Are all LEPs created equal? Workplace and job dynamics DRAFT not for quotation

Transcription

1 Are all LEPs created equal? Workplace and job dynamics DRAFT not for quotation Michael Anyadike-Danes & Mark Hart Aston Business School & Enterprise Research Centre Version: 13 May 2014 The research reported here was supported by the ESRC Secondary Data Analysis Initiatives programme, grant: ES/K00400X/1. 1

2 Abstract Towards the end of 2011 the UK government announced a list of 39 Local Economic Partnerships (LEPs) for England, joint local authority/business bodies which were to promote the local growth agenda, replacing the nine Regional Development Agencies whose abolition had been announced in the previous year. The LEPs are very new and information about their economies is quite sparse, so our purpose is to provide some baseline measurements of the LEP landscape, which can feed into the policy-making process and, in the longer term, provide a useful context for evaluating the progress they have made. Using workplace-level data we investigate two measures of economic vitality: the dynamics of workplaces and the dynamics of jobs. We find that there is virtually no relationship across LEPs between the business birth rate and net job creation LEPs with large birth rates have proportionally larger death rates although variations in the difference between the birth rate and death rate do account for some of the variation in rates of net job creation, the contribution to net job creation by continuing firms is more important almost everywhere. JEL codes: L25; E24; M13 2

3 1 Motivation & approach 1. The commitment to the creation of Local Enterprise Partnerships (LEPs) was announced in the UK government s Coalition Programme for Government: We will support the creation of Local Enterprise Partnerships joint local-authority-business bodies brought forward by local authorities themselves to promote local economic development to replace Regional Development Agencies. Cabinet Office [2010, p. 10] Just over a year later, at the end of 2011, the process of LEP creation was complete and a list of 39 LEPs was approved. 2. This paper does not discuss the motivation for the creation of LEPs, nor does it discuss their design, or their funding. 1. Our concerns are more narrowly focused, albeit of central concern to LEPs. We investigate two measures of economic vitality: the dynamics of workplaces and the dynamics of jobs. The purpose is to provide some baseline measurements of the LEP landscape, which can feed into the policymaking process and, in the longer term, provide a useful context for evaluating the progress they make. 3. In brief, we find that there is virtually no relationship across LEPs between the business birth rate and net job creation. The (proximate) reason is that the death rate is very strongly (positively) correlated with the business birth rate across LEPs, and the difference between the birth rate and death rate is uncorrelated with the birth rate: in other words, LEPs with large birth rates have proportionally larger death rates. Whilst variations in the difference between the birth rate and death rate do account for some of the variation in rates of net job creation, the contribution to net job creation by continuing firms is more important almost everywhere. 4. This deliberately empirical paper starts with a discussion of data sources and construction. The first step in the analysis introduces 1 On the motivation see the White Paper on local growth Department of Business, Innovation and Skills [2010]. On design and implementation see the two reports of the House of Commons Business, Innovations and Skills Committee House of Commons, Business, Innovation and Skills Committee [2010] House of Commons Business and Committee [2013], which include volumes of evidence and thoroughgoing witness interviews (including with government ministers). 3

4 the LEPs by summarising data on workplaces and jobs, we then examine in more detail in two successive sections workplace dynamics and job dynamics. The next section draws together the analysis of workplaces and jobs and the final section sketches some implications for policy. 2 Data sources & definitions 1. We use the UK Business Structure Database 2 (compiled by the Office for National Statistics) 3 which records annual data on employees for the entire population of firms in the UK and their constituent workplaces. This data is a series of annual snapshots of the Inter- Departmental Business Register (IDBR), an administrative database which captures information from a range of sources, amongst them VAT returns and employer Pay As You Earn (PAYE) tax and social security records. The unit of analysis here is a local unit with at least one employee 4 Local units are plants, for example, a retail outlet or factory. The plant is the source of business activity. It may be a factory that produces nished goods or an accountancy ofce, for example. Evans and Welpton [2009, p. 72] We refer to these local units as workplaces rather than plants which seems to suggest manufacturing operations. 5 In our analysis of LEP performance we use workplace-level data, rather than firm-level data, because we are interested in the location of jobs and in firm- 2 The statistical data used here is from the Office of National Statistics (ONS) and is Crown copyright and reproduced with the permission of the controller of HMSO and Queens Printer for Scotland.The use of the ONS statistical data in this work does not imply the endorsement of the ONS in relation to the interpretation or analysis of the statistical data. The analysis upon which this report is based uses research datasets which may not exactly reproduce National Statistics aggregates. 3 For a full, official, account of the Business Structure Database and its compilation, see Evans and Welpton [2009] 4 Since an employee can work for more than one workplace summing over firms produces an estimate of jobs rather than employment, we ignore this distinction here and use the terms employment and jobs inter-changeably. 5 The ONS seems occasionally to use the term site, whilst the term establishment is used in the United States (see Sadeghi [2008]) but in the EUROSTAT system local units can have more than one location, their equivalent term for the UK s local units appears to be a local unit of homogeneous production (see EUROSTAT-OECD [2007, pp ]). 4

