Wednesday, May 29, 2013

Here is the full text of my article on the QS Subject Rankings published in the Philippine Daily Inquirer.


The QS university rankings by subject: Warning needed


 107  360  8
It is time for the Philippines to think about constructing its own objective and transparent ranking or rating systems for its colleges and universities that would learn from the mistakes of the international rankers.

The ranking of universities is getting to be big business these days. There are quite a few rankings appearing from Scimago, Webometrics, University Ranking of Academic Performance (from Turkey), the Taiwan Rankings, plus many national rankings.

No doubt there will be more to come.

In addition, the big three of the ranking world—Quacquarelli Symonds (QS), Times Higher Education and Shanghai Jiao Tong University’s Academic Ranking of World Universities—are now producing a whole range of supplementary products, regional rankings, new university rankings, reputation rankings and subject rankings.

There is nothing wrong, in principle, with ranking universities. Indeed, it might be in some ways a necessity. The problem is that there are very serious problems with the rankings produced by QS, even though they seem to be better known in Southeast Asia than any of the others.
This is especially true of the subject rankings.

No new data

The QS subject rankings, which have just been released, do not contain new data. They are mostly based on data collected for last year’s World University Rankings—in some cases extracted from the rankings and, in others, recombined or recalculated.

There are four indicators used in these rankings. They are weighted differently for the different subjects and, in two subjects, only two of the indicators are used.

The four indicators are:
A survey of academics or people who claim to be academics or used to be academics, taken from a variety of sources. This is the same indicator used in the world rankings. Respondents were asked to name the best universities for research.
A survey of employers, which seem to comprise anyone who chooses to describe himself or herself as an employer or a recruiter.
The number of citations per paper. This is a change from the world rankings when the calculation was citations per faculty.
H-index. This is something that is easier to give examples for than to define. If a university publishes one paper and the paper is cited once, then it gets an index of one. If it publishes two or more papers and two of them are published twice each, then the index is two and so on. This is a way of combining quantity of research with quality as measured by influence on other researchers.

Out of these four indicators, three are about research and one is about the employability of a university’s graduates.

These rankings are not at all suitable for use by students wondering where they should go to study, whether at undergraduate or graduate level.

The only part that could be of any use is the employer review and that has a weight ranging from 40 percent for accounting and politics to 10 percent for arts and social science subjects, like history and sociology.

But even if the rankings are to be used just to evaluate the quantity or quality of research, they are frankly of little use. They are dominated by the survey of academic opinion, which is not of professional quality.

There are several ways in which people can take part in the survey. They can be nominated by a university, they can sign up themselves, they can be recommended by a previous respondent or they can be asked because they have subscribed to an academic journal or an online database.
Apart from checking that they have a valid academic e-mail address, it is not clear whether QS makes any attempt to check whether the survey respondents are really qualified to make any judgements about research.

Not plausible

The result is that the academic survey and also the employer survey have produced results that do not appear plausible.

In recent years, there have been some odd results from QS surveys. My personal favorite is the New York University Tisch School of the Arts, which set up a branch in Singapore in 2007 and graduated its first batch of students from a three-year Film course in 2010.In the QS Asian University Rankings of that year, the Singapore branch got zero for the other criteria (presumably the school did not submit data) but it was ranked 149th in Asia for academic reputation and 114th for employer reputation.

Not bad for a school that had yet to produce any graduates when the survey was taken early in the year.

In all of the subject rankings this year, the two surveys account for at least half of the total weighting and, in two cases, Languages and English, all of it.

Consequently, while some of the results for some subjects may be quite reasonable for the world top 50 or the top 100, after that they are sometimes downright bizarre.

The problem is that although QS has a lot of respondents worldwide, when it gets down to the subject level there can be very few. In pharmacy, for example, there are only 672 for the academic survey and in materials science 146 for the employer survey. Since the leading global players will get a large share of the responses, this means that universities further down the list will be getting a handful of responses for the survey. The result is that the order of universities in any subject in a single country like the Philippines can be decided by just one or two responses to the surveys.

