What is the Impact of Universities on the UK Economy?

Universities UK alert me to a report, The impact of universities on the UK economy, PDF here, from April 2014.

What does it say?  Here is a quote

The report highlights universities’ increasingly significant impact on the economy in terms of output, contribution to GDP, job creation, and overseas investment. It also estimates the economic activity generated elsewhere in the economy through the knock-on effects of expenditure by universities, their staff, and international students.
The report finds that in 2011–12, the UK higher education sector:
• generated over £73 billion of output – up 24% from £59 billion in 2009
• contributed 2.8% of UK GDP in 2011 – up from 2.3% in 2007
• generated 2.7% of all UK employment and 757,268 full-time-equivalent jobs
• generated £10.7 billion of export earnings for the UK

In the report, P.4 we have some multiplier or rate of return type calculations

For every 100 full-time jobs within the universities themselves, another 117 full-time-equivalent jobs

were generated through knock-on effects. 373,794 full-time-equivalent jobs in other sectors of the

UK economy were dependent on the expenditure of the universities.

• For every £1 million of university

output a further £1.35 million of output was generated in other

sectors of the economy. This meant that anadditional £37.63 billion of output was generated

outside the universities as a result of theirexpenditure.

•For every £1 million of university GVA a further £1.03million of GVA was generated in other industries.

And this enables the website to say

Universities generate more GDP per unit of expenditure than other sectors including health, public administration and construction

What was done to get these estimates? P.6 says

The model used was a purpose designed and specially constructed ‘type II’ input-output model

based on actual UK data derived from the Office forNational Statistics’ input-output tables together with

data from its Blue Book.

And this is in fact a method which is widely used in many other studies, e.g. the impact of tourism.  

 I would think that most economists don't know anything about this method, despite it using Economic data and answering an economic question. And you never read this kind of thing in reputable economics journals.  Indeed, in a review of the method, Siegfried et al say,
The Economic Impact of Colleges and Universities, that this kind of work is damaging to universities

"If these economic impact studies were conducted at the level of accuracy most institutions require of faculty research, their claims of local economic benefits would not be so preposterous, and, as a result, trust in and respect for higher education officials would be enhanced."

 So what is going on?  Siegfried et al have a very good review, which I try to outline here.

1. The basic method in this work uses input/output tables.  All statistical agencies produce such tables.  Suppose the economy consist of dairy farmers, milk bottlers and opera singers.  An input/output table describes the inputs and outputs of industries and products, so for example, the farmers produce milk, which is an input to the bottlers, who then sell to the opera singers.  At the same time, farmers like opera, so the output of the opera industry is an input into farming.  Such a method can therefore tell you the payment flows between sectors. (Wonkish note: such data depends on detailed purchase information, current UK tables use purchase data from 2004).

2. Does this answer the question "what is the Impact of Opera Singing on the Economy"?  In a some sense it does.   The farmers want to listen to opera and so that is a first pass estimate of the impact on the economy.  But the opera input into farming is then an indirect input into bottling.  So one can calculate the indirect contributions as well.

3. So this is commonly how the impact of universities or football stadia or making movies is calculated.  The university is in location X.   Students pay £100 to attend.  The university buys £20 of goods and employs £80 worth of workers, say 80 workers. P.15 of the UK report therefore sets out primary and secondary or "knock-on" effects.

• The primary effect is the university spending and employment itself, (here £100 and 80 workers), the actual UK figure being £26.68bn in 2011-2. 
• the secondary effect is two types which the report calls
• indirect effects: purchasing of goods and services by the university who then buy from others e.g. buying paper, but the paper industry then buys machines etc.
• induced effects: the 80 workers buy goods, which then employs other workers etc. e.g. if a worker buys a bottle of milk, this supports farmers and opera singers via the input/output relations set out above).  

This is how the report can claim that (p.16)

"Universities spent some £26.68 billion in 2011–12. This expenditure generated £37.63 billion of output in other UK industries"

4. All this sounds very plausible.  However, but a moment's thought will show that falls prey to a key objection: what is the counterfactual? Whilst this sounds abstract it is important. As Siefried et al say

The key question posed in studies designed to measure the local impact of a

college is how much better off are area residents with the institution there than they

would be in its absence. “Better off” is usually defined as higher employment, per capita

income or local tax revenue. Both common sense and standard regional economic

analysis say the proper procedure is to compare economic indicators in the presence of

the institution with predictions of those same indicators “but for” the college – that is,

compare actual to “counterfactual” outcomes.