5 level data jobs are located at the firm s headquarters not at actual workplaces. 2. We have linked together the annual snapshots from the BSD using local unit-level identifiers to form a longitudinal workplace-level database for the each of the English LEPs and have devised algorithms to produce workplace-level demographic markers for birth and death. The birth of a workplace is dated by the first appearance of non-zero employment and its death is treated symmetrically and dated by the disappearance of the last employee. Although the data start in 1997, workplaces alive in 1997 could have been born in any previous year, so the first birth year we can identify with certainty is Workplaces are classified as either private or public sector and we make this split using the classification by industrial sector. All employees in: public administration and defence; education; and health and social work (SIC92 6 sections L, M, N); are classified as public sector. Of course, some workplaces in these sectors (in health and/or education for example) are private, and some workplaces in our private sector are public, but ours is a reasonable approximation and ensures that most typically longer lived public entities (like schools and hospitals) do not distort our results, all of which are for the aggregate private sector. It should also be noted that the dataset being discussed here has only recently been compiled, so its properties (and idiosyncrasies) are not yet well understood or documented. However the key findings are likely to be quite robust, after all they represent averages over millions of unit-level records. 4. There are two important caveats. Firstly, as mentioned earlier, the formation of LEPs was a bottom-up process involving discussions between business and local authority partners. LEPs were built up as aggregates of local authority units with their boundaries to be determined by boundaries of local economic activity and in some cases this has led to LEPs with overlapping membership. 7 For example, York is a member of York, North Yorkshire and East Riding LEP as well as Leeds City Region LEP (in fact there are 37 local authority areas which are members of two LEPs). One consequence of this 6 the UK version of the EU NACE rev.1 7 For a discussion of the boundary issue and of the process of LEP formation see House of Commons Business and Committee [2013]. 5

6 overlap is that summing data over LEPs (rather than individual local authorities) involves a degree of double-counting. 8 Secondly, the ultimate data source, the IDBR is a live register. Consequently the date attached to data from the the BSD refers to the date at which the IDBR snapshot is taken (March each year), individual unit-level records may be reporting job numbers some time in the previous year. This means in practice is that it would not be appropriate to rely on the dating of observations to (for example) the precise identification of turning points in the business cycle. 9 3 Getting to know the LEPs 1. Irrespective of the indicator used, there is considerable variation between LEPs, and Table 1 displays some basic measures for 2011: in the first two columns, and of central interest here, are counts of jobs and workplaces; and in the next column population, a contextual variable. Notice first of all the extraordinary range in each of these variables. The maximum and the minimum are the last two rows of the table, and in each case, as we can see from the ratio row on the table, the maximum is around 20 times the minimum. Moreover, the distribution of each of these three indicators is highly skewed, with a median (also at the foot of the table) the middle of the distribution which is much closer to the minimum than the maximum. Some ratios between the indicators are also reported in the table and, as we shall see, these turn out to be an effective means of identifying patterns in the data. 2. Organising the data in size order can assist its interpretation and here the LEPs are listed in decreasing order by number of jobs. So London Enterprise Panel (lond) 10 is at the top of the table and, unsurprisingly, accounts for the bulk of all jobs: 2.8 million, almost 20% of the LEP total. 11 The South East LEP (seast), close to London, is 8 Using population in 2011 as a measure: population summed over LEPs is about 10% larger than the population of England. 9 The PAYE and VAT data on the IDBR are updated more or less continuously but nonetheless it is not possible to associate an aggregate, summed over large numbers of individual records, with a particular month of a particular year. 10 On its first mention each LEP is given its official name, thereafter, the names are shortened, typically to a single word. All the official names are listed in the Appendix table. 11 The firm-level jobs figure for London is one third larger, about 4 million, illustrating 6

7 next with a further 900 thousand jobs, and then two other LEPs centred on cities, Leeds City Region LEP (leeds) and Greater Manchester LEP (manch), which each account for between 700 and 800 thousand. Taken together the top 10 (down to North Eastern LEP (noreast) on the list, a quarter of all LEPs) account for half of all jobs. At the bottom of the jobs list is Cornwall and the Isles of Scilly LEP (corn), with just over 112 thousand, less than 1% of the LEP total. 3. An ordered line plot provides a clearer picture of the shape of the distribution. Figure 1 displays the jobs data in decreasing order (like Table 1), and the numbers are plotted against a logarithmic scale to aid interpretation. Even on a log scale the exceptional size of London is striking but, more significantly, we can see how clearly the big four stand out from the rest there is a drop of two hundred thousand from Manchester, at 717 thousand, to Derby, Derbyshire, Nottingham and Nottinghamshire LEP (dernot), at 507 thousand. Notice too that the rest are (roughly) symmetrically distributed around the median, which is Liverpool City Region LEP (liver, 308 thousand jobs) Derby is roughly two thirds larger than the median, and Cornwall is roughly one third the median. 4. If the size of workplaces as measured by the jobs/workplace ratio were uniform across LEPs, then the distribution of the number of workplaces across LEPs recorded in column (2) would match that of jobs in column (1). However, as we can see from column (4), and although the jobs/workplace ratio is not exactly uniform, it varies very much less than does the jobs figure, and its variation is not correlated with job numbers. So the workplace numbers decline as we move down the table, more or or less in parallel with jobs. The big four from the jobs column, although still forming a distinct group, are less detached from the rest of the distribution, but the rest is still symmetric about the median (West of England LEP (west), 27.7). Of course, there are some anomalies comparing columns (1) and (2): for example the Coast to Capital LEP (coast) (covering parts of Surrey and Sussex), ninth on the jobs list, has 52.1 thousand workplaces, which would place it fifth if the workplaces list were in size order. Typically though the larger a LEP s job numbers the larger is its number of workplaces, and this is also true of the LEPs at the small end of the distribution. In summary, the distribution of workthe significance of the difference between the workplace-level and firm-level datasets, see discussion Anyadike-Danes et al. [2013, p. 60]. 7

