BestCourse4Me is a not-for-profit website which aims to allow UK students to make more informed choices about university courses.
Contents
Using information collected nationally, it shows the link between what people study at University and the jobs they go into afterwards. It claims to be totally independent, impartial, and free to use. Users of the website can compare:
The site is somewhat controversial in that it allows direct comparisons of supposedly equivalent courses at different institutions, and may also encourage over-simplistic comparisons of courses. Others argue that having some data is a great improvement on having none. A stated aim of the site is to assist those from disadvantaged backgrounds, who may not have sufficient information about universities and careers from other sources.
The site has won high-level support from some UK politicians, most notably David Cameron (the Prime Minister of the United Kingdom) and David Willetts (the Minister of State for Universities and Science), and from celebrities such as Andy McNab, who does the voiceover for some of the site's video tutorials.
The site is provided by Student Information Services Ltd (SIS). SIS Ltd was established by the philanthropists Steve and Ros Edwards as a not for profit organisation aiming to provide clear information to prospective students about the outcomes and value of a higher education so they can make better informed choices.
Data
The site describes their data processing as follows:
The site is currently driven by two key data sources: the UK Quarterly Labour Force Survey (QLFS) conducted by the Office for National Statistics, and the Higher Education Statistics Agency Destination of Leavers from Higher Education survey (DLHE). QLFS data is used mainly to populate the “Career” section of the site, whereas DLHE data is used in the “Subject”, “Uni” and “MyChoice” sections.
The Quarterly Labour Force Survey captures data from thousands of households across the UK and includes information about employment status, occupation (using the Standard Occupational Classification System (SOC)), industry, hourly earnings, age, gender and qualifications.
The processing of the raw QLFS data involves combing the data of four sequential quarters together, and then aggregating over one or more specific fields, such as SOC code, to derive results based on Occupation (e.g. average hourly earnings for bus drivers), or SOC code and Age to derive results based on Occupation and Age (e.g. average weekly earnings for bus drivers aged 39).
When aggregating over fields, care must be taken to ensure that the averages and percentages are calculated from a reasonably-sized population. BestCourse4Me follows the HESA aggregation strategy and ensures that any average is calculated from more than 10 members, and any percentage is calculated from a population greater than 52. Population counts are quantized following HESA rules and are rounded to the nearest multiple of 5. Where actual (i.e. un-quantized) population counts are smaller than these limits, the results are suppressed. Observing and allowing for these rules is important, particularly when the data is ‘drilled-down’ to finer levels of granularity (e.g. average salary of male bus drivers who have graduated from university and are aged 39 in 2007). The finer the granularity of aggregation, the more likely the limits are reached.
For some results, such as average hourly pay by age and occupation, there are insufficient data to satisfy the statistical limits when the data is partitioned into separate years. Thus for a few results, all quarters (2002–2009) are combined together before the aggregation is performed. At present, the effect of inflation on wages is not taken into account over these multi-year aggregations: this shortcoming will be addressed in a forthcoming update.
The DLHE data records what UK graduates are doing six months after graduation. Employment status and earnings, occupation, subject and University are captured by the survey. The data is partitioned into yearly blocks, running from 1 August until 31 July of the following year, and so unlike the QLFS data, no special combining is required before aggregation. However, most results based on DLHE data have required aggregation over all years (2002–2008) in order to meet the averaging and percentage minimum population thresholds. Aggregations follow a similar pattern to those for QLFS data, except that they are usually orientated around Subject (via Joint Academic Classification of Subjects (JACS) code) or Higher Education Institution, or both.
One issue of the DLHE data is that it captures the annual salary (of employed graduates), rather than the hourly earnings captured by the QLFS. This makes earnings comparisons across the datasets quite difficult; it is unsafe to presume that an annual salary can be simply divided by the typical number of working hours in a year.
An interesting feature of BestCourse4Me is its ability to show the graduate “premium” for specific subjects - i.e. the increase in earnings due to being a graduate in a subject (LFS data on its own is not sufficient to answer this question as it does not capture the subjects studied in great detail). BestCourse4Me does this in the “Subject” section of the website by linking the DLHE data with the QLFS data via occupational codes. There is typically a significant difference between the earnings of graduates and non-graduates, and in some subjects (e.g. Finance) the difference is very pronounced. The aggregation algorithm is as follows:
For a given subject, e.g. English, and from the DLHE data, find all the weighted occupations in which employed graduates are working. For each of these occupations, find the hourly earnings from the QLFS data for graduates and for non-graduates. Average together, accounting for the weights of the occupations.
Technology
The site is built in Ruby on Rails and uses a PostgreSQL database. It is hosted on a Microsoft Windows platform and is currently serviced by the Amazon cloud (AWS). The website uses Google Visualisations to display graphs and tables in multiple web browsers. Proprietary display technologies were specifically avoided to reduce technical barriers to the site’s uptake. There are two databases: (i) a master database which contains all the “raw” QLFS and DLHE data (ii) a production database which contains derived and aggregated data calculated from the raw data The production database is heavily indexed and is loaded onto a hosted server where it is used by the website mainly for lookup operations, rather than calculations. The raw database is held separately. From time to time, when a new aggregation is defined or when new data is added to the raw database, the production database is regenerated. The production database is approximately 20mb in size, whereas the raw database is over 1gb in size.