• A
  • A
  • A
  • ABC
  • ABC
  • ABC
  • А
  • А
  • А
  • А
  • А
Regular version of the site
Important announcements 1

News

Back To Multilevel, Back To Essex

Anna Shirokanova tells about Essex Summer School!

Essex Summer School in Social Science Data Analysis and Collection is considered by many as one of the best and prestigious schools in statistics in the field of social sciences. This year the 45th School has been held. Among its students were researchers of the LCSR. Anna Shirokanova shares her impressions about Essex's courses and peoples.

 

“This time next year” is something I will hardly be able to predict after this summer school. Last year the Lab provided me with funding for attending a course on regression analysis at the Essex Summer School in Data Analysis, which I was happy to attend, together with Anna Nemirovskaya and Daria Oreshina. That was a lot of experience, both academically and socially! A very international class, leading teaching staff from all over the world, up-to-date computation techniques, and a great deal of improvement of my own research – no doubt all this is something to look forward to again! While being in Essex for the first time, I met people attending the Summer School for the second or more times, and I could never imagine at that time I would have such an opportunity myself – another dream came true with the Laboratory!

 

This year the Lab was sending its heralds again to the best summer schools in Europe and the US in search for the best practice and advanced methods in social science data analysis. Imagine my surprise when I realized I was going back to Essex, this time to study “Multilevel Analysis with Applications” (taught by Kelvyn Jones from the University of Bristol, UK). This was not the first bite into the method for me, after a thorough and remarkable course on multilevel modeling (MLM) in HLM taught by Hermann Duelmer at the LCSR 1st Summer School last year, and Eduard Ponarin’s R sessions for MLM at the 2nd LCSR Summer School this year. Overall, this experience is definitely something to reckon!

 

However, the course in Essex was also a step forward, as it was all done in MLwiN (a perfectly British but also intuitively friendly programme for MLM), which was new to me. The lecturer, Kelvyn, made every effort to make us understand more of MLM, with the help of his spiral curriculum, turning back time after time to an empirical example at hand to demonstrate the strengths of MLM. According to Professor Jones, this course was taught for the 20th time in Essex already, which is why probably it was so thoughtfully laid out and so successfully presented.

 

It is worth saying that Essex is not just a summer school, but one of the long-lasting ones, known for its variety of courses, sessions, and students. It is not rare here that a person sitting next to you at lunch could be “your soul-mate” from another country tackling the same research problems as you do – what a surprise but not surprise at all! It is here that we met again Peter Schmidt who had just taught us SEM at the 2nd LCSR Summer School. It is also remarkable that all LCSR delegates to Essex Summer School were able to choose the courses suitable for their specific research interests. This time there were four of us at the session – Natalia Firsova, Olga Gryaznova, Maria Kravtsova, and myself.

 

The sun was behaving truly summerlike these days. We were enjoying sunlight day after day during breaks between sessions throughout the whole first week, but next week the notorious English summer rain took over. In both cases this was a pleasant framework for the courses   “Math for Social Sciences” early in the morning, and “Multilevel Analysis” afterwards.

 


There were traditional social activities for everyone in the evenings: fish and chips at the nearby Wivenhoe on Tuesday, and a bus trip (this time to Cambridge) on Saturday. In Cambridge, “everything was beautiful and nothing hurt”. We visited the historical King’s College Chapel, wandered around the ancient streets to Market Hill, enjoyed punting along the backs of colleges, walked through the yards of colleges on land, delving into antique bookshops with 1749 Shakespeare editions on the shelves. The green and vibrant city made an extremely charming – and refreshing pause from classes.

 

The sessions were full of intriguing paths of decisions on the way to multilevel modelling. This MLM course paid special attention to visual data analysis and interpreting covariance, while MLwiN provided rich opportunities for this. MLwiN has been developing in Bristol for many years now and has grown into a handy and quite friendly environment for MLM. By the way, under the surface there is not only the user interface, but also original algorithms (that can be easily imported into STATA using the “runmlwin” package). To assist researchers, the Bristol Centre for MLM has also created a hugely navigable gallery of articles that are devoted to MLM in various fields.

 

Meanwhile, MLM itself is becoming ever more popular for a social scientist! To use the words of Ita Kreft and Jan de Leeuw, “Once you know hierarchies exist, you see them everywhere” [Kreft, de Leeuw, 1998]. And even though there is a known simplification (data are not always hierarchical and are often cross-sectional), the effervescent development of MLM supported by programmes such as MLwiN promises a whole revolution to the social sciences, where regression analysis has been reigning so far. For, with the strengths of MLM and limitations of regressions in mind, it is quite hard to continuously trust regression models.

 

The consequences of applying MLM to social and political expertise can be already estimated as substantial. Even the old, established data can be successfully re-analyzed and possibly revised. One vivid example is a British discussion on the styles of teaching dating back to the 1970s. In 1976, Nevill Bennett published his study where he used regression analysis to find out that the “traditional” teaching style is most effective in the pupils’ outcome grades. The study spurred up a wave of public interest and appeared in PM’s speeches. However, a ML analysis of the same data by Neil Spencer has showed that, taking into account the endogeneity (and hierarchies) in the data, there is no significant difference in the effectiveness of different teaching styles. Thus, MLM could become a precise and powerful leverage in hands of a researcher.

 

An important note about the software. While one could argue about the aesthetics, users’ level requirement, or, broadly speaking, the ideology behind MLwiN, HLM, and R, at this point all the attempts to run MLM in SPSS have been (not obviously) unsuccessful. Professor Jones has had a few sorrowful passages on the fact that SPSS does not allow the variance of independent variables higher than the 1st level, with reference to a fresh and thick 2012 volume on the subject…

 

In conclusion, one last jump to the future (and the present) of MLM. Working in different packages and exporting data is still rather a necessity for data analysis involving MLM today. They have been developing and have recently released a beta-version of a system called Stat-JR at the Bristol Centre for MLM. The goal of this project is to provide the fullest interoperability for the data analysis software. What is more, anyone can try it today (and have it later for free)! Another piece of good news (especially for the beginners) is the possibility of automatic creation of individually tailored, illustrated manuals that would be available right after loading the data and fixing the model!

 

Thus, self-education is the limit. Meanwhile, this year the Essex MLM course is fully booked. Next year there are plans for an advanced MLM course with Kelvyn Jones (including multilevel factor analysis, a closer look at MCMC models, etc.) I am glad to say now that many of the Lab’s researchers are ready for these advanced courses in data analysis, and that within the LCSR we are delivering a refreshingly new quality to quantitative models. LCSR is now in the forefront of data analysis, with its brave researchers sometimes fighting windmills like Don Quixote, but much more often taking over the difficulties of the data, like St. George defeating the dragon. Bon chance à nous tous!

by Anna Shirokanova