8 places across LEPs is skewed too, but a little less than the distribution of jobs. 5. The distribution of population, another measure of size, also matches the distribution of jobs reasonably well, although it is slightly more compact (the maximum to minimum ratio is just 16.4). The similarity between the two distributions is a by-product of the relative uniformity of the ratio of jobs to population (see column (5)). Of course there are some striking examples where the two distributions depart. For example, the North Eastern LEP is a little lower down the jobs ordering (10th) than its population ordering (seventh) would suggest, since it has amongst the lowest number of jobs per head of population. 6. The number of jobs per workplace (column (4)) is, as mentioned earlier, relatively uniform across the LEPs: at the top end, the ratio for Leeds is 12.3, is about two thirds larger than that for Buckinghamshire Thames Valley LEP (bucks), at the bottom with 7.6 jobs per workplace. But Buckinghamshire is clearly an outlier, the second smallest value is Cornwall with 8.6, about three quarters of the jobs per workplace in Leeds: so the range of variation is really very limited. To emphasis this characteristic of the data the ratio of jobs per workplace has been added to Figure 1. To make the variation in the ratio series directly comparable with the jobs figures it has also been plotted a against a log scale (on the right hand axis). The uniformity of the numbers is immediately apparent: none of them stray very far from the dashed line which denotes 9 jobs per workplace (a little below the median of 10.5). Moreover, the pattern of departures from that dashed line is, clearly, entirely uncorrelated (either positively or negatively) with the overall number of jobs. 7. Jobs and population are, like jobs and workplace numbers, strongly positively correlated, so as already noted, the jobs to population ratio (column (5)) is also relatively uniform. Again the ratio between the top Thames Valley at 37.5% and the bottom Liverpool, at 20.5% is roughly two. The top end of this distribution is largely southern, with most of the LEPs close to the top are accounted for by London and its surrounds; whilst the LEPs at the bottom, relatively job poor, end of the distribution are mostly northern. But there important exceptions in both categories, for example Cheshire in the north west is relatively job rich whilst the South East (Essex, Kent 8

9 and East Sussex) is relatively job poor. However, some of these effects may have arisen because, as mentioned earlier, the jobs figures are based on the location of the workplace, so the jobs/population ratio is not a simple reflection of the state of the local labour market: it includes in its numerator jobs located in the LEP but held by those who live elsewhere and travel into the LEP to work. 4 workplace dynamics 1. In this section we explore some of the key properties of the population of workplaces: their vital rates. Annual birth and death rates are calculated as a ratio between the numbers of birth and deaths and the stock of workplaces at the beginning of the year. We start with annual time series data for all the LEPs taken together and then LEPlevel averages over the period 1998 to 2012 to compare performance across LEPs. 4.1 the LEP total 2. Data for the LEP total is displayed in Figure 2. From 2001 to 2008, the birth ratio fluctuated between 15% and 20%, and the death ratio between 10% and 15% (with the death ratio plotted as negative numbers). After 2008 though the birth ratio moved decisively downward, by 2011 it had fallen as low as 12.5%. In contrast, over that period, the death ratio remained in the 10% to 15% range, indeed in 2011 the death ratio was close to the average of the decade up to So the proximate cause of the decline in the stock of workplaces post is lower birth rates not higher death rates. We can see this from the bars in the middle of the plot which measure net birth the difference between the birth ratio and the death ratio there are negative values for net birth in 2009, 2010 and zero in 2011, whilst in previous years, between 2001 and 2008, the stock was typically increasing (except for a marginal fall in 2002). Between 2001 and 2008 the stock of workplaces rose by about 25 percentage points, but after 2008 around two percentage points were wiped off the stock. 12 In 12 The pattern here with decreased births rather than increased deaths accounting for the post-2008 decline is similar to that found in firm-level data, see Anyadike-Danes et al. [2011, Fig. 2, p.13]. 9

