of frequentist statistics as the only approach, and are not even aware that Bayesian inference exists. We cannot claim that Bayesian inference is too difficult to teach to these students with limited mathematical ability. The basic idea of Bayes' theorem does not require calculus, and Bayesian methods can be taught if the emphasis is placed on interpretation of models and results, rather than on the inner workings. Bayesian posterior credibility intervals have a much more direct interpretation than confidence intervals – our consumers of statistics in any case interpret it this way to their managers. It is still shocking to see some lecturers even interpret it this way to their students! Any case, frequentist hypothesis testing is not easy to teach to these consumers of statistics.

To summarise, Bayes and frequentist ideas are important for good statistical inference, and both sets of ideas need to be developed and taught.

22

Papers co-authored by staff 13.

HARVEY J, VAN DER MERWE AJ. Modelling Occupational Exposure Using a Random Effects Model: A Bayesian Approach. South African Statistical Journal 2014; 48: 61-71.

14.

HATTINGH M, UYS DW. In-Season Retail Sales Forecasting Using Survival Models. ORiON: Journal of the Operations Research Society of South Africa 2014; 30(2): 59-71.

15.

JOOSTE MM, ROHWER EA, KIDD M , HUYSAMER M. Comparison of antioxidant levels and cell membrane composition during fruit development in two plum cultivars (Prunus salicina Lindl.) differing in chilling resistance. Scientia Horticulturae 2014; 180: 176-189.

16.

KANSKY R, KIDD M, KNIGHT AT. Meta-Analysis of Attitudes toward Damage-Causing Mammalian Wildlife. Conservation Biology 2014; 4: 924-938.

17.

KHAN Z, CLOETE KJ, HARVEY J, WEICH LEM. Outcomes of adult heroin users v. abstinent users four years after presenting for heroin detoxification treatment. South African Journal of Psychiatry 2014; 20(3): 82-87.

18.

KOTZE MJ, MARNEWICK J, KIDD M, FISHER LR, VAN VELDEN DP. Assessment of the impact of hereditary factors on biochemical parameters of cardiovascular risk in relation to moderate alcohol consumption. Nutrition and Aging 2014; 2: 189-195.

19.

KRUGER M, REYNDERS D, OMAR F, SCHOEMAN J, WEDI O, HARVEY J. Retinoblastoma Outcome at a Single Institution in South Africa. SAMJ South African Medical Journal 2014; 104(12): 859-863.

20.

DANNATT L, CLOETE KJ, KIDD M, WEICH LEM. Frequency and correlates of comorbid psychiatric illness in patients with heroin use disorder admitted to Stikland Opioid Detoxification Unit, South Africa. South African Journal of Psychiatry 2014; 20(3): 77-82.

LE ROUX NJ, LUBBE S, GOWER J. The analysis of distance of grouped data with categorical variables: Categorical conanical variate analysis. Journal of Multivariate Analysis 2014; 132: 9-24.

21.

DE WAARD L, BUTT JL, MULLER CJB , CLUVER CA. Retrospective review of the medical management of ectopic pregnancies with methotrexate at a South African tertiary hospital. South African Journal of Obstetrics and Gynaecology 2014; 20(3): 84-87.

LUCKHOFF HK, JANSE VAN RENSBURG S, BOTHA K, KIDD M, KOT ZE MJ. T he pro-inflammatory T NFA -308G>A (rs1800629) Polymorphism is associated with an earlier age at onset in patients with major depressive disorder. African Journal of Psychiatry 2014; 17(3): 1000124.

22.

DE WET T, TEUGELS JL, VAN DEVENTER PJU. Historic bells in Moravian Missions in South Africa's Western Cape. Historia 2014; 59(2): 94-119.

MARTIN LI, VILJOEN M, KIDD M, SEEDAT S. Are childhood trauma exposures predictive of anxiety sensitivity in school attending youth? Journal of Affective Disorders 2014; 168: 5-12.

23.

