Service Lives of R&D Assets: Background and Comparison of Approaches

Abstract

This paper draws together ONS research into the service lives of R&D assets which will be capitalised in the National Accounts from 2014. The characteristics of service lives are considered and results from two sources are presented; a new question on service lives implemented on the UK R&D surveys, and estimates derived from Patent administration data. These are compared and theoretical underpinnings, strengths, and limitations discussed.

Acknowledgements

The author would like to thank Christopher Steer, Helen Meaker, and Walter Mkandawire of the Office for National Statistics for their valued support and contributions.  Thanks are also due to John Jankowski of the US National Science Foundation for his input.

Introduction

The 2008 revision of the System of National Accounts (United Nations, 2009) requires Research and Development (R&D) expenditures to be treated as investment in knowledge assets.  This reflects the contribution of knowledge produced by R&D to production and output over multiple years.  It will provide insight into the role of knowledge in the economy, the allocation of resources to R&D, and will improve understanding of the relationship between knowledge and productivity.  Capitalisation also recognises R&D assets in the National Balance Sheet and therefore affects the measurement of National Wealth.

The change poses a variety of challenges for National Accountants.  In particular, much effort has been invested in developing methods for measuring the knowledge assets produced (also referred to as ‘R&D assets’).  Having established an appropriate valuation for R&D assets, it is then necessary to establish which sectors have economic ownership of them (see Steer & Ker (2013)) and to establish how long they are useful for; the subject of this work.

This publication is from a suite of three which provide a comprehensive overview of ONS research into R&D service lives.  This paper provides context for the research, compares the methods for the UK, and draws conclusions.  ‘Service Lives of R&D Assets: Questionnaire Approach’ (Ker, 2013b) describes the implementation of a new survey question on R&D lives and presents its results, while ‘Service Lives of R&D Assets: Patent Approach’ (Ker, 2013c) derives estimates from patent renewals data.  These related papers provide extensive detail on the approaches and estimates discussed in this paper and should be read alongside.

Service Lives

The service life of an asset is ‘the total period during which it remains in use, or ready to be used, in a productive process’ (OECD, 2001, p. 95).  Put simply, the service life of an asset is the number of years for which it is useful to its owner(s) in generating returns.  During its service life an asset may have more than one owner.

‘Asset life is understood here as an economic notion, and not as a physical or engineering notion of capital goods. This is important because it implies that assets can change over time simply due to economic considerations even if the asset remains physically unchanged’ (OECD, 2009, p. 106).

Typically, the end of an asset’s useful life may result from economic obsolescence, wear and tear, or loss through accidental damage. For R&D, the majority of these concepts have no meaning due to its intangible nature; knowledge does not suffer from ‘wear and tear’, nor can it be ‘accidentally damaged’ or subsequently ‘repaired’.  It has a theoretically infinite lifespan.  However, economists are concerned with the length of time over which knowledge directly contributes to economic output and to companies’ profits.  This productively useful ‘economic service life’ will be finite.

The finite nature of R&D service lives can be explained by ‘creative destruction’ where knowledge can be rendered obsolete and displaced by new discoveries (Bitzer & Stephan, 2007).  Peleg (2008) found evidence that use of knowledge may be gradually scaled back over time suggesting a gradual superseding of the R&D or a process of gradual spill-over where, over time, knowledge becomes available more widely until it is common knowledge and of no remaining specific value to its owner.  R&D ‘depreciates’ therefore because it makes a decreasing contribution to owners’ profits and to economic output over time.

There are clearly challenges here; much of the R&D enabling human flight still benefits companies and society as a whole for example.  However, the principles of flight are now common knowledge available to anyone.  Any on-going benefits are spillovers from the original R&D which are specifically excluded from the National Accounts; although new R&D which builds incrementally upon the original research (new wing designs for example) would be captured.

Why Service Lives Matter

Aside from the volume of R&D investment itself, service lives of R&D assets are the main determinant of the R&D stock used in analyses of productivity and National Wealth.  Service lives are key inputs to the Perpetual Inventory Method (PIM) used to compile asset stocks.  They are key parameters to survival functions which adjust for obsolescence; age-price (depreciation) functions which estimate the decline in assets’ values as they age; and age-efficiency functions which estimate the age-related decline in efficiency.  As a result:

 ‘the accuracy of capital stock estimates derived from a PIM is crucially dependent on service lives - i.e. on the length of time that assets are retained in the capital stock, whether in the stock of the original purchaser or in the stocks of producers who purchase them as second hand assets’ (OECD, 2009, p. 106). 