10 2012 the birth ratio moved steeply up, the death ratio declined, and the stock of workplaces added eight percentage points. 4.2 LEP-level averages 3. Annual data on the vital rates of the 39 LEPs is a little noisy in particular a small handful of LEPs record quite extraordinary birth rates in some years 13. but the median of each year s 39 observations, on both births and deaths, tracks reasonably well the aggregate series plotted on Figure 2. Since our principal concern here is with longer run relationships, rather than explore the LEP-level annual time series we focus on each LEP s average performance over the 1998 to 2012 period. 14 The birth and death ratio data for each LEP plotted in descending order by birth ratio on Figure 3 reveal that there are some perceptible long term differences between LEPs. 4. The average birth ratio ranges from 19.4% in Coast to Capital, to 13.5% in New Anglia LEP (newang); for comparison, the period average of the total LEP figure computed in the same way is 16.0% (which is a plausible average of the annual data plotted on Figure 2). With a range of just 5.9 percentage points and 39 LEPs there is clearly not much difference between most observations. Indeed there is a very wide but compact middle to this distribution between Hertfordshire LEP (hert, (ranked five) and New Anglia the difference is only three percentage points. So, on average, there is less than 0.1 percentage points between neighbours in the ranking. At the top of the ranking though there are four LEPs which stand out: aside from Coast to Capital, London, Cheshire, and Liverpool, all have birth ratios over 17.5%. 5. Whilst we can see from Figure 2 that the death ratio ordering clearly differs from that of the birth ratio, the plot is qualitatively similar, with the top four from the birth ratio rankings also forming a discrete group with higher death ratios. The clear parallel between the 13 Indeed, some of these outliers are so extraordinarily large that they are, frankly, a little implausible. However, the numbers of observations affected are so small that they are unlikely to affect any substantive conclusions. Also, as mentioned earlier, this dataset is quite new and work on it is still in the developmental phase. 14 Each of the LEP averages here has the same dimension as the ratios plotted on Figure 2 the cumulated sum of births (deaths) is divided by the cumulated sum of the opening stock. 10

11 birth and death ratios suggest it is worth looking explicitly at the relationship between the two, and Figure 4 is a scatter plot of the death ratio against the birth ratio. Clearly there is a strong positive correlation between the two (the multiple correlation coefficient is 0.89): LEPs with larger birth ratios have larger death ratios too. 6. Two lines have been added to the plot to assist its interpretation. The first is the dashed line (though the origin with a slope of 45 ) which joins all the points where deaths equal births. If a LEP is above the dashed line the death ratio exceeds the birth ratio, so the birth ratio less the birth ratio the net birth ratio is negative and the stock of workplaces has contracted. As we already knew (from Figure 3) none of the LEPs recorded a birth ratio less than the death ratio on average over the period 1998 to 2012, so all the LEPs are below the dashed line. 7. A least squares fit to the datapoints has also been added to the plot the solid line to aid visualisation and the slope of this fitted line is So for example, Solent LEP (birth 15.2%) which has a 0.8 percentage point larger birth ratio than Worcestershire LEP (worcs, birth 14.4%) will have (on average) a 0.7 ( ) percentage point higher death ratio than Worcestershire. Moreover, since the slope of the least squares line is less than unity we can infer that the difference between the birth ratio and death ratio the net birth ratio is, on average, positive. 8. The size of a LEP s net birth ratio is equal to the vertical distance from its position to the line. It is quite clear from the plot that, for example, the net birth ratio of Coast to Capital (id number 4 ) is smaller than London ( 23 ) since the Coast to Capital point is clearly closer to the dashed line than London; the London/Liverpool ( 22 ) comparison is more difficult to judge. Whilst it looks as if Black Country LEP (id number 1 ) has the lowest net birth ratio (it is closest to the dashed line), it is generally quite difficult to compare other LEPs from the low end of the birth/death scale to those at the top end. It is not obvious by inspection alone that Cumbria ( 7 ) has a larger net birth ratio than Coast to Capital. 9. Since the net birth ratio is a measure of the expansion of the stock of workplaces, and it clearly varies independently of the birth ratio it is worth looking at a further scatter plot: the net birth ratio against 11

12 the gross birth ratio. From Figure 5 we can now, by inspection, determine that Cumbria ( 7 ) has a larger net birth ratio than Coast to Capital ( 4 ). Two further points are worth noting here. First it is clear that all the net birth ratios fall within quite a narrow range, so that although births and deaths are very strongly correlated, the net births are hardly correlated at all with (gross) births. It follows then, by definition, that the rate of expansion of the stock of workplaces is only very slightly correlated with the birth ratio (the multiple correlation coefficient is 0.19). Second, at the top end, it is now clear that although Cheshire ( 3 ) and Liverpool ( 22 ) have an identical birth ratio, their net birth ratios are different by about 1.25 percentage points (and similarly for Coast to Capital ( 4 ) and London ( 23 )). 10. Though net birth ratio differences between LEPs are typically relatively small from 3.8% in Cheshire at the top to 1.5% in the Black Country at the bottom the cumulative effect over 15 years can produce a considerable divergence in the growth of the stock of workplaces. Over the period 1998 to 2012 the stock in Cheshire has grown by about 75%, that of the Black Country by just 25%. Of the top four LEPs, each with a stock expansion rate greater than 60%, the three (other than Cheshire) are in south east England (London, Swindon and Thames Valley). Of the bottom five, with a stock expansion rate less than 40%, all except the Black Country are in the north: Tees Valley, North East, Humber and Lancashire. The other LEPs, the 30 in the middle, have net birth ratios packed into the interval between 2.2% and 3.2%, with stock expansion ratios between 40% and 60%. Certainly, looking at the detail, does nothing to undermine the earlier impression that there is little obvious connection between the overall growth in workplace numbers and the (gross) workplace birth rate. 5 job dynamics 1. The conventional approach to accounting for job creation and destruction distinguishes two categories of job creation entry, jobs added by new born firms; and expansion, jobs added by continuing firms. There are also two categories of job destruction exit, jobs lost as firms go out of business; and contraction, jobs shed by continuing firms. Finally, there is a fifth category net job creation, the difference between job creation and destruction which, in turn, 12