Another problem is that, after a few obvious choices like Harvard, MIT (Massachusetts Institute of Technology), Tokyo, most respondents probably rely on a university’s general reputation and that can lead to all sorts of distortions.

Many of the subject rankings at the country level are quite strange. Sometimes they even include universities that do not offer courses in that subject. We have already seen that there are universities in the Philippines that are ranked for subjects that they do not teach.

Somebody might say that maybe they are doing research in a subject while teaching in a department with a different name, such as an economic historian teaching in the economics department but publishing in history journals and getting picked up by the academic survey for history.
Maybe, but it would not be a good idea for someone who wants to study history to apply to that particular university.

Another example is from Saudi Arabia, where King Fahd University of Petroleum and Minerals was apparently top for history, even though it does not have a history department or indeed anything where you might expect to find a historian. There are several universities in Saudi Arabia that may not teach history very well but at least they do actually teach it.

These subject rankings may have a modest utility for students who can pick or choose among top global universities and need some idea whether they should study engineering at SUNY (State University of New York) Buffalo (New York) or Leicester (United Kingdom) or linguistics at Birmingham or Michigan.

But they are of very little use for anyone else.

Tuesday, May 28, 2013

The QS university rankings by subject: Warning needed

My article in the Philippine Daily Inquirer can be accessed here.

Monday, May 27, 2013

The QS Subject Rankings: Not Everybody is Impressed

The subject rankings just released by QS seem to be a shrewd marketing move. Dozens of universities around the world have learnt that they have been ranked for something by the renown and revered QS, which will look good in their promotional literature.

Some people are not impressed. Brian Leiter, the law scholar and philosopher asks whether they are a fraud on the public. See here for his answer.

Why are they so worried?

The UK HE International Unit represents the views of the British university sector and is cooperating with Times Higher Education and Thomson Reuters in the organising of this week's Global University Summit in London, a Prestigious Event in a Spectacular Setting.

It has just issued a policy statement about the slowly emerging U-Multirank project, which is also discussed by David Jobbins in University World News.

The Unit has expressed a number of concerns. These include the overcrowding of the league table market, reliance on self-reported data which lack validity, the combining of incommensurate variables to create a league table, the risk of " becoming a blunt instrument that would not allow different strengths across an institution to be recognised" and diverting EU funds from other priorities. It claims that U-Multirank may "harm rather than benefit the sector."

It is difficult to see why the International Unit is getting so concerned. I agree that self reported data may lack validity but the QS and Times Higher Education global rankings also include such data. The combining of incommensurate variables is the essence of ranking. Sometimes blunt instruments are appropriate. A scalpel is of little use for hammering nails and a tool like U-Multirank may have uses which existing rankings do not.

As for the 2 million Euros, this is trivial compared with some of the things the EU has been wasting money on in recent years.

Saturday, May 25, 2013

The Efficiency Rankings

Times Higher Education has a story about a study by Dirk Van Damme, head of the Centre for Educational Research and Innovation at the OECD. This will be presented at the Global University Summit  held in Whitehall, London from the 28th to the 30th May.

The Summit "brings an invitation-only audience of leaders from the world’s foremost universities, senior policy-makers and international business executives to London in 2013." It is a "prestigious event" held in a "spectacular setting" and is sponsored by the University of Warwick, Times Higher Education, Thomson Reuters and UK Universities International Unit. Speakers include Vince Cable, Boris Johnson, the Russian ambassador and heads of various universities from around the world.

What Professor Van Damme has done is to treat the THE World University Rankings Research Indicator scores as an input and the Research Influence (Citations) scores as an output. The output scores are divided by the input scores and the result is a measure of the efficiency with which the inputs are turned into citations, which, as we all know, is the main function of the modern university.