MIHRET A, LOXTON AG, BEKELE Y, KAUFMANN SHE, KIDD M, HAKS MC , OTTENHOFF THM, ASEFFA A, HOWE R, WALZL G. Combination of Gene Expression Patterns in Whole Blood Discriminate Between Tuberculosis Infection State. BMC Infectious Diseases 2014; 14: 257.

24.

MYERS MM, HOFMEYR F, GROENEWALD CA, NEL DG, FIFER WP, SIGNORE C, HANKINS GDV, ODENDAAL HJ. Fetal heart rate patterns at 20 to 24 weeks gestation as recorded by fetal electrocardiography. Journal of Maternal-Fetal & Neonatal Medicine 2014; 27(7): 714-718.

25.

NEL WS, BRUWER BW, LE ROUX NJ. An emerging market perspective on key value drivers in the valuation of crossborder transactions into South Africa. Economics, Management and Financial Markets 2014; 9(4): 92-111.

26.

NEL WS, BRUWER BW, LE ROUX NJ. An emerging market perspective on peer group selecton based on valuation fundamentals. Applied Financial Economics 2014; 24(9): 621-637.

The following papers were co-authored by staff of the Department of Statistics and Actuarial Science and the Centre for Statistical Consultation, and published in 2014 in accredited journals. (Their names appear in bold.) 1.

2.

3.

4.

5.

6.

7.

8.

9.

ACKERMANN C, ANDRONIKOU S, LAUGHTON B, KIDD M, DOBBELS E, INNES S, VAN TOORN R, COTTON M. White matter signal abnormalities in children with suspected HIVre late d ne urologic dise ase on e arly combination antiretroviral therapy. Pediatric Infectious Disease Journal 2014; 33(8): e207-e212. ARCHER E, BEZUIDENHOUT J, KIDD M, VAN HEERDEN BB. Making use of an existing questionnaire to measure patientcentred attitudes in undergraduate medical students: A case study. African Journal of Health Professions Education 2014; 6(2): 150-154. CHILIZA B, ASMAL L, OOSTHUIZEN PP, VAN NIEKERK Y, ERASMUS RT, KIDD M, MALHOTRA AK, EMSLEY RA. Changes in body mass and metabolic profiles over 12 months in patients with first-episode schizophrenia with assured antipsychotic adherence. Schizophrenia Research 2014; 153(Suppl 1): S305. COETZER RL, ROUSSOUW RF, LE ROUX NJ. Reference set selection with generalized orthogonal Procrustes analysis for multivariate statistical process monitoring of multiple production processes. Chemometrics and Intelligent Laboratory Systems 2014; 132: 52-62. CONRADIE M, CONRADIE MM, KIDD M, HOUGH FS. Bone density in black and white South African women: contribution of ethnicity, body weight and lifestyle. Archives of Osteoporosis 2014; 9(193): 1-12.

DE WET T. EM-based Likelihood Inference for Some Lifetime Distributions Based on Left Truncated and Right Censored Data and Associated Model Discrimination. South African Statistical Journal 2014; 48(2): 191-195.

10.

DICKS A, CONRADIE WJ, DE WET T. Value at risk using Garch volatility models augmented with extreme value theory. Studies in Economics and Econometrics (SEE) 2014; 38(3): 1-18.

11.

FRIGATI L, VAN DER MERWE JP, HARVEY J, RABIE H, THERON GB, COTTON MF. A retrospective review of group B streptococcal infection in the Mero East area of the Western Cape Province: 2010 to 2011. Southern African Journal of Epidemiology and Infection 2014; 29(1): 33-36.

12.

GOWER JC, LE ROUX NJ, LUBBE S. The Canonical Analysis of Distance. Journal of Classification 2014; 1(1): 107-128.

23

27.

NEL WS, BRUWER BW, LE ROUX NJ. Precision, Consistency and Bias in Emerging Equity Markets. Journal of Economics and Behavioral Studies 2014; 6(5): 386-399.

28.

NEL WS, BRUWER BW, LE ROUX NJ. The valuation performance of equity-based multiples in South African context. Risk Governance and Control: Financial Markets & Institutions 2013; 3(3): 35-47.

29.