13 is equal to the change in job numbers. 15 Here we record these same five aggregates, but for workplaces rather than firms the LEP total 2. Studies of firm-level data using annual time series typically find that expansion and contraction are the two largest components of job creation and destruction with entry and exit (typically) playing a relatively minor role (for a recent UK example see Anyadike-Danes et al. [2011, Figure 5, p. 19]). However, the workplace data picture looks slightly different. Specifically, entry and exit become relatively more important. This suggests, by implication, that much of the job expansion and contraction recorded by continuing (multiworkplace) firms involves the opening and closing of workplaces Job creation and destruction figures by component for the years 1998 to 2012 are displayed on Figure 6 where all the series are expressed as ratios to the year s opening stock of jobs. In virtually every year the number of jobs created by entry are greater than (or equal to) the number created by expansion, equally in virtually every year the number of jobs destroyed by exit are greater than the number destroyed by contraction. Moreover the contributions of expansion and contraction do not change much, both the ratios are around 5% to 10%. By contrast the ratios for entry and exit fluctuate quite widely, although they too have been in the 5% to 10% range since about Net job creation (recorded on the bars in the middle of the chart) is always in the -5% to +5% range, and was negative in 2010 and After 2008 the contribution of entry is typically lower, whilst expansion remained roughly flat, exit and contraction dipped in 2010, but otherwise remained fairly flat. The rebound we saw in the 2012 workplace data is clearly evident, with a rise in the contributions of entry, expansion and a dip in the exit contribution, only contraction remained flat. 15 The standard account can be found in Davis et al. [1996]. 16 For a recent review of US Bureau of Labor Statistics data on workplaces see Sadeghi [2008]. 17 However, there may be a data quality issue. Workplace IDs can change for administrative reasons (as distinct from business-connected reasons) and this can generate false death and birth pairs for continuing workplaces. This seems to have been a more important phenomenon in the earlier years of the database. 13

14 5.2 LEP-level averages 5. We use the same approach based on period averages to explore the cross-lep variation in job creation and destruction, as we did to investigate workplace vital rates (and for the same reasons). The data is displayed on Figure 7, ordered by the entry ratio. Evidently the contribution of expansion and contraction fluctuate within quite narrow bands: the range for expansion is 2.6 percentage points, for contraction 2.0 percentage points. In contrast, entry and exit vary more widely. For entry the maximum of 11.6% is recorded by London, and the minimum of 7.2% by Cumbria, a range of 4.4 percentage points. The range for exit is even wider at 4.8 percentage points, but the maximum and minimum are recorded by the same two LEPs, London at 11.7% and Cumbria at 6.9%. 6. There is a strong similarity between the ordering of LEPs by entry on Figure 7 and the ordering by birth on Figure 3. Three out of the top four on the entry list London, Coast to Capital and Cheshire are also in the top four on the birth list. At the other end, two out of the bottom four Cumbria and the North Eastern are at the bottom four of the birth list. The rank correlation between the birth and entry series is 0.79 and hugely significant. 7. The pattern of cross-lep variation in job creation and destruction seems to resemble that of the birth and death ratios in two respects. The apparent positive association between entry and exit and the lack of association between the components of job creation and destruction and the net job creation series recorded by the bars in the middle of the plot. Notice though here we have two LEPs which record negative net job creation, that is net job destruction the Black Country (which had the lowest net birth rate) and Tees Valley. 8. In the absence of any clear pattern across the LEPs in net job creation it is worth investigating the relationship between job creation and destruction (as we did between birth and death) using a scatter plot. To keep the picture relatively simple we have added together the two destruction components exit and contraction and plotted their sum against job creation the sum of entry and expansion. Since there is relatively little cross-lep variation in expansion and contraction this seems a reasonable simplification. 9. Figure 8 the scatter plot of job destruction against job creation 14

