This article uses published estimates from the Annual Business Survey (ABS) to provide more detailed information on recent trends in labour productivity by industry than those available from other sources. The detailed results reveal a great deal of volatility in labour productivity over time and large differences in productivity levels across different industries. Some of the volatility is likely to be genuine, reflecting cyclical and structural dynamics affecting particular industries; and some may reflect greater statistical imprecision as industry definitions become narrower and sample sizes smaller. The article finds that productivity levels are fairly well correlated with labour remuneration, in line with economic theory that wages reflect marginal productivity and that skills vary across industries. No ABS data are available on capital inputs, but it is highly likely that differences in capital intensity account for much of the residual differences in labour productivity between industries. An innovation in this article is the use of 'double deflation' to construct real (inflation adjusted) measures of value added from the ABS nominal estimates. This method is conceptually correct for productivity purposes, but its provisional application in this article reveals very large differences between labour productivity derived from the ABS and according to the ONS National Accounts.
This article uses published estimates from the Annual Business Survey (ABS) to provide more detailed information on recent trends in labour productivity than those available in the Office for National Statistics (ONS) quarterly Labour Productivity release.
This is the latest in an occasional series of such articles. There are a number of innovations from the last article published in November 2011 (Acheson, 2011). First, this article focuses exclusively on the period 2008-12, being the period over which detailed ABS estimates are available on a consistent industry taxonomy, following the switch to SIC07 in 2008.
Second, this article is the first to analyse labour productivity by 'double deflation' — that is, by constructing real measures of industry level gross value added (GVA) from separately deflated series of gross output and intermediate consumption. This method has been facilitated by recent publication of Input-Output Analytical Tables for 2010 (ONS, 2014b). This is the first application of double deflation by ONS since 2002 and is necessarily provisional, since ONS does not currently produce balanced National Accounts (NA) in real terms. However, double deflation is conceptually preferable for the purpose of analysing productivity and the results in this article represent an important step forward.
Third, although the published ABS results contain detailed employment estimates (sourced from the Business Register and Employment Survey (BRES), which utilises a similar sampling frame to ABS), here, we prefer to use estimates of hours worked derived from the ONS labour productivity systems. There are a number of reasons for this:
Hours worked provide a more comprehensive measure of labour input than headcount- or jobs-based measures since they implicitly take account of changes in average hours worked due, for example, to changes in proportions of full- and part-time employment.
Our measures of hours worked are benchmarked at the whole economy level to estimates from the Labour Force Survey (LFS), which is ONS's preferred source of overall labour engagement in economic activity.
There is a discontinuity in the BRES estimates between 2009 and 2010, such that BRES estimates to 2009 contain an element of double counting, the size of which is unknown.
It is important to be aware of a number of differences between the estimates in this article and aggregate measures of GVA, labour inputs and productivity, including:
ABS does not cover the entire industry spectrum and has only partial coverage in some areas such as health and education. In this article we restrict our attention to those industries where ABS coverage is complete.
ABS does not cover unincorporated business below the VAT and PAYE thresholds. In practice this means that ABS covers only a small proportion of economic activity from self-employment.
Although ABS is a key data source for current price supply-use balancing, National Accounts output measures are subject to balancing and coherence adjustments that may result in reduced consistency between the ABS source data and published GVA in current price terms.
ABS does not provide any information on prices, or any information on real (inflation-adjusted) economic activity.
ABS estimates are not revised after publication of final estimates 30 months after the end of the reference year. By contrast, National Accounts estimates can be revised much further back, for example, reflecting methodological changes.
Current price supply-use balancing has not been carried out for 2012 and ABS estimates are provisional for this year.
More information on differences between ABS and the National Accounts framework is available in the ABS Technical Report (ONS, 2012) and in Ayoubkhani (2014).
This article is arranged as follows. The following section provides an overview of the detailed results at 2-digit, 3-digit and 4-digit level (detailed results are available in the Reference Tables (1.13 Mb Excel sheet) component of this release). The next section provides some context and comparisons with National Accounts measures at the broad industry level. Following this there is a short discussion of next steps and future development priorities. Some notes on data sources and methodology are provided as an appendix.