Shorter service lives give greater weight to more recent innovations in capital stock estimates (Australian Bureau of Statistics, 2009, p. 48), as (OECD, 2010, p. 33) clarifies;

‘Specifying a service life of 10 years rather than 5 years would make a huge difference to estimates of capital measures. Net capital stock would be approximately double, and with a typical scenario of strong growth, consumption of fixed capital would be appreciably smaller.’

The importance of reliable service lives for R&D assets is heightened by the use of sum-of-costs valuation for R&D (where the R&Ds value is set equal to the total costs incurred in its production).  In a fully integrated system, the contribution of existing R&D capital to the production of further R&D is included amongst the costs summed; the effect of this on output and hence Gross Domestic Product is partially determined by the R&D service life used.

European Task-Force Recommendation

Relatively few countries have produced R&D life estimates of any sort and the recommendation of the Second European Task-Force on R&D Capitalisation offered the pragmatic alternative of ten years as a default in the absence of empirical information:

Service Life estimates used in the calculations of R&D should be based on dedicated surveys or other relevant research information, including information of other countries with comparable market/industry characteristics.  In cases, where such information is not available, a single average Service Life of 10 years should be retained.

It is also recommended that the above mentioned Service Life estimates should be investigated regularly, e.g. every 10 years. (Eurostat, 2012)

However, there is no reason to assume that R&D service lives are equal between sectors in the same economy and much less between different countries with varied industrial mixes.  Nor is it likely that the ‘European average service life of R&D’ happens to be 10 years – the minimum of the range suggested by the OECD (2010, p. 62).    Furthermore, standardising the service life across countries removes one of the key determinants of the knowledge capital stock leaving the volume of R&D investment as the sole explanatory variable for countries’ differing economic performance.

Estimating R&D Service Lives

Each individual asset will have a unique service life; two identical machines need not last the same amount of time before wearing out.  In the case of R&D, no two knowledge assets are the same (by definition) and thus will have different useful lives.  However, it is not viable to collect data at the asset level and information is collected on typical lives for different types of assets (such as buildings, machines, vehicles, R&D, etc).

Though average lives receive the most attention, in general three descriptive statistics are needed for the estimation of the various functions used in the PIM;

‘The average or mean service life has to be distinguished from the maximum service life of a cohort of assets because the service lives of the same assets within a cohort are normally described by a retirement or mortality function which defines the distribution of asset retirements around the mean and between the minimum and maximum service lives’ (OECD, 2009, p. 106).

These values are likely to vary across:

• asset types – the average service live for R&D asset is likely to be less than for a building for example; similarly ‘Basic Research’ could generally have a useful life different to ‘Applied Research’ for example

• industries and products – aeronautical knowledge may generally be useful for longer than knowledge in software for example

• Institutional sectors - government-produced R&D may last longer on average than businesses’ R&D for example

• Time periods – the pace of technological change is likely to vary over time

A practical and affordable source is needed to provide sufficient detail to capture the essential variation in R&D lives over these characteristics.

Sources of Information

As R&D is not usually treated as capital investment in financial reporting standards, tax lives and information from company accounts commonly used for other assets are unavailable.  Expert opinions have provided useful information for some types of assets, while data from patent administration systems and survey questions are generally the most favoured sources for R&D service lives.

This research compares and contrasts two approaches for estimating service lives of UK R&D assets:

1. The paper ‘Service Lives of R&D Assets: Questionnaire Approach’ (Ker, 2013b) presents one of the largest surveys of business R&D service lives to date with 650 responses from a new question introduced to the UK Business Enterprise R&D Survey (BERD).  The wording was developed from a question used in the previous ‘Investment in Intangible Assets’ survey and was improved through two rounds of pilot testing.  It asked how long the business expects to benefit from a typical investment in ‘Basic Research’, ‘Applied Research’, and ‘Experimental Development’.  These different ‘types’ of R&D are defined in the Frascati Manual (OECD, 2002) and are also used in other questions on UK the R&D surveys.