15 looks very much like Figure 4, the earlier vital rate scatter plot, and is composed of the same elements. In addition to the datapoints there is a solid line, the least squares fit to the observations which has a slope of 0.89 (the coefficient of multiple correlation between the series is 0.82). There is also a dashed line through the points where job destruction equals job creation: a LEPs net job creation is equal to the vertical distance from its position to this line. 10. Not only does this scatter plot look qualitatively similar to the birth/death plot, we know (given the correlation between ranks mentioned earlier) that some of the same LEPs are likely to feature at the top and bottom of Figure 8 as we saw at the top and bottom end of Figure 4. At the top end of Figure 8 we have both London ( 23 ) and Coast to Capital ( 4 ), whilst at the very bottom we have Cumbria. What does differ between the two scatter plots, though, is the typical distance from the 45 line, and, as we saw on Figure 7, there are two LEPs (Tees Valley ( 34 ) and the Black Country ( 1 )) which recorded (on average) net job destruction. 11. There are two obvious conclusions here. First is that churn in the job market the sum of job creation and job destruction is, given the strong positive correlation between job creation and destruction, positively associated with the scale of job creation: for example, the churn figure for London is about 45%, and is one third larger than of Cumbria ( 7 ) which is about 30%. Second, since net job creation is very weakly correlated with (gross) job creation (the multiple correlation coefficient is less than 0.2) it is hardly surprising that the scale of net job creation the change in employment is almost entirely uncorrelated with churn : the coefficient of multiple correlation is less that The lack of association between net and gross job creation is clear from the scatter of LEP datapoints on Figure 9. We can see London ( 23 ) and Cumbria ( 7 ) at opposite ends of the gross job creation scale, but with almost identical rates of net job creation. Equally, we can see Cornwall ( 5 ) and Tees Valley ( 34 ) with virtually the same rates of gross job creation but at opposite ends of the net job creation scale. 15

16 6 Drawing the threads together 6.1 background facts 1. Our cross-lep comparisons have yielded some reasonably clear findings there is a strong positive relationship between birth and death rates, but net birth rates the change in the stock of workplaces are hardly correlated with birth rates there does appear to be a spatial pattern in births and deaths, with larger cities and LEPs close to London recording higher rates and more northerly and less populated places recording lower rates, but there is no obvious spatial pattern in net births gross job creation and destruction rates are strongly positively correlated but net job creation the change in employment appears uncorrelated with gross job creation there is a spatial pattern in the contribution of workplace entry and exit rates to gross job creation and destruction which resembles that between birth and death rates but there is no obvious spatial pattern in net job creation. 2. The key linkage, which ties variation in birth rates and job creation together, is the relative uniformity across LEPs in the ratio of jobs to workplaces (a feature of the data discussed earlier, see Table 1 column (5) and Figure 1). The connection is not precise, it also depends on the proportion of continuing firms expanding, and the degree of uniformity in the size of continuing firms. However, assuming that births and jobs created through entry, and indeed job creation, will move in parallel, seems to be a usable first approximation. Similarly, and for the same reasons, there are parallels between deaths, jobs destroyed by exit, and job destruction more generally. To tease out this connection we build a simple model consisting of a collection of interpretable ratios (computed as averages over the period 1998 to 2012) which link the business birth rate to the accounting identity which defines net job creation. 3. We can begin by looking at the data on the business birth rate and net job creation. As might have been anticipated there is no simple relationship between these two. We can see in Figure 10 a scatterplot of the net job creation rate against the birth rate a fairly featureless 16

17 cloud of points around the centre of the figure, where the extremes of both series are clearly uncorrelated. For example, the largest values of the birth rate Coast to Capital ( 4 ) and London ( 23 ) are close to the middle of the net job creation rate distribution, as are the LEPs with the smallest rates, Cumbria ( 7 ) and New Anglia ( 24 ). Notice too that Cornwall ( 5 ), the LEP with largest net job creation rate, is close to the minimum birth rate, whilst Tees Valley ( 34 ) the LEP with the smallest job creation rate is around the middle of the birth rate distribution (and Black Country ( 1 ), the LEP with the second smallest net job creation rate is quite close by). 6.2 the model 4. Our analysis begins with the conventional definition of the net job creation rate, where all the components are expressed as ratios to the opening stock of jobs (opening j ), netjcr opening j entry opening j exit + expansion contraction (1) opening j opening j opening j First we express the workplace death rate (defined as the ratio of deaths to the opening stock of workplaces (opening w )) as a ratio to the workplace birth rate (defined, analogously, as the ratio of births to the opening stock of workplaces). So we can write, death birth ( ) = α ( ) opening w opening w For convenience in what follows we denote the ratio of jobs to workplaces as s (for size ), so the average jobs per worker (measured using the opening stock) can be written as, s jobs opening w We can link the entry rate of workplaces to the birth rate and, the exit rate of workplaces to deaths, where the subscripts on s denote birth ( b ) and death ( d ), entry opening j birth s b opening w s 17 exit opening j death s d opening w s

18 To simplify notation the difference between expansion and contraction rates is replaced by a single term which represents the net expansion rate of continuing firms, netexp expansion opening j contraction opening j Finally, we replace the ratio of jobs per workplace at death as a ratio to jobs per workplace at birth, and the ratio of birth size to average size, as single parameters, β s d s b δ s b s Substituting for the rates of entry, exit, expansion and contraction, the relative size terms and the the birth, death relationship into equation (1) yields, netjcr opening j = birth opening w δ α birth opening w β δ + netexp which can be re-arranged to, netjcr opening j = where, γ α β, a size-adjusted death ratio. birth opening w δ (1 γ) + netexp (2) 5. This relationship divides the sources of net job creation into two: the net jobs produced by entering and exiting firms; and net job creation by continuing firms. The ratio between the birth and death of firms, adjusted for the relative sizes of births and deaths, is at the core of the entry/exit term, so a useful consistency check is: under what conditions would this term go to zero? If deaths were equal to births (i.e. α = 0), and average size at death were equal to average size at birth (i.e. β=1), then γ would equal unity, and the entry/exit term 18