Detailed results are available in the Reference Tables (1.13 Mb Excel sheet) accompanying this article. Results are presented in terms of current price GVA per hour (£/hour) and indices of double-deflated GVA per hour based on 2008=100, and for three different industry disaggregations. In general, the more disaggregated the series presented in this release, the more volatile they are. This reflects both the real world volatility of narrowly defined industries’ output and labour input, and the smaller sample of survey respondents for such industries. As well as volatility in individual series, there is a great deal of variation in trends across similar industries at the more detailed levels. Additionally ABS estimates are subject to disclosure control at finer levels of disaggregation. In some cases missing ABS estimates have been replaced by values derived by interpolation, by residual or other means. This is not always feasible, for example, where estimates have been suppressed for all years or for all or most of the components of a broader aggregate.
The current price productivity series in the reference tables are comparable across industries but do not distinguish between price and volume movements over time. Thus they should not be used to compare year on year productivity movements of a single industry, but should be used to analyse productivity level differences between industries at a point in time. For this reason, the current price tables should be read vertically.
The constant price productivity results in the reference tables are presented as indices (indexed to 2008=100), year-on-year percentage changes, and an average growth rate over the period 2008-2012. These results should be read horizontally, and show productivity movements of a single industry over time. Comparisons of constant price productivity movements of different industries can show how different industries fared over the period, but say nothing about the absolute levels of labour productivity between industries.
|Average Rank||2-digit SIC07 industry||Industry description||GVA/hour, 2012 (£)|
|1||6||Extraction of crude petroleum and natural gas||665|
|2||12||Manufacture of tobacco products||196|
|3||36||Water collection, treatment and supply||117|
|4||60||Programming and broadcasting activities||99|
|5||35||Electricity, gas, steam and air conditioning supply||106|
|68||14||Manufacture of wearing apparel||19|
|69||80||Security and investigation activities||19|
|70||56||Food and beverage service activities||14|
|71||81||Services to buildings and landscapes activities||12|
|72||94||Activities of membership organisations||10|
Table 1 provides an indication of the distribution of current price GVA per hour at the 2-digit industry level, based on the average ranking over 2008-12. As for all tables in this section, a full set of results is available in the Reference Tables (1.13 Mb Excel sheet) accompanying this article.
At the 2-digit level current price productivity estimates are available for 72 of the 74 industries for which the ABS provides full coverage1. In general the current price productivity rankings conform to a priori expectations: capital intensive production industries display the highest levels of GVA per hour, and low capital-intensive, low-skilled service industries feature at the bottom of the productivity distribution.
|Average Rank||2-digit SIC07 industry||Industry description||Average YoY %, 2008-12|
|1||13||Manufacture of textiles||11.0%|
|2||33||Repair and installation of machinery and equipment||10.6%|
|3||39||Remediation activities and other waste management services||10.0%|
|4||11||Manufacture of beverages||9.9%|
|5||16||Manufacture of wood and of products of wood and cork, except furniture; manufacture of articles of straw and plaiting materials||8.1%|
|68||92||Gambling and betting activities||-7.7%|
|70||60||Programming and broadcasting activities||-8.5%|
|71||06||Extraction of crude petroleum and natural gas||-9.3%|
|72||21||Manufacture of basic pharmaceutical products and pharmaceutical preparations||-10.6%|
Table 2 provides equivalent rankings in terms of average annual growth of real (double-deflated) GVA per hour over the period 2008-12. It is important to note that in some cases these average growth rates disguise very large movements in real GVA per hour from year to year. Such movements are not confined to the steep economic downturn in 2008-09, although the downturn is likely to have been a contributory factor in the overall degree of volatility over this period.
Over the period 2008-12, average productivity growth was fastest among a range of medium-sized manufacturing industries together with industry 39, which is very small and for which the ABS estimates are especially volatile. The highest ranking service industry is 96 (other personal services) where real GVA per hour grew at an average rate of 6.9% per year over 2008-12.