The question was included on the BERD long form which is sent to the businesses which spend the most on R&D (making around 80 percent of all business R&D expenditure) and gained responses from 86 per cent of these firms.  The question was also included on all forms in Northern Ireland which adds some coverage of firms which spend less on R&D, though the response rate was much lower.  In total, the results directly represent over 66 per cent of UK business R&D (by 2011 current expenditure).

2. The paper ‘Service Lives of R&D Assets: Patent Approach’ (Ker, 2013c) uses information on annual patent renewals extracted from the UK Intellectual Property Office (IPO) and European Patent Office administration systems to estimate the useful lives of patents filed and/or approved between 1986 and 2010.  It estimates patent life by identifying patents which had ‘died’ (renewal payments had ceased or the patent expired) and estimating the ‘age at death’ as the difference between the year of death and the year the patent was applied for.  Estimates weighted by patent value are also produced.

This work adds to the existing literature in two ways; by using patent data that has been matched to businesses to break down estimates by industrial section, and by implementing Kaplan-Meier survival analysis techniques which reduce downward bias in the lives estimated by examining all patents including those which had not died by the end of the period covered by the data (2010).

The patent lives estimated can be used to draw conclusions about the service lives of R&D assets by assuming that patents represent R&D.  The paper explains and discusses this assumption in the context of available evidence.

This paper compares the two methods both theoretically and in the context of results for the UK, and looks at the implications of choosing one over the other.

Comparison of Approaches

Table 1 summarises the key features, assumptions, benefits, and limitations of the two different approaches.

The questionnaire approach more closely targets R&D assets and may be more forward looking and responsive to changes in the pace of technological change if regularly updated (eg. every five years).  However estimates are based on expectations rather than observations and there are costs involved with designing, testing, and implementing new survey questions.  The approach relies upon respondents’ ability to meaningfully average across their R&D projects and programmes and the sample design will affect how well results can be generalised to R&D more widely.

If the necessary data can be accessed, the patent approach offers the possibility of producing estimates without the need to wait for data to be gathered through a survey.  It provides a large number of direct observations on the lives of patents.  However, it may not always be the case that patent holders only renew when the patent continues to be of use; some companies, especially those holding large numbers of patents, may renew by default as the cost of annually reassessing each patent may outweigh the relatively small renewal fees.  Furthermore, with patent litigation continuing to increase amid growing awareness of the value of patents (PwC, 2012) patent holders may also choose to renew as a precautionary measure or because the actions of other businesses might give the patent value in future.

The choice of methods for analysis has a considerable affect on the downward bias in the results, especially with data covering shorter periods; analysing only patents which ‘died’ during the observation period biases estimates downward considerably compared to estimates based on Kaplan-Meier Survival analysis techniques which uses all information available – both on patents which died during the period and on those which existed but did not die before the end of the period covered by the data.

More fundamentally, using patent lives as an indicator of R&D lives requires the assumptions that all patents embody R&D (as opposed to resulting from other processes).  There is a logical relationship between R&D and the production of patents. However, the evidence suggests that in practice this link is not firm and will vary between industries.  Furthermore, generalisation effectively requires the assumption that all R&D is patented (so that patents represent all R&D). Meanwhile as the analysis is ‘backwards looking’ it requires that the patent lives derived for the period 1986 – 2010 can be applied to other periods, foregoing the possibility that lives may be altered by changes in the pace of technological advance.

Table 1: Comparison of methods for estimating R&D service lives

Results

The analysis showed that both the questionnaire and patent data were highly positively skewed, even after the optimal Box-Cox transformation (Box & Cox, 1964) (Osborne, 2010) had been applied.  Kitchen (2009) explained that in the presence of only one arbitrarily large outlier the mean becomes arbitrarily large; by contrast, the median will not breakdown as long as only a minority of observations are corrupted.  As such, in the presence of skewed data such as this, the median is likely to provide a better estimate of central location.  Table 2 presents median service lives derived using the different approaches.

Table 2: Comparison of median lives from questionnaire and patent sources

Table notes:

1. Source: ONS BERD 2011 (2012) and IPO Patent data
2. '- not published: low sample size
3. * median undefined (survival probability has not declined below 50%)
4. ** for patent estimates this category is an aggregation of individual industrial sections provided for comparison purposes

It is immediately apparent that the questionnaire data was only sufficient to sustain disaggregation to a number of the industrial sections although disaggregation is possible for a number of key R&D performing industries including manufacturing.  By contrast the patent data facilitates the complete breakdown, though sample sizes are very small for ‘Public administration and defence’ and ‘Activities of households as employers’ and so these estimates should be treated with caution.  The Kaplan-Meier survival method benefits from larger samples as it makes use of the information on patents observed to exist but which did not die before the end of the observation period in 2010.