19 would be zero, and (ceteris paribus) net job creation would be equal to the contribution of continuing firms alone. Equally then it follows from the algebra (as well as common sense) that the larger the ratio of firm deaths to firm births, and/or the larger are dying firms relative to those being born, the smaller will be the contribution of the entry/exit term to net job creation. 6. Since the terms in equation (2) are not generally familiar (with the possible exception of netexp) a summary can be helpful. Table 2 sets out data on each of the terms for selected quantiles of its distribution. As noted earlier, the maximum rate of net job creation (recorded by Cornwall) is 0.024, and the minimum is actually negative, (recorded by Tees Valley). A more informative guide to the shape of the distribution is the ratio of the third quartile to the first quartile (the Q3/Q1 row in the table) Q3 is 40% larger than Q1. By contrast, the birth rate distribution varies much less, Q3 is less than 10% larger than Q1. We can see from the δ column that firms at birth are about 60% the size of existing firms, and the variation is quite limited. The average ratio of deaths to births, α, is even more tightly distributed about its median value of The ratio between the average size of exiting and entering workplaces, β, is everywhere greater than unity. The middle value of the distribution is 1.18, which implies that exiters are almost 20% larger than entrants, again though there is relatively little variation around the middle. Since it is the product term γ, rather than α or β, which plays a key role, it has been tabulated as well. Below the third quartile it is less than unity, so for about three quarters of LEPs the entry/exit of firms will be contributing to net job creation. The net expansion term in the last column stands out from all the other components: Q3 is more than 50% larger than Q1. So although the central value is quite small, just 0.089, we can see that the maximum is more than ten times the minimum. 7. Of course, these statistics on univariate distributions cannot inform us about the relationships between the components, but correlation coefficients can, and a correlation matrix is displayed in the upper triangle of Figure 11. Most of the correlations are very small. Only two are relatively large: the correlation between γ and net job creation at 0.80; and correlation between the net expansion term and net job creation at The correlation coefficient measures the strength of a linear relationship, so here is it supplemented, in the lower triangle of Figure 11, by a non-parametric ( lowess ) fit to each of the 19

20 scatterplots. Although most of the bivariate plots show quite dense and featureless clouds of points, the fit does suggest that there may be a (non-horizontal) slope to the relationship in the two cases with the largest correlation coefficients: the relationship between net job creation and net expansion and γ. For netexp the slope of the fit is positive, for the other it is negative. So net job creation is (to some extent) positively related to net expansion (of continuing firms), and negatively related to the (size-adjusted) death rate of firms. Notice, most importantly, there appears to be no relationship between the birth rate and net job creation the lowess fit to the cloud of points is flat and the correlation coefficient is just Whilst we can compute the correlation between net job creation and each of these terms, such measures may still not be very informative since the relationship between most of them is non-linear. Counterfactual decomposition provides an alternative strategy for investigating the relative importance of the different factors in accounting for the variation across LEPs in net job creation. For example, we can compute the difference between the net job creation rate in the top LEP Cornwall and another LEP attributable to a difference in birth rates by substituting Cornwall s birth rate into equation (2) for that LEP and computing the difference it makes to the value of net job creation. The sum for a LEP of these differences over the four terms in the equation is approximately equal to the difference between Cornwall s net job creation rate and that of the LEP The results of the counterfactual calculations can be summarised quite simply. Most notably (but hardly surprising given the evidence from the scatterplots), the birth rate and δ contribute virtually nothing to the cross-lep variation in net job creation. For only two of the LEPs is the birth rate contribution greater than 0.001, Cheshire and Liverpool, but in both those cases it was less than For all LEPs δ was always less than (and often considerably less). Necessarily then, the γ and netexp terms contributed virtually all of the cross-lep variation in net job creation, and Figure 12 provides some insight into their relative importance. The continuous line is a plot of the excess of net job creation in Cornwall above that in each of the LEPs. The LEPs are plotted in declining net job creation order, from 18 Here the discrepancy between the sum of differences and the actual difference is very small, the non-linearities are, apparently, too small to be empirically important (in virtually all cases less than 0.01) and we ignore them here. 20