The lowest ranking 2-digit industries in terms of productivity growth form an eclectic group from across the industry spectrum. The heterogeneity of performance across manufacturing is illustrated by the low rank of industry 21 (pharmaceuticals), and the poor productivity performance of this industry is also echoed in ONS's macro productivity estimates. According to the latest quarterly Labour Productivity release, output per hour in the combined 2-digit industries 20 and 21 (chemicals and pharmaceuticals) fell by 17% between 2008 and 2012. It will be noted that industry 06 (oil and gas extraction) has experienced the second worst productivity growth over the period, despite it being top of the current price rankings in Table 1. This provides some informal support for the argument that aggregate productivity performance has been subject to compositional effects insofar as some of the weakest industries in terms of productivity growth are among the highest in terms of productivity levels. Overall there is a negative correlation of -0.25 between the two sets of rankings at the 2-digit level.
|Average Rank||3-digit SIC07 industry||Industry description||GVA/hour, 2012 (£)|
|1||49.5||Transport via pipeline||197|
|2||35.1||Electric power generation, transmission and distribution||123|
|3||35.2||Manufacture of gas; distribution of gaseous fuels through mains||72|
|4||61.9||Other telecommunications activities||*|
|5||9.1||Support activities for petroleum and natural gas extraction||*|
|178||56.1||Restaurants and mobile food service activities||14|
|179||56.3||Beverage serving activities||14|
|181||94.9||Activities of other membership organisations||4|
|182||70.1||Activities of head offices||0|
Table 3 repeats the rankings set out in Table 1 at the 3-digit level. At this level the detailed ABS results together with some imputations support current price rankings for 182 of the 215 3-digit industries for which the ABS provides full coverage. Missing industries due to suppression of estimates for disclosure purposes can be seen in the Reference Tables (1.13 Mb Excel sheet) and include sub-components of industry 06 (that is, the split between oil and gas), some manufacturing industries, some construction industries and the components of industry 60, which featured in the bottom section of Table 2. Also, industries 12 (tobacco) and 36 (water supply) do not feature in Table 3 despite their appearance in the top section of Table 1 because there are no 3-digit breakdowns of these industries.
Three of the top five industries are production industries, but interestingly none is in manufacturing. Assuming that GVA per hour was roughly equal in 06.1 (crude petroleum) and 06.2 (natural gas), the highest ranking 3-digit industry 49.5 (transport via pipeline) would rank third overall, behind these extractive industries but fractionally ahead of the 2-digit industry 12.
There are fewer omissions from the 3-digit industry taxonomy at the bottom of the current price GVA per hour rankings, and hence more consistency between Tables 1 and 3. All of the bottom five places are occupied by service industries, including 3-digit components of industry 94 (membership organisations), industry 81 (building services) and industry 56 (food and beverage services), all of which also feature at the bottom of Table 1. Last place however is reserved for industry 70.1 (activities of head offices). This is something of an anomaly insofar as the ABS results show significant numbers of enterprises and considerable employment for this industry, but hardly any value-added. This may suggest issues with how ABS survey respondents apportion value added across the value chain for integrated businesses with head offices separate from their production facilities. We return to the issue of value chains below.
|Average Rank||3-digit SIC07 industry||Industry description||Average YoY %, 2008-12|
|1||51.2||Freight air transport and space transport||61.6%|
|2||47.4||Retail sale of information and communication equipment in specialised stores||24.2%|
|3||46.2||Wholesale of agricultural raw materials and live animals||21.0%|
|5||94.9||Activities of other membership organisations||17.0%|
|170||68.1||Buying and selling of own real estate||-24.7%|
|171||81.1||Combined facilities support activities||-26.7%|
|172||78.3||Other human resources provision||-32.9%|
|173||80.2||Security systems service activities||-35.4%|
|174||70.1||Activities of head offices||-40.9%|
As shown in Table 4, it is possible to compute average productivity growth (double deflated) for 174 of the 215 3-digit industries, slightly fewer than those for which current price productivity levels are available due to missing values in 2008 or 2012. Comparing Table 4 with the 2-digit rankings in Table 2 it is apparent that there is much more variability in productivity performance at the 3-digit level. Moreover, none of the industries featured either at the top or the bottom of Table 4 are sub-industries of those featured in Table 2. And perhaps surprisingly, given that at the macro level productivity is more volatile in the production industries than in services, all of the five highest and lowest ranking industries are services industries.
Comparing Tables 3 and 4, industry 70.1 props up both tables but on the other hand industry 94.9 features in the top panel of Table 4 despite being second from bottom in Table 3. As noted above, this is another industry with very low estimates of value added that are erratic from year to year.