Weighting questionnaire responses by firms’ average share in total business expenditure on each R&D type makes the overall median life longer.  The effect is similar at the industrial section level, although the median life actually falls from five to four years in software.

Comparing the different sources, the questionnaire estimates are in the same approximate “ball-park” as the unweighted patent medians.  This suggests that unweighted patent estimates may provide similar results to questionnaires, at least at the aggregate level; in some industries the spread between patent and questionnaire estimates is considerable (eg. three years in ‘Information and communications excluding software).  This might be because only longer-lived R&D is patented in this industry and so the survey results better reflect the average life of all R&D.  The value-weighted patent estimates are considerably higher as they place greater weight on longer-lasting patents.  This may lead to upward bias of estimates, although selecting the median will mitigate this effect somewhat compared to the mean.  Even so, it seems preferable to also take the ‘average of averages’ (the mean of the weighted and unweighted median estimates) to moderate the bias introduced.

The Kaplan-Meier survival estimates vary in their proximity to the other various patent estimates, in some cases falling towards the lower end of the range between the unweighted median patent life and in other cases falling in line with the value-weighted estimates.  This may result from industrial sections generally holding the long-lived patents which are given more weight by taking values into account.

There is considerable variation across industry sections, suggesting that this dimension is important.  The estimates generally agree that service lives are shorter in software and longer in the R&D sector, which consists of both specialist research firms and research branches of larger firms which have been separately identified and classified to this sector.  As a result, the R&D performed by these businesses is likely to be highly varied.

Table 3 compares mean lives estimated from the different sources and approaches.  The picture here is generally similar to that given by the medians in Table 2, although the questionnaire means and the unweighted patent mean life are even closer, lying within 1.3 years of each other.  There is still considerable variation across industries.

Table 3: Comparison of mean lives from questionnaire and patent sources

Table notes:

1. Source: ONS BERD 2011 (2012) and IPO patent data
2. '- not published: low sample size
3. * median undefined (survival probability has not declined below 50%)
4. ** for patent estimates this category is an aggregation of individual indistrial sections provided for comparison purposes

Comparing the survival profiles of R&D/patent lives from each source shows that survival from the questionnaire is quite similar to the profile given by unweighted patent estimates.  By comparison, using the unweighted patent estimates would under estimate obsolescence in the first 7 years before falling approximately into line with the questionnaire based profiles.  The average patent profile moderates the upward bias in the weighted profile and lies in the middle.

The figure also illustrates the impact of using patent data, which is (effectively) capped at a maximum life of around 21 years.  The survey responses suggest that around four per cent of R&D survives beyond 21 years but, though this proportion is small, the number of years further life is considerable at up to 35 years.

Figure 1: Comparison of R&D Survival Profiles

Notes:

1. Source: ONS BERD 2011 (2012) and IPO patent data

Download chart

• XLS format
  (19.5 Kb)

Comparison to Other Studies

Peleg (2008) reviewed depreciation rates and service lives from ten different studies covering differing geographical areas and time periods.  They used various econometric models including approaches based on patent renewals. Findings showed (or implied through depreciation rates) average service lives ranging between 3.8 years and 18.2 years.  Results varied considerably between industries and by R&D field (eg. ‘Chemicals’, ‘Electrical Equipment’, etc.), emphasising the need to consider these differences.

The majority of estimates lie within this range or slightly above it, although this is as much a product of the width of this range in the first place.  These results also re-confirm the presence of considerable variation between industrial sections.

The average lives of R&D from UK questionnaires are all greater than that found by the United States National Science Foundation (NSF) which also conducted a large-scale survey (covering over 6,000 firms) in 2012.  The question used asked about the life of products embodying R&D (that is, over how many years the business sold a product which made use of R&D).  Although many firms did not provide responses, the survey gained almost 1,000 answers - making it the largest survey to date.  The results gave an average of 5.46 years and showed variation between the 25 industries covered (Li, 2012).  The most comparable estimate from the UK survey, the unweighted mean service life, is 8.2 years.