According to THE:

"The input indicator takes scaled and normalised measures of research income and volume into account, and also considers reputation, while the output indicator looks at citations to institutional papers in Thomson Reuters’ Web of Science database, normalised for subject differences.
Professor van Damme said that the results - which show that university systems outside the Anglo-American elite are able to realise and increase outputs with much lower levels of input - did not surprise him.
“For example, Switzerland really invests in the right types of research. It has a few universities in which it concentrates resources, and they do very well,” he said.
Previous studies have found the UK to have the most efficient research system on measures of citation per researcher and per unit of spending.
But Professor van Damme explained that under his approach, productivity - output per staff member - was included as an input.
“With efficiency I mean the total research capacity of an institution, including its productivity, divided by its impact. The UK is not doing badly at all, but other countries are doing better, such as Ireland, which has a very low research score but a good citations score,” he said.
Given the severity of the country’s economic crisis, Ireland’s success was particularly impressive, he said.
“I think it is really conscious of the effort it has to make to maintain its position and is doing so.”
Low efficiency scores for China and South Korea reflected the countries’ problems in translating their huge investment into outputs, he added."

One hesitates to be negative about a paper presented at a prestigious event in a spectacular setting to an invitation only audience but this is frankly rather silly.

I would accept that income can be regarded as an input but surely not reputation and surely not volume of publications. Also, unless Van Damme's methodology has undisclosed refinements he is treating research scores as having the same value regardless of whether they are composed mainly of scores for reputation or for number of publications or for research income.

Then there is the time period concerned. Research income is income for one year, Publications are drawn from a five year period. These are then compared with citations over a six year period. So the paper is asking how research income for 2010 produces citations in the years 2006 - 2011 of papers published in the years 2006 - 2010. A university is certainly being remarkably efficient if its 2010 income is producing citations in 2006, 2007, 2008 and 2009.

Turning to the citations side of the equation, it should be recalled that the THE citations indicator includes an adjustment by which the citation impact score for universities is divided by the square root of the citation impact score for the country as a whole. In other words a university located in a country where papers are not cited very much gets a big boost and the lower the national citation impact score the bigger the boost for the university. This is why Hong Kong universities suffered reduced scores when Thomson Reuters took them out of China when counting citations and put them in their own separate category.

So, it is not surprising that universities from outside the Anglo-Saxon elite do well for citations and thus appear to be very efficient. Thomson Reuters methodology gives such universities a very substantial weighting just for being located in countries that are  less productive in terms of citations.

None of this is new. In 2010 Van Damme did something similar at a seminar in London.

Van Damme is just analysing the top 200 universities in the THE rankings. It would surely be more interesting to analyse the top 400 whose scores are obtainable from an iPad/iPhone app.

So here are the top ten universities in the world according to the efficiency with which they turn income, reputation and publications into citations. The procedure is to divide the citations score from the 2012 THE rankings by the research indicator score.

1.  Tokyo Metropolitan University
2.  Moscow State Engineering Physics Institute
3.  Florida Institute of Technology
4.  Southern Methodist University
5.  University of Hertfordshire
6.  University of Portsmouth
7.  King Mongkut's University of Technology
8.  Vigo University
9.  Creighton University
10. Fribourg University

No doubt the good and the great of the academic world assembled in Whitehall will make a trip to Portsmouth  or even to Vigo or Creighton if they can find them on the map.

And now for the hall of shame. Here are the bottom ten of the THE top 400, ranked according to efficiency as measured by citations indicator scores divided by research scores. The heads of these failing institutions will no doubt be packing their bags and looking for jobs as junior administrative assistants at technical colleges in Siberia or the upper Amazon

391.  Tsinghua University
392.  Chinese University of Hong Kong
393.  National Taiwan University
394.  National Chiao Tung University
395.   Tilburg University
396.  Delft University of Technology
397.  Seoul National University
398.  State University of Campinas
399.  Sao Paulo University
400.  Lomosonov Moscow State University

In a little while I hope to publish the full 400 after I have finished being sarcastic about the QS subject rankings.

Update on IREG Approval 

  • The International Ranking Experts group has also given its approval to the national ranking produced by the Perspektywy Education Foundation of Poland.

  • The approval given to the QS World, Asian and Latin American Rankings does not apply to the QS Stars.
Details here.

Saturday, May 18, 2013

The First IREG Audit 

QS is the first ranking organisation to get the seal of approval from the International Ranking Experts Group (IREG) for its World, Asian and Latin American rankings

The IREG audit process would appear on the surface to be quite rigorous. Take a look at the audit manual. There are a  number of criteria some of which sound quite daunting but are not really so. For example, Criterion 8 says:

 "If rankings are using composite indicators the weights of the individual indicators have to be published. Changes in weights over time should be limited and due to methodological or conception-related considerations."