The correlation between rankings in terms of levels and growth rates is weakly positive at the 3-digit level, suggesting that industries which rank higher in terms of current price GVA per hour have also tended to rank higher in terms of growth of real GVA per hour.
|Average Rank||4-digit SIC07 industry||Industry description||GVA/hour, 2012 (£)|
|1||59.13||Motion picture, video and television programme distribution activities||*|
|2||24.41||Precious metals production||*|
|3||11.01||Distilling, rectifying and blending of spirits||133|
|4||77.32||Renting and leasing of construction and civil engineering machinery and equipment||78|
|5||35.13||Distribution of electricity||*|
|334||91.01||Library and archive activities||-3|
|335||94.91||Activities of religious organisations||-5|
|336||91.04||Botanical and zoological gardens and nature reserve activities||-1|
|338||59.11||Motion picture, video and television programme production activities||*|
Results in terms of current price GVA are available or can be imputed for 338 of the 411 4-digit industries for which the ABS has full coverage (Table 5). Missing industries are shown in the Reference Tables (1.13 Mb Excel sheet) . Comparisons between the 4-digit and other levels of the SIC07 industry taxonomy should bear in mind that a significant number of 3-digit industries are not broken down into 4-digit headings, while some 2-digit industries such as 62 (computer programming) are parents of 4-digit breakdowns but not of 3-digit breakdowns.
There are a number of interesting features of Table 5, perhaps the most interesting being that industry 59.13 is at the very top of the rankings and industry 59.11 at the very bottom. These 4-digit components represent different stages of the value chain in industry 59.1 (Motion picture, video and television programme activities) but the respective rankings could be considered to be counter-intuitive. ABS estimates imply that 'distribution activities' is at the top of the rankings and 'production activities' is at the bottom, whereas one might have thought it would be the other way round. On the other hand, industry 35.13 (distribution of electricity) also features in the top panel, but in this case the related production activity (35.11 - production of electricity) also ranks highly, in ninth place.
Three of the bottom five industries are components of the 2-digit industry 91, which is one of the industries suppressed from Tables 1 and 2 because aggregate GVA oscillates around zero.
|Average Rank||4-digit SIC07 industry||Industry description||Average YoY %, 2008-12|
|1||94.92||Activities of political organisations||87.2%|
|2||42.91||Construction of water projects||43.3%|
|3||13.95||Manufacture of non-wovens and articles made from non-wovens, except apparel||43.2%|
|4||43.13||Test drilling and boring||43.0%|
|5||24.54||Casting of other non-ferrous metals||38.6%|
|260||28.11||Manufacture of engines and turbines, except aircraft, vehicle and cycle engines||-21.9%|
|261||46.16||Agents involved in the sale of textiles, clothing, fur, footwear and leather goods||-25.2%|
|262||24.43||Lead, zinc, and tin production||-25.6%|
|263||33.19||Repair of other equipment||-26.1%|
|264||93.19||Other sports activities||-37.3%|
As shown in Table 6, growth rates of constant price GVA per head can be derived for 264 of the 411 ABS 4-digit component industries. There is an interesting contrast between two closely located industries, 24.43 (lead, zinc and tin production, ranked 262nd) and 24.54 (casting of non-ferrous metals, ranked fifth), and a further contrast between these rankings and the respective rankings of their 3-digit parent industries, 24.4 (ranked 160th out of 174) and 24.5 (ranked 74th).
Several of the entries in Table 6 are characterised by very small and volatile ABS estimates on either the output side or the labour input side, leading to large movements in productivity. A partial exception is industry 28.11 (manufacture of engines and turbines etc) which is fairly substantial (turnover of £8.4bn in 2012) but where the employment estimate in 2008, but not the GVA estimate, is much lower than in later years.
Using ABS reported employment costs to estimate employment costs per hour, we can see that at the 2-digit level, in general relatively high employment costs are well correlated with higher productivity levels. The relationship between employment costs and the productivity level of 2-digit industries is shown in Figure 1. The results conform well to expectations: assuming that factors receive their marginal products, hours worked which are remunerated relatively highly would contribute more to production, and have more value added attributed to them. At the top and bottom levels of the distribution this is highly pronounced.
Three of the five industries with the highest employment costs are in the mining and quarrying sector, in which productivity tends to be high owing to high levels of skill and capital intensity. The industry with the highest employment costs per hour overall is industry 06 (extraction of oil and gas), at £80.70 in 2012. This industry (not shown in Figure 1) also has the highest productivity level by an order of magnitude.
In general, the lower end of the distribution contains low-skilled and low capital intensive manufacturing and service industries such as accommodation, manufacturing of wearing apparel and two of the three 2-digit components of retailing and wholesale (Section G). Again, the bottom end of the distribution in terms of employment costs is highly correlated with the distribution in terms of productivity levels.