The unweighted mean patent life is 9.5 years in the UK, a little lower than the 11 years found by the Australian Bureau of Statistics (2009).  Meanwhile, the unweighted median patent life of eight years is greater than the seven years found in The Netherlands (Tanriseven, van Rooijen-Horsten, & de Haan, 2010).  The value-weighted median patent life was longer in the UK than The Netherlands at 20 years compared to 18 years in The Netherlands; influenced by comparatively higher values of UK patents.  As a result, the UK ‘average of averages’ was also longer at 14 years compared to 12.5 years.

Comparing the unweighted survival profile to the Australian Bureau of Statistics and that implied by the survival probabilities provided by Tanriseven, van Rooijen-Horsten, & de Haan, all share a very similar smooth declining pattern with little attrition over the first four years and an acceleration between 19 and 21 years.

Looking at the top level results in the context of the EU default of 10 years and the 10-20 year range suggested by the OECD (2010, p. 62) in Table 4 it is clear that while many of the estimates lie between 10 and 20 years, only in one case is the life estimated equal to the suggested default of 10 years.  The median, weighted to reflect the relative shares of different respondents in total R&D output is also the most appealing questionnaire-based estimate.  However, this could largely result from the ‘clustering’ at focal numbers observed in businesses’ responses.  Furthermore, the breadth of alternative estimates available (the confidence intervals of which show that they are significantly different from 10) suggests that this is somewhat of a coincidence in the case of the UK; estimates vary by industry and are likely to vary across countries too.  The 10-20 year range is a useful guide but researchers should attach less significance to being within or outside this range than to the robustness of the methods chosen.

Table 4: Comparison of results by source

Table notes:

1. Source: ONS BERD 2011 (2012) and IPO patent data

Impact of Alternative Estimates

This research has provided considerable number of different estimates of R&D service life for the UK, they range from 6.0 to 20.0 years.  Aside from the volume of investment in R&D itself, the choice of average service life is a key determinant of the size of the R&D stock, which has implications for analysis of productivity and National Wealth.

Figure 2 presents a stylised example illustrating the impact of using these different service life estimates in the Perpetual Inventory Method used to compile capital stocks in the National Accounts.  The ‘double-declining’ geometric rate used here offers a convenient approximation of the integrated survival/age-price/age-efficiency profile (OECD, 2009, p. 97) accounting for the retirement of assets and falling values (depreciation) and efficiency.   By construction, annual R&D investment is set at £20 to exactly offset this adjustment when the life is 10 years so that the R&D stock remains constant over time, with the initial stock set at £100.  The vertical axis can therefore be interpreted as an index and shows that shorter lives then imply a shrinking R&D stock, while the longer lives imply different degrees of R&D stock growth.

Figure 2: Illustrative Impact of Different Service Lives on R&D Stocks

Notes:

1. Source: ONS BERD 2011 (2012) and IPO patent data

Download chart

• XLS format
  (20.5 Kb)

In this example, over 20 years, a service life of 6 years (rather than 10) is consistent with a 40 per cent decline in the stock of R&D and therefore a move away from knowledge driven growth.  By contrast, an R&D service life of 20 years would contribute to a rapidly expanding national stock of knowledge, with the stock continuing to grow beyond 20 years and doubling in 40 years.  This is an extremely stylised example but it serves to demonstrate that in the real world the stock of R&D capital depends upon both the volume of investment in R&D and the period over which the resulting knowledge is useful.

Differentials in service lives will also therefore also be a determinant of international variation in R&D stocks and so the effect of R&D in different economies; as such the choice of service life is non-trivial and it is desirable to investigate the specific details in different countries.

Evaluation and Conclusions

This research has outlined two different approaches to estimating R&D service lives:

• survey questions gather expected service lives based on the assumption that respondents can correctly interpret the question and can provide consistent and meaningful answers

• patents provide observations on renewals based on the assumption that agents will only renew as long as the benefits outweigh the costs

Each offers different benefits and limitations; for example survey questions require money and time for design and implementation, and can also add to respondent burden through the different, more abstract approach used.  Patent research may benefit from lower cost and time requirements, and offers a large amount of observed information.  Industry analysis is also possible, though patents must first be matched to business information.