Fair enough, but there is nothing about how weighting should be distributed across the indicators in the first place. Forty per cent for the academic survey in the QS rankings?

Some indicators are obvious -- providing a contact address. Others are so vague that they mean very little -- organisational measures that enhance the credibility of rankings.

The basic principle of the audit is that ranking organisations are given scores ranging from 1 (not sufficient/not applied) to 6 (distinguished) for the various criteria, with a double weighting for core criteria. The maximum score is 180 and 

"On the bases of the assessment scale described
above, the threshold for a positive audit decision will
be 60 per cent of the maximum total score. This
means the average score on the individual criteria
has to be slightly higher than “adequate”. In order
to establish the IREG Ranking Audit as a quality
label none of the core criteria must be assessed
with a score lower than three."

So a positive result could mean that an organisation is distinguished in everything. It could also mean that it is on average slightly higher than adequate. It would be interesting to know which applies to QS.

I do not know whether the auditors had any criticisms to make. If not it is difficult to see the point of the exercise. If they did it would be nice to know what they were.

QS are to be commended for submitting to the audit although it probably was not very searching but it still seems that the ranking world needs more and better monitoring and observation.

Wednesday, May 15, 2013

QS Rankings by Subject

QS have produced their annual subject rankings. At the top there are no real surprises and, while there is certainly room for argument, I do not think that anyone will be shocked by the top ten or twenty in each subject.

The university with the most number ones is Harvard:

Pharmacy and Pharmacology
Earth and Marine Sciences
Politics and International Studies
Economics and Econometrics
Accounting and Finance

MIT has seven:
Computer Science
Chemical Engineering
Electrical Engineering
Mechanical Engineering
Phys and Astronomy
Materials Science

Then there is Berkeley with exactly the four you would expect:
Environmental Science
Statistics and Operational Research
Communication and Media Studies

Oxford has three:

Modern Languages

Cambridge another three:

Imperial College London is top for Civil Engineering and University of California, Davis for Agriculture and Forestry.

These rankings are based on the academic opinion survey, the employer survey, citations per paper and h-index, a measure of both output and influence that eliminates outliers, in proportions that vary for each subject. They are very research-focused which is unfortunate since there seems to be a consensus emerging at conferences and seminars that the THE-TR rankings are for policy makers, the Shanghai ARWU for researchers and the QS rankings for undergraduate students.

Outside the top fifty there are some oddities resulting form the small number of responses when we leave the top fifty or top one hundred. I will leave it to specialists to find them.

Wednesday, May 08, 2013

Unsolicited Advice

There has  been a lot of debate recently about the reputation survey component in the QS World University Rankings.

The president of University College Cork asked faculty to find friends at other universities who "understand the importance of UCC improving its university world ranking". The reason for the reference to other universities is that the QS survey very sensibly does not permit respondents to vote for their own universities, those that they list as their affiliation.  

This request appears to violate QS's guidelines which permit universities to inform staff about the survey but not to encourage them to nominate or refrain from nominating any particular university. According to an article in Inside Higher Ed QS are considering whether it is necessary to take any action.

This report has given Ben Sowter of QS sufficient concern to argue that it is not possible to effectively manipulate the survey.  He has set out a reasonable case why it is unlikely that any institution could succeed in marching graduate students up to their desktops to vote for favoured institutions to avoid being sent to a reeducation camp or to teach at a community college.

However, some of his reasons sound a little unconvincing: signing up, screening, an advisory board with years of experience. It would help if he were a little more specific, especially about the sophisticated anomaly detection algorithm, which sounds rather intimidating.

The problem with the academic survey is not that an institution like University College Cork is going to push its way into the global  top twenty or top one hundred  but that there could be a systematic bias towards those who are ambitious or from certain regions. It is noticeable that some universities in East and Southeast Asia do very much better on the academic survey than on other indicators. 