Across the middle of the distribution, most industries employment costs and productivity level are well correlated, but there are some notable exceptions. It is likely that some industries could be capital intensive but would not require high levels of labour skill. If factors receive their marginal products then this would materialise as a relatively high return on capital and a relatively low return on labour, and as a consequence relatively low employment costs. Some examples of such industries are 12 (manufacture of tobacco products), 77 (rental and leasing activities) and 37 (sewerage).
Examples can also be found where the opposite is true; industries which require highly skilled labour but are not capital intensive, where employment costs are relatively high. Industry 72 (Scientific research and development) illustrates this, ranking eighth in terms of employment costs and 48th in terms of productivity at the two-digit level. A similar scenario is seen in industry 70 (activities of head offices; management consultancy services), which ranks 16th in terms of employment costs and 28th in terms of productivity.
Retail and wholesaling (section G) comprises a large weight in both employment and output, but in level terms, many of its more disaggregated components are relatively low in productivity. At the 4-digit level, industry 47.11 (Retail sale in non-specialised stores with food, beverages or tobacco predominating), which includes supermarkets, is the largest in terms of employment and output, but in productivity level terms ranks 297 out of 338 industries. In growth terms, industry 47.11 performs relatively well over the downturn, averaging 0% growth and a rank of 162 out of 264 4-digit industries.
In level terms, the two highest productivity retail industries at the 4-digit level are 47.77 (retail sale of watches and jewellery in specialised stores) and 47.91 (retail sale via mail order houses or via internet) respectively. The more interesting of these is 47.91, as it is larger in terms of output and employment, and allows a comparison between traditional retailing and e-commerce. Industry 47.91 ranks 114 out of 338 4-digit industries in level terms, putting it quite above the average for retailing. This could reflect a relatively lower volume of labour input required for online retailing, where labour input would be mostly confined to a distribution network, compared to traditional retailing, where front line sales staff would also be required. The weight of labour in the factors of production would therefore be relatively lower for online retailing. Complementing this, the capital assets used in online retailing are likely to be more concentrated upon assets with shorter lives and higher rates of return such as software than less productive assets such as buildings. It is reasonable to expect that the composition of capital used in the traditional retailing sector would be more concentrated upon less productive assets.
However, in terms of productivity growth, the performance of industry 47.91 is poorer than the retail industry in aggregate, averaging -2% over the period 2008-2012, and this masks considerable year on year volatility. It is also outperformed by industry 47.11, which averages zero growth over the period. In terms of rankings of growth, industry 47.91 is 187th out of 264 4-digit industries for which results were available, and this is poorer than the other main components of retailing, such as industry 47.11 (ranked 162nd).
|B-J, L-N, RS||All ABS (excl partial coverage)|
|B||Mining and quarrying|
|D||Electricity, gas, steam and air conditioning supply|
|E||Water supply, sewerage, waste management, and remediation activities|
|G||Wholesale and retail trade; repair of motor vehicles and motorcycles|
|H||Transport and Storage|
|I||Accommodation and food service activities|
|J||Information and communication|
|L||Real estate activities|
|M||Professional, scientific and technical activities|
|N||Administrative and support service activities|
|R||Arts, entertainment and recreation|
|S||Other service activities|
Table 7 provides a list of section level industries for which results are reported in this article. In addition to the sections shown in Table 7, ABS provides results for part of sections A (agriculture, forestry and fishing), P (education) and Q (health), but since we have no reliable method of allocating hours worked to the parts of these industries covered by ABS, or of making comparisons with National Accounts estimates, we exclude these sections from the analysis in this article. In addition, ABS covers parts of section K (finance and insurance) but estimates for this section were removed from the ABS publication to allow for more quality assessment work to take place. In any case, the conventional measurement of turnover (sales) and GVA (sales minus purchases) do not apply to activities such as banking intermediation and parts of the insurance sector. Finally, ABS does not cover sections O (public administration), T (activities of households as employers) and U (extra-territorial organisations).
|B-J, L-N, RS||29||28||29||31||30|
Table 8 provides some context for the more detailed results in the Reference Tables (1.13 Mb Excel sheet) , showing ABS estimates of current price GVA per hour worked at the sectional level. Even at this broad section level there is striking variation in levels of GVA per hour worked, with sections B and D greater by an order of magnitude than sections I and S. This reflects much higher ratios of capital to labour in mining and quarrying (where output is dominated by oil and gas extraction) and the utilities, compared with labour intensive service activities such as accommodation and food services.