NELSON MC, ISAACS F, HASSAN S, KIDD M, CRONJE FJ, JANSE VAN RENSBURG S. Prevalence of abnormal bloodflow patterns and effects of biochemistry and lifestyle factors on the major neck vessels in patients with multiple sclerosis in the Western Cape, South Africa. Medical Technology SA 2014; 28(1): 26-33.

temperature on the ascorbic acid content, total phenolic content and antioxidant activity in lettuce (Lactuca sativa L.). Journal of Animal and Plant Sciences 2014; 24(4): 1173-1177. 36.

VAN NIEKERK E, AUTRAN CA, NEL DG, KIRSTEN GF, BLAAUW R, BODE L. Human Milk Oligosaccharides Differ between HIV-Infected and HIV-Uninfected Mothers and Are Related to Necrotizing Enterocolitis Incidence in Their Preterm Very-Low-Birth-Weight Infants 1–3. Journal of Nutrition 2014; 144(8): 1227-1233.

37.

VAN NIEKERK E, KIRSTEN GF, NEL DG , BLAAUW R. Probiotics, feeding tolerance, and growth: A comparison between HIV-exposed and unexposed very low birth weight infants. Nutrition 2014; 30(6): 645-653.

30.

NIEUWOUDT MM, VAN DER MERWE JL, HARVEY J, HALL DR. Pregnancy Outcomes in Super-Obese Women - an Even Bigger Problem? A Prospective Cohort Study. South African Journal of Obstetrics and Gynaecology 2014; 20: 54-59.

38.

VIVIERS MZ, BURGER BV, LE ROUX NJ, MORRIS J, LE ROUX M. Temporal changes in the neonatal recognition cue of Dohne merino lambs (Ovis aries). Chemical Senses 2014; 39(3): 249-262.

31.

PICKEN SC, WILLIAMS S, HARVEY J, ESSER M. The routine paediatric human immunodeficiency virus visit as an intervention opportunity for failed maternal care, and use of point-of-care CD4 testing as an adjunct in determining antiretroviral therapy eligibility. Southern African Journal of Epidemiology and Infection 2014; 29(2): 70-74.

39.

VON GERHARDT K, VAN NIEKERK A, KIDD M, SAMWAYS MJ, HANKS J. The role of elephant Loxodonta africana pathways as a spatial variable in crop-raiding location. Oryx 2014; 3: 436-444.

40.

WASSERMAN E, ORTH H, SENEKAL M, HARVEY J. High prevalence of mupirocin resistance associated with resistance to other antimicrobial agents in Staphylococcus aureus isolated from patients in private health care, Western Cape. Southern African Journal of Epidemiology and Infection 2014; 29(4): 126-132.

41.

WASSERMAN E, ORTH H, SENEKAL M, HARVEY J. High Prevalence of Mupirocin Resistance Associated with Resistance to Other Antimicrobial Agents in Staphylococus aureus Isolated from Patients in Private Health Care, Western Cape. South African Journal of Infectious Diseases 2014; 29: 126-132.

42.

WESSELS CB, MALAN FS, NEL DG, RYPSTRA T. Variation in strength, stiffness and related wood properties in young South African-grown Pinus patula. Southern Forests 2014; 76(1): 37-46.

32.

PRETORIUS WB, DAS S, MONTEIRO PMS. Investigating the Complex Relationship Between In Situ Southern Ocean pC O2 and Its Ocean Physics and Biogeochemical Drivers Using a Nonparametric Regression Approach. Environmental and Ecological Statistics 2014; 21: 697-714.

33.

RANGANAI E, VAN VUUREN JO, DE WET T. Multiple Case High Leverage Diagnosis in Regression Quantiles. Communications in Statistics-Theory and Methods 2014; 43(16): 3343-3370.

34.

REISINGER AK, VAN HEERDEN JD. Is liquidity a pricing factor on the JSE? Studies in Economics and Econometrics (SEE) 2014; 38(1): 17-34.

35.

SEREA C, BARNA O, MANLEY M, KIDD M. Effect of storage

www.sun.ac.za/english/faculty/economy/statistics

24