21 the Heart of the South West at the top to Tees Valley at the bottom. The bars represent the contribution of the netexp term to the difference, and by implication γ accounts for most of the distance from the top of the bar to the line (since we know the other components in the decomposition are negligible. 10. Some patterns are immediately visible. For example, for around two thirds of LEPs the netexp term accounts for more than half the difference, in other words these LEPs recorded rather lower values of netexp than Cornwall. There are some obvious exceptions to this generalisation clearly the West of England, Hertfordshire, and London indeed, these are three LEPs which (after Cornwall) recorded the largest values for netexp. By implication, in those LEPs, the γ term plays an atypically large role in accounting for the difference net job creation. Both extremes of the figure are interesting too. At the bottom end, and in two cases in particular (Sheffield and North Yorkshire), the γ contribution is negative (the netexp bar is above the netjcr line), so it adds to the difference. At the top end, the LEPs where net job creation is substantially lower than in Cornwall, we can see that γ and netexp play a roughly equal role. 11. The intuition behind these findings is straightforwardly explained. Since the death rate is very strongly (positively) correlated with the business birth rate across LEPs, it is the difference between the birth rate and death rate, not the birth rate itself, which contributes to net job creation. In our framework the difference between the birth rate and death rate is captured by the ratio α, and we have seen from the accounting framework, its influence is modified by the relative size of exiting firms (β). However, if we compute the correlation between α and net job creation, it is just 0.46, about half the correlation recorded for the compound γ term. Alternatively if we use the method of counterfactual decomposition to compute the contribution of variations in α to net job creation, almost everywhere it accounts for less of the variation less than nextexp. This should hardly be surprising. You will recall that the size of counterfactual differences depend on the size of a term relative to its size in Cornwall and we know from Table 2 that α varies very little (and the value for Cornwall is quite close to the median), consequently its influence is likely to be quite limited. 21

22 7 Discussion and implications for policy 1. We have shown that there is a reasonably clear relationship, on average, and over the longer term, between the workplace birth and death and job creation and destruction. However, we are not claiming that these associations, which are rather more the implications of accountancy than structural behavioural (causal) relationships, could be used to underpin LEP-level policy interventions (see the discussion in Cartwright and Hardie [2012]). Especially since the principal policy outcome of interest growth in jobs is the net result of two very much larger, and opposite signed, flows: job creation and destruction. Rather the purpose here has been to uncover some stylised facts about variations in workplace and job dynamics across the LEPs which might serve to inform the policy-making process. This investigation is best understood as a response to a need identified by the House of Commons BIS Committee in its first enquiry into LEPs to: calibrate LEP performances against one another by way of consistent nationally coordinated data. House of Commons, Business, Innovation and Skills Committee [2010, para. 149] 2. However, there is one important question, related to those which we have investigated here, but which we are not yet able to answer. This concerns variations in the relative importance in different LEPs of single workplace firms, as distinct from branch workplaces of multi-workplace firms and the connected question of the location of a multi-workplace firm s headquarters. Policy-makers are used to dealing with a firm, which means its headquarters, its locus of decision-making. But it is workplaces which are linked directly with jobs in the local area and with job creation and destruction dynamics. Localising policy requires a more detailed understanding of the spatial character of business operations. 3. The (now abolished) RDAs commitment to reduce the persistent gap in growth rates between regions was a key part of their mission and embodied in their Public Service Agreement (see House of Commons, Business, Innovation and Skills Committee [2010, p. 8]), but there seems to have little discussion of the specific role that LEPs might play in the spatial re-balancing of the English economy. We have previously drawn attention to the degree of persistence in the spatial pattern of business vital rates using county data for the 1980s and 1990s (see Anyadike-Danes et al. [2005]) and local author- 22

23 ity data for the decade 1994 to 2003 (see Anyadike-Danes and Hart [2006]), and we now see a similar pattern in workplace births and deaths across LEPs. Perhaps the silence on the spatial re-balancing issue represents a recognition of the difficulty that the spatial stickiness 19 in business vital rates might pose for policy? Or perhaps it represents a more realistic appreciation of the reach of the business support policy agenda? 19 This term has been suggested by Fotopoulos [2013] in an examination of UK new firm formation rates at the NUTSII level. 23

24 References Anyadike-Danes, Michael, Karen Bonner, and Mark Hart (2011) Job Creation and Destruction in The UK: , report, Department for Business Innovation and Skills. (2013) London Business Demography Project, research report, London Enterprise Panel. Anyadike-Danes, Michael and Mark Hart (2006) The impact of sector, specialisation, and space on business birth rates in the United Kingdom: a challenge for policy? Environment and Planning C: Government and Policy, Vol. 24, pp Anyadike-Danes, Michael, Mark Hart, and Maureen O Reilly (2005) Watch that Space! The County Hierarchy in Firm Births and Deaths in the UK , Small Business Economics, Vol. 25, pp Cabinet Office (2010) The Coalition: our programme for government, report, Cabinet Office. Cartwright, Nancy and Jeremy Hardie (2012) Evidence-Based Policy A Practical Guide to Doing it Better, Oxford, England: Oxford University Press. House of Commons Business, Innovation and Skills Committee (2010) The New Local Enterprise Partnerships: An Initial Assessment First Report of Session Volume I, report, House of Commons 434-I. (2013) Local Enterprise Partnerships Ninth Report of Session Report, report, House of Commons 598. Davis, S J, J C Haltiwanger, and S Schuh (1996) Job Creation and Destruction, Cambridge, MA: MIT Press. Department of Business, Innovation and Skills (2010) Local Growth: realising every place s potential, report, HMSO Cmd EUROSTAT-OECD (2007) EUROSTAT OECD Manual on Business Demography Statistics, Luxembourg: EUROSTAT. 24