Both approaches require assumptions about generalisation of results to other R&D.  However, while results from the BERD can claim to be representative of 80 per cent of UK business R&D, the strength of the link between R&D and patents is unclear; there are various reasons why business (and other types of organisation) may choose not to patent R&D including the costs involved, lags between application and approval, and the renewal limit.  Additionally, low renewal fees and precautionary motives may lead firms to renew by default rather than only when there is clear value to the patent, a fundamental assumption of the approach.

Generalisation over time is also an issue, and while survey questions might easily be re-run every five or ten years to collect benchmarks, patent data is backwards looking and requires waiting to observe death. The Kaplan-Meier survival analysis techniques demonstrated address this to some extent.

Both methodologies classify based on the industry producing the R&D (or patent) but R&D may be sold or otherwise transferred across industries.   However, the general nature of the question posed should mean that survey results are fairly representative of R&D invested in and used by the different industrial groupings (whether it is created on own account or bought in).  The more specific nature of patent analysis, coupled with the potential for bias in the ‘fuzzy matching’ of patents to businesses does not afford this approach the same level of resilience.

On the basis of this full and thorough analysis of Business R&D service lives results, the ONS proposes to adopt weighted median service lives estimated from the survey data.  While the level of industry disaggregation is less satisfactory, this approach offers the advantage of being clearly linked to R&D. This It is intended that these will be supplemented by results covering the Government and Private Non-Profit sectors which will be published separately in 2013.

Correct understanding of the impact of R&D on the economy and on productivity can only be gained by using data and methods that produce results of sufficient detail and accuracy.  One key parameter determining the size of the R&D stock and thus the economic impact of R&D capitalisation is the service life adopted.  Estimating R&D service lives is challenging, but this research provides a detailed comparison of two popular methods.  It is hoped that this will provide a useful foundation for our colleagues internationally.

References

Australian Bureau of Statistics. (2009). Implementation of New International Statistical Standards. Canberra: Australian Bureau of Statistics.

Bitzer, J., & Stephan, A. (2007). A Schumpeter-Inspired Approach to the Construction of R&D Capital Stocks. Applied Economics , 153-167.

Box, G., & Cox, D. (1964). An Analysis of Transformations. Journal of the Royal Statistical Society , 211-252.

Eurostat. (2012). Second Task Force on the Capitalisation of Research and Development in National Accounts: Final Report. Luxembourg: European Commission, Eurostat.

Gambardella, A., Giuri, P., & Mariani, M. (2005). The Value of European Patents.

Ker, D. (2013c). Service Lives of R&D Assets: Patent Approach. Newport: Office for National Statistics.

Ker, D. (2013b). Service Lives of R&D Assets: Questionnaire Approach. Newport: Office for National Statistics.

Kitchen, C. (2009). Nonparametric vs Parametric Tests of Location in Biomedical Research. Los Angeles: UCLA School of Public Health.

Li, W. (2012). Summary of the Survey on the Service Life of R&D Assets. Bureau of Economic Analysis.

OECD. (2002). Frascati Manual. Paris: OECD Publications.

OECD. (2010). Handbook on Deriving Capital Measures of Intellectual Property Products. Paris: OECD Publishing.

OECD. (2001). Measuring Capital (first edition). Paris: OECD Publications Service.

OECD. (2009). Measuring Capital. Paris: OECD Publishing.

Osborne, J. W. (2010). Improving your data transformations: applying the Box-Cox transformation. Practical Assessment, Research & Evaluation , 1-9.

Peleg, S. (2008). Service Lives of Research and Development. New York: United Nations.

PwC. (2012). 2012 Patent Litigation Study. Delaware: PwC.

Schumpeter, J. A. (1943). Capitalism, Socialism, and Democracy. Taylor & Francis e-Library.

Steer, C., & Ker, D. (2013). Ownership of R&D Assets. Newport: Office for National Statistics.

Tanriseven, M., van Rooijen-Horsten, M., & de Haan, M. (2010). Capitalisation of R&D: Preparing the new ESA. The Hague: Statistics Netherlands.

United Nations. (2009). System of National Accounts 2008. New York: United Nations.

Whittard, D., Franklin, M., Stam, P., & Clayton, T. (2009). Testing an Extended R&D survey: Interviews with Firms on Innovation Investment and Depreciation. London: National Endowment for Science Technology and the Arts.

Background notes

1. Details of the policy governing the release of new data are available by visiting www.statisticsauthority.gov.uk/assessment/code-of-practice/index.html or from the Media Relations Office email: media.relations@ons.gsi.gov.uk