|% differences from ABS||2008||2009||2010||2011||2012||Average|
|B-J, L-N, RS||15%||13%||15%||14%||15%||14%|
Table 9 compares hours attributed to the ABS sample frame with aggregate hours by industry taken from the ONS labour productivity system. Since the ABS sample frame does not include unincorporated businesses below the VAT and PAYE thresholds, it is not surprising that the National Accounts estimates are generally higher. Over the period 2008-12, the National Accounts measure of hours worked in the industries fully covered by ABS is some 14% higher than the estimated hours worked in firms covered by the ABS sample frame.
For the period 2008-12 as a whole we estimate that the ABS sample frame captures about 21% of the self employed working in the aggregate matched industries. Over this period self employment in these industries averaged some 3.3 million, implying that the economic activity of some 2.6 million self-employed workers are not captured within the ABS sample frame for the aggregate matched industries.
Table 9 also shows that the proportion of hours worked attributed to the ABS sample frame is lowest relative to National Accounts estimates in section F (construction), R (arts and entertainment) and S (other services), all of which are industries with high proportions of self employment. The ABS share of hours is highest in section L (real estate, where ABS hours worked are actually higher than National Accounts estimates in 2011 and 2012). The relatively high coverage of ABS for section G (retail and wholesale) may reflect the fact that the nature of these activities is such that many of the self employed will be registered for VAT.
As well as differences in levels, Table 9 shows significant movements in differences from year to year, implying quite different growth rates in labour input between ABS and the equivalent industry within the National Accounts framework. These movements are especially notable in 2009 for a number of industries including F, L and R. And there is a saw-tooth pattern of differences for industry B.
|% differences from ABS||2008||2009||2010||2011||2012||Average|
|B-J, L-N, RS||7%||9%||11%||8%||9%||9%|
Table 10 shows differences between National Accounts estimates of GVA and those published in the latest ABS release. In interpreting this table and Table 11 below it should be borne in mind that the National Accounts for 2012 have not yet been balanced through the supply-use framework. In the light of Table 9 one might expect that the National Accounts GVA estimates would be consistently larger than ABS, reflecting output of that element of the self employed outside the ABS sampling frame. It can be seen, however, that while this is the case for the matched aggregate as a whole and for a majority of industries, there are large negative differences for industries D, M and N. In these industries, the National Accounts estimates of current price GVA are significantly smaller than the ABS estimates. The hugely positive difference for L reflects the output of housing services from owner-occupied dwellings, which is included in the National Accounts estimate of GVA but has no equivalent in ABS.
It is also apparent from Table 10 that the time series properties of the two sets of data are quite different in a number of cases. These differences reflect a number of factors, some of which were covered in the Introduction section above. More information on the differences between ABS and the National Accounts estimates of GVA is available in Section 9.1 of the ABS Technical Manual (ONS, 2012) and in Ayoubkhani (2014).
|PPT differences from ABS||2009||2010||2011||2012||Average|
|B-J, L-N, RS||13.8||-32.0||3.3||6.2||-2.2|
Table 11 brings together the information on labour input and GVA and additionally takes account of price movements in comparing growth of constant price GVA per hour between estimates produced top-down within the National Accounts framework and estimates derived, bottom-up, from published ABS estimates. In interpreting this table it is important to note that productivity estimates from the ABS are derived from double-deflated1 estimates of constant price GVA while those in the National Accounts use published constant price GVA series derived from single deflation. This methodological difference injects more volatility into the ABS estimates, as discussed further below.
The top right-hand cell of Table 11 shows that, for the aggregate matched industries, average growth of constant price output per hour was 2.2 percentage points lower in the National Accounts estimates over 2008-12 than according to the ABS. For comparison, productivity growth for this aggregate in the National Accounts averaged -0.1% per year. Differences are especially large in 2009 and 2010, with ABS suggesting a much steeper fall in productivity in 2009 and a massively larger recovery in 2010.
There are eye-catching differences in productivity growth rates throughout the table, affecting almost all industries and all years. One interesting feature is that the differences tend to cancel out partially over time, so that the average differences in the last column, while still large, are not as large as those for a single year.