The QS academic survey is getting overly complicated and incoherent. It began as a fairly simple exercise. Its respondents were at first drawn form the subscription lists of World Scientific, an academic publishing company based in Singapore. Not surprisingly, the first academic survey produced a strong, perhaps too strong, showing for Southeast and East Asia and Berkeley. 

The survey turned out to be unsatisfactory, not least because of an extremely small response rate. In succeeding years QS has added respondents drawn from the subscription lists of Mardev, an academic database, largely replacing those from World Scientific, lists supplied by universities, academics nominated by respondents to the survey and those joining the online sign up facility. It is not clear how many academics are included in these groups or what the various response rates are. In addition, counting responses for three years unless overwritten by the respondent might enhance the stability of the indicator but it also means that some of the responses might be from people who have died or retired.

The reputation survey does not have a good reputation and it is time for QS to think about revamping the methodology. But changing the methodology means that rankings cannot be used to chart the progress or decline of universities over time. The solution to this dilemma might be to launch a new ranking and keep the old one, perhaps issuing it later in the year or giving it less prominence.

My suggestion to QS is that they keep the current methodology but call it the Original QS Rankings or the QS Classic Rankings. Then they could introduce the  QS Plus or New QS rankings or something similar which would address the issues about the academic survey and introduce some other changes. Since QS are now offering a wide range of products, Latin American Rankings, Asian Rankings, subject rankings, best student cities and probably more to come, this should  not impose an undue burden.

First, starting with the academic survey, 40 percent is too much for any indicator. It should be reduced to 20 per cent.

Next, the respondents should be divided into clearly defined categories, presented with appropriate questions and appropriately verified.

It should be recognised that subscribing to an online database or being recommended by another faculty member is not really a qualification for judging international research excellence. Neither is getting one’s name listed as corresponding author. These days that  can have as much to do with faculty politics as with ability.  I suggest that the academic survey should be sent to:

(a) highly cited researchers  or those with a high h-index who should be asked about international research excellence;
(b) researchers drawn from the Scopus database who should be asked to rate the regional or national research standing of universities.

Responses should be weighted according to the number of researchers per country.

This could be supplemented with a survey of student satisfaction with teaching based on a student version of the sign up facility and requiring a valid academic address with verification.

Also, a sign up facility could be established for anyone interested and asking a question about general perceived quality.

If QS ever do change the academic survey they might as well review the other indicators. Starting with the employer review, this should be kept since, whatever its flaws, it is an external check on universities. But it might be easier to manipulate than the academic survey. Something was clearly going on in the 2011 ranking when there appeared to be a disproportionate number of respondents from some Latin American countries, leading QS to impose caps on universities exceeding the national average by a significant amount. 

"QS received a dramatic level of response from Latin America in 2011, these counts and all subsequent analysis have been adjusted by applying a weighting to responses from countries with a distinctly disproportionate level of response."

It seems that this problem was sorted out in 2012. Even so, QS might consider giving   half the weighting for this survey to an invited panel of employers. Perhaps  they could also broaden their database by asking NGOS and non-profit groups about their preferences.

There is little evidence that overall the number of international students has anything to do with any measure of quality and it also may have undesirable backwash effects as universities import large numbers of less able students. The problem is that QS are doing a good business moving graduate students across international borders so it is unlikely that they will ever consider doing away this indicator.

Staff student ratio is by all accounts a very crude indicator of quality of teaching. Unfortunately, at the moment there does appear to be any practical alternative. 
One thing that QS could do is to remove research staff from the faculty side of the equation. At the moment a university that hires an army of underpaid research assistants  and sacks a few teaching staff, or packs them off to a branch campus, would be recorded as having brought about a great improvement in teaching quality.

Citations are a notoriously problematical way of measuring research influence or quality. The Leiden Ranking shows that there are many ways of measuring research output and influence. It would be a good  idea to combine several different ways of counting citations. QS have already started to use the h- index in their subject rankings starting this year and have used citations per paper in the Asian University Rankings.

With the 20 per cent left over from reducing the weighting for the academic survey QS might consider introducing a measure of research output rather than quality since this would help distinguish among universities outside the elite and perhaps use internet data from Webometrics as in the Latin American rankings.