They continue with a key point

From this perspective that portion of an institution’s economic activity that would remain in the local area even if the institution were not there is not a contribution to the local economy. Few studies of the local economic impact of colleges and universities explicitly articulate such a counterfactual.

Take then the contribution of  college X in London.  It employs, say Y cleaners.  Is the contribution to include those cleaners?  The correct counterfactual is to ask "would those cleaners be employed without college X?"  Or take students.  The correct counterfactual is to ask "would students be students without college X?" If the answer is that cleaners would anyway be cleaners and students anyway students, then the college is making no additional contribution.

Or, take a university hospital.  Without the university, would there be no hospital?

So, the method in principle traces through the network of payments currently existing (if the data are correct etc. etc. ).  But if the question is: what would those payments be if the university did not exist, the method cannot answer that without additional assumptions.

5.  Let us make an assumption then.  In an economy operating at full employment we might then, on this measure, score the university additional contribution, that is, the contribution relative to its not being there, as, well, zero. Sure the university buys cleaning services, books and electricity (who then spend out of their sectors).  But if it were not there then in a full employment economy by assumption cleaners, books and electricity would be bought anyway.



6. Even if one wants to use the input/output method, Siegfried have some very useful points (suppose you think about a rural university, which if disappeared, would not be replaced).

• studies often count spending by students and also the college. Be careful of double counting.  When a student spends on fees that then leads to spending by the college, so counting both is wrong (counting a student spend on haircuts is ok)
• be careful if a student would have gone elsewhere in the area.  The additional contribution to London of a university in London should not include a student who would have gone to another college in London.  
• In practice many University hospitals spending dominates spend.  These should not be included.  They say "
  
  The revenues and expenditures of university hospitals usually dwarf the rest of the institution. Seldom
  do medical center expenditures contribute much to local economic development, however. Teaching
  hospitals usually are surrounded by other acute care medical facilities. In such circumstances, were the
  university hospital to evaporate, most of the medical services provided by it would be assumed by other
  local hospitals. Only patients with specialized medical problems would likely turn to hospitals outside the
  area. Thus, most university hospital expenditures should not be included in the first-round of expenditures,
  perhaps an exception being isolated university hospitals that serve broad geographic areas in the plains and
  mountain states". 

 7. Note that ascribing benefits of educated workers, which is often done, is not clear either.  Educated workers get some of the benefits from themselves in terms of higher earnings, but they might move from the area and might be in the area anyway even if there was no college (and such local benefits might just be capitalised in local house prices).  If there are spillover benefits to lower crime etc. that might be included however, again, as long as those graduates would not have been in the area.

8.  Finally, a note from my time at the Treasury.  One week an official might get this kind of report, suggesting universities support 10% of GDP.  Next week, movies support 15% of GDP and the following week the car industry supports 25% of GDP.  Pretty soon, these reports account for 150% of GDP.  This tells you there is a fallacy of composition, which is again a consequence of the counterfactual: all these reports typically assume that particular industry in question would not exist at all and no other indirect effects would occur.  So reports like this will not convince officials in the Treasury at least.

These are the main objections to these types of methods and suggest different methods might be used that properly show the additional contributions.  One such is the point that university knowledge spillovers to firms are often local and specific to the university: our work on spillovers from the science base for example. 

Various teaching links

1. Professors on Corporate Boards Increase Profits.  of course.

2. the FT on the UK productivity puzzle: Osborne's unorthodox solution to the UK productivitypuzzle

3. Creative destruction, the London Knowledge Centre is closing. And a letter in the FT makes the point that 

If they really want to learn something useful, these people should learn Chinese as they will only need to show knowledge of 2,600 characters to pass instead of 25,000 street names and locations for “the Knowledge".

Various teaching links: broadband and innovation

1. Broadband raises productivity and demand for the skilled, lowers for the unskilled (Quarterly Journal of Economics (2015), 1781–1824).

This is an interesting paper that uses matched employee and employer data for Norway, using variation in Broadband availability across regions to measure the broadband effects on various meaures, including skilled and unskilled output elasticities.   Broadband lowers (raises) the output elasticity of the unskilled (skilled) with an overall effect on total factor productivity of (p.1809) a 10 point rise in availability of 0.4%.  The overall change in availability is a bit hard to see, but figure 1 suggests that most areas of Norway had zero availability in 2001 but 75% or above in 2005.  So if availability rose by, say 80 percentage points in 4 years, roughly TFP rose by 0.8% per year.