Overall, National Accounts productivity growth is lower than that derived from ABS over the period 2008-12 for all industries except L and R. Average differences in productivity growth are largest for industries D, F and H, and smallest for industries I, J and M.
Using double deflation produces more volatility in the estimates of GVA than simply deflating GVA by a gross output deflator. One reason for this is that price movements of inputs and outputs, while being related to each other, do not match perfectly. As GVA is often a relatively small proportion of gross output and intermediate consumption relatively large, this means that a small difference in price movements of inputs and outputs is amplified when GVA is computed using double deflation.
To illustrate this, Figure 2 plots the relationship between the standard deviation of each two digit double deflated GVA series and the corresponding share of intermediates in gross output. There is a positive relationship between them across a meaningful range, indicating that, as the intermediates share of gross output increases, so does the volatility of the time series of double-deflated GVA. This phenomenon is identified in the OECD productivity manual (OECD, 2001) as a likely consequence of double deflation.
Analysis conducted for this article highlights two areas for possible further investigation. As always, ONS welcomes feedback from users on these areas and indeed on any aspect of measuring productivity.
The framework for double deflation developed for this article can be seen as a small but important step towards constant price supply-use balancing. The absence of constant price supply-use balancing in the UK National Accounts has the consequence that some of the implied deflators between current and constant price GVA estimates are not consistent with the source data on gross output deflators. This is because, while the current price GVA estimates are fully balanced through the supply-use process, the constant price series are constrained to a growth path given by deflation of the expenditure components of GDP. Deflated expenditure is preferred for the path of GDP as a whole partly because, at the time this methodology was adopted, the coverage and reliability of final expenditure deflators for components such as consumers’ expenditure and fixed investment were superior to output deflators (at that time there were no services producer prices, for example). Moreover, as noted by Oulton (2004), the ONS method of using real gross output as a proxy for real GVA induces an error in the estimate of GDP(O) growth1.
To some extent we have skirted around this issue in this article by using implied GVA deflators to deflate ABS turnover estimates. But a better approach would be to confront the issue head on: deflating turnover by a set of independent output deflators. The advantage of such an approach is that it would yield methodologically sound estimates of growth of value added which could then be reconciled within a constant price supply-use framework with the deflated expenditure components. One technique would be to use the same output deflators as used in the GDP(O) system (before adjustment of the constant price growth path). However, another technique would be to use the supply-use tables to weight output deflators in the same way as we have used them to weight deflators for intermediate consumption. This would mean using appropriate final expenditure deflators for the share of industry output that goes to meet final demand, and using consistent deflators between industry y outputs that are inputs to industry z, and industry z inputs from industry y. This would likely mitigate one of the issues highlighted by the preliminary double-deflation framework developed for this article, namely the occurrence of large disparities in deflator movements on the input and output side of certain industries, including industries where inputs represent a large proportion of the value of output.
A second area of development, of particular interest from the productivity perspective, is the treatment of the self employed. Until recently ONS did not produce a consistent industry breakdown of labour input between employees and the self employed, but work in 2013 to address a long-standing requirement under ESA95 has remedied this (Allen & Franklin, 2013). As noted above we are now able to estimate that around 680,000 of the self employed are captured in the ABS sample for the comparable aggregate industries. This compares with an estimated 3.5 million self employed workers in these industries in 2012. ONS plans to build on the framework developed for this article with a view to producing experimental productivity statistics for employees and the self employed.
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: email@example.com
The main source of data for this release are the published data tables from the ABS, which can be found on the ONS Website. Currently the most recent finalised data are for 2010 and revised data for 2011, however the provisional results for 2012 are also included in these estimates.
The output data source used in this article is ‘Approximate’ Gross Value Added (aGVA) at basic prices. These differ from the GVA estimates in the National Accounts, which are subject to a number of quality and consistency adjustments. More information on the differences between GVA estimates is available in Ayoubkhani (2014).
The employment figures published in the ABS are taken from the Business Register Employment Survey (BRES), and are total employment 'point in time', and 'average during the year'. For this article we use 2-digit estimates of hours worked by employees from a framework which is benchmarked to overall hours worked from the Labour Force Survey, with a small adjustment to account for the proportion of self-employed that is captured within the ABS/BRES sample frame. We use published BRES point in time estimates to apportion hours worked below the 2-digit level.
Creating intermediate consumption deflators requires having information on the breakdown of each industry’s consumption of other industries’ products. This is provided in the intermediate consumption matrix part of the supply-use tables published by ONS, and can be used to weight PPI and SPPI deflators to account for the price changes of different products.
This is complicated by the lack of detailed information on imports in the Supply Use tables, as each industry will consume a mix of domestically sourced products and products imported from abroad in its intermediates. The prices of Imported products and domestically produced products will change at different rates, and while separate PPIs are available for domestically and imported products, this is of little use without weights of intermediate consumption of domestically and imported products of each product group. Input Output Analytical tables published by ONS in February 2014 (ONS, 2014b) allow this to be done for 2010, but estimation had to be used to derive import weights for other years.
In order to compare productivity estimates over time, the aGVA figures must be deflated to remove price effects and leave a volume index. In the National Accounts, this is currently done by deflating current price GVA by gross output deflators. A more complete approach is to deflate turnover and intermediate consumption using different deflators before computing real GVA, a process known as double deflation, as done in this article. The ABS provides current price turnover (Gross output) and current price GVA.
Double deflation is a conceptually a more correct method of producing real output figures than simply deflating nominal GVA with gross output deflators, as to do so would ignore price changes of products used by an industry but are produced by other industries.
For this article, the computation of a constant price index of GVA is done using a Törnqvist index. For this, current price weights of intermediates and GVA in gross output are calculated, then averaged over each comparison period. This is the preferred method of calculating a series of constant price GVA outlined in the OECD measuring productivity manual (OECD, 2001).
Equation 1 shows how the constant price series of GVA is calculated using a Törnqvist index. The resulting index is a geometrically weighted average of an index of gross output and of intermediate consumption.
An alternative method of calculating double deflated constant price GVA would be to use a Laspeyres form of double deflation, which would involve subtraction of a constant price index of intermediates from a constant price index of gross output. In this form, weights enter in prices of the base period, which places restrictive assumptions on the underlying production technology of the industry. This has a number of effects. It means that over a comparison period, any substitution of labour for intermediate consumption (as occurs in outsourcing of factors of production) does not manifest correctly in GVA estimates as it should. Also, the Laspeyres form can result in negative values of GVA in constant prices even when current price estimates are positive. This is because when imposing different year prices on a current year production process, there is no guarantee that the production process will still be efficient. The Törnqvist calculation method deals with this problem.
While more conceptually sound than single deflation of value added, double deflation has a tendency to result in much more volatile estimates than single deflation. This is due in part due to the sensitivity of estimates to the intermediate consumption share of gross output. It is also a consequence of the large differences in deflator movements of gross output and intermediate consumption, as described previously.
Producing intermediate consumption deflators requires estimates of industries consumption of products of other industries, so that a basket of input deflators can be weighted up to reflect price changes of products used by each industry. This is done for this article by deriving current price weights for the use of each product group by each industry (with each weight separated further into domestically sourced and imported products) and using these weights to aggregate selected input product (and service product) deflators. This results in price indices for each supply-use level industries’ intermediate consumption; these are then mapped and used to produce volume indices of intermediate consumption for each industry in ABS.
This method suffers from some weaknesses, namely the reliance on full industry coverage for deflators as for some industries, particularly in the services sector, supply use level deflators are not available. For services sector industries for which no SPPI is available, a number of different sources have been used, including CPIs, and in a small number of cases gross output deflators from GDP(O). This is a problem which cannot be overcome without fuller supply use level coverage of service industries with SPPIs.
A number of PPIs used in this article do not contain data for before December 2008, so the 2008 figures are imputed from the December 2008 figure.
A further difficulty in producing these estimates are that the supply-use tables for 2012 are not available at the time of writing, and will be released later this year. This means that the figures for 2012 had to be imputed. The supply-use figures used in this release from which current price weights are derived are an average of the 2008-2011 figures.
Products used as intermediate consumption include domestically sourced products and imported products. This makes deflating them more challenging, but it can be done if the intermediate consumption of each product can be split into that which is domestically sourced and that which is imported. Information needed to construct weights to do this are available in the Input-Output Analytical Tables for 2010 (ONS, 2014b). These weights are used for this article for 2010, and it is assumed that changes in the proportion of imported products in intermediate consumption would vary with changes in the proportions of imports in the total supply of products. It is intuitive to assume that firms would respond in a similar way to changes in relative prices of imports and domestically produced products as would consumers of final products.