Update.
A new paper by Rosa Sanchis and co-authors does not however find such good results for Broadband speed on learning by kids.(summary here)

The abstract

Governments are making it a priority to upgrade information and communication technologies (ICT) with the aim to increase available internet connection speeds. This paper presents a new empirical strategy to estimate the causal effects of these policies, and applies it to the questions of whether and how ICT upgrades affect educational attainment. We draw on a rich collection of microdata that allows us to link administrative test score records for the population of English primary and secondary school students to the available ICT at their home addresses. To base estimations on exogenous variation in ICT, we notice that the boundaries of usually invisible telephone exchange station catchment areas give rise to substantial and es-

sentially randomly placed jumps in the available ICT across neighboring residences. Using this design across more than 20,000 boundaries in England, we find that even very large changes in available broadband connection speeds have a precisely estimated zero effect on educational attainment. Guided by a simple model we then bring to bear additional microdata on student time and internet use to quantify the potentially opposing mechanisms underlying the zero re-duced form effect. While jumps in the available ICT appear to increase student consumption of online content, we find no significant effects on student time spent studying online or offline, or on their learning productivity.

2. McKinsey have a new report on China Innovation.  They focus on innovation, measured by TFP growth, this from the Executive Summary.

Without labor force expansion and investment to propel growth, China must rely more
heavily on innovation that can improve productivity. We use multifactor productivity—growth that does not come from factors of production such as labor and capital investment—as a proxy for the macroeconomic impact of innovation broadly defined (including productivity gain from catch-up). The contribution to GDP of multifactor productivity has been falling in China, from nearly half of yearly GDP growth in the 1990 to 2000 decade to 30 percent in the past five years. To reach the growth target of 5.5 to 6.5 percent per year (the current consensus view from five leading economic institutions), multifactor productivity growth will need to contribute 35 to 50 percent of GDP growth, or two to three percentage points per year of GDP (Exhibit E1).

Various teaching links

Spencer Dale, Chief Economist at BP on the New Economics of Oil. Very interesting: paper here.  Thanks to my past student Sav for the link.

Have ONS data revisions solved the productivity puzzle?

(This is a corrected post following my earlier post, for readers of the earlier post, please see Update note below)

The ONS yesterday revised growth up with new GDP data.  Today, they have released new productivity data which uses this new output data.  What difference does it make to the productivity puzzle?  Answer: it changes the dates of it and solves some of it, but not all.

1. The revisions are to real output, mostly of the service sector, says the ONS. Very little revisions to hours/jobs.
2. The figure shows annual average growth in real output per hour, whole economy, using the new and the older data.  You can see the following

• with the new data productivity growth was lower in the 2000-07. 
• with the new data, the downturn in productivity came earlier, in 2008.
• with the new data, the dip in productivity and recovery in 2009/10 was not as large
• with the new data, there is a dip down in 2012, but recovery since then.

All this means that the averaged periods look like this (all data, output per hour CAGRs)

Years	old data	new data
1995-00	2.17	2.29
2000-07	2.02	1.97
2007-10	-0.13	-0.37
2010-14	-0.08	0.30

Finally, the productivity gap, that is the productivity we would have expected in 2014 had the trend 1985-2007 (2.17%pa) continued after 2007  was 16.8 on the old data, but 16.1 on the new, hardly reducing the gap.

But we might do another calculation, which is to project forward productivity on the basis of trends 2000-07.  If we do that the old data trend was 2.02% giving a gap of 15.7.  But the new data trend is 1.97% giving a gap of 14.6%, a reduction in the gap of 7%.    So the gap is reduced, but mostly because we were doing worse before the recession than we thought we were (if we use market sector data, which might be better measured the gap falls from 21.4 to 20.1 points, a fall of 6%).  (The source of this reduction is basically a fall in the pre-recession productivity growth of the service sector, from 2.1%pa to 1.9%pa).


So the revisions look like a better representation of the immediate timing and do reduce the gap by around 7%.  In our earlier work we found the things like structural change, utilisation and scrapping can account (using pre Blue Book 2015 data) for around 50% of the puzzle, see here.  Perhaps we are getting closer to a solution.

Update
Mea culpa and apologies to earlier readers of this post.  After kindly getting some comments, I checked my spreadsheet. The original ONS data contained an error in the market sector data  (the data were displaced by a cell) and I had done the revised gap calculations incorrectly: I had calculated the gap for the service sector.  The implied gap due to revisions should be about 7% less than the original one and not 20% as I estimated earlier.  I have therefore revised the data in the table and three paragraphs above.