Back to Africa

"All present-day estimates of Africa’s mid-twentieth century population seem to agree broadly that the continent contained around 280 million people in 1960 and 220m or so in 1950", I wrote here. I should of course (in addition to raising each figure by five million or so) have added "except those of Frankema & Jerven".

In their 2014 paper "Writing history backwards or sideways", the two authors echo Manning in querying reported African growth rates for the first half of the 20th century: this time they go one better by concluding that even the continental population in 1950 is underestimated today by up to a tenth.

The paper suffers from the same fundamental problem as Manning's: reluctance to credit late colonial-era African populations with the growth rates implied by contemporary population returns as adjusted where necessary by subsequent scholarship, a perspective I find slightly dispiriting in seeming at risk of sidelining the resilience and adaptability of former colonial peoples in times of enormous change.

Frankema & Jerven start by considering three cases of relatively well-documented populations: Ghana in 1948, Nigeria in 1953, and Kenya, for which they work back from the seemingly fairly comprehensive 1969 census to the less complete 1962 count and beyond. They propose respective upward revisions of 12%, 28% and 8% in the official numbers. But at least two of the cases, there are serious questions concerning the appropriateness of the suggested resizing, still more so when applied to the generally-accepted regional population total.

To start with Ghana, the authors are entirely right to question the 1948 census return in light of the results for 1931 and 1960, both considered highly successful undertakings, and their proposed correction of the result is if anything modest. The problem is that nobody today uses the unadjusted 1948 figure in any case: their revision is already incorporated into all present-day estimates for the period: in fact their estimate represents the very minimum that might be gleaned from the UN data by abandoning the 1950 figure as excessive and substituting improbably high annual growth of 2.8% over 1948-52.

Much the same goes for at least part of the 1952/53 Nigerian count: the raw figure of 31 million is likewise little used today as an accurate count of numbers present, implying as it does a mid-century total of only 29 million. The latest UN projection implies 39.24 million at October 1952 (the weighted mid-point of the census period) - above Frankema & Jerven's 39m despite incorporating an improbably sluggish 1.67% annual growth rate.

In fact analysis of Nigeria’s long-run demographic growth indicates a 1952/53 population of some 36-37 million, with 35 million in 1950 – and such a result is (unlike the higher figure) compatible with both high and low reckonings for 1911-31 and 1991-2006, depending on the timing of growth. Indeed even this may be an over-generous estimate: comparison by the Africapolis project of census data with satellite evidence suggests that Nigeria’s latest return may represent a substantial overcount. No such assumption is required to conclude that both the UN and Frankema-Jerven estimates for earlier periods are needlessly high, sustained by a large 1991 adjustment and low growth rates in the 1950s and 1960s – 1.8% in the earlier decade according to the UN tabulation.

A result 2-2½ million below the authors' represents an upward adjustment of around 18% to the official returns, an exceptionally large correction factor in its own right. But more to the point, the needed correction is already more than incorporated in UN and other international data, none of which – as with the 1948 Gold Coast enumeration – use the unadjusted tally in the first place. If anything, the latest UN number for 1950 – the highest released to date - is almost certainly excessive, not omitting millions but rather inserting 2½-3m who there’s no reason to believe were actually there at all.

And so to Kenya. Here Frankema & Jerven’s conclusion is similarly shaky, at least for 1960:

Assuming a population growth rate of 2.5 per cent for Kenya in the 1960s—which is still a rather high rate—and taking the 11.1 million as a reliable estimate of the 1969 level, one has to adjust the 1962 estimate to 9.35 million (an upward adjustment of 8 per cent).

In fact we need do nothing of the sort, because there is no reason to consider a 2½% annual growth rate high or even sufficient for 1960s east Africa: all of the region’s counts of the period indicate a rate nearer 3%, grounds indeed to raise the 1962 population to 8.7 million or so, with 8.2m in 1960, little above the latest UN estimate. The authors do have a valid point however when it comes to 1950, for which the UN's figures seem low for each of Kenya, Uganda, Tanganyika and Rwanda, understating the regional total by more than three-quarters of a million.

And so Frankema & Jerven’s sweeping critique of modern estimates of mid-century African population estimates turns out to be a damp squib. Their proposed net revision for the three countries studied amounts to not the promised ten million but little over a hundred thousand against the UN numbers, all Kenyan – indeed on balance as few as a quarter of the Kenyans omitted in the UN’s 1950 projection. For the biggest country, Nigeria, which as they observe accounts for most of the uncertainty, their reworking implies a lower figure than the world body's, as does their result for Ghana.

The authors could have found stronger support for higher numbers in the data for the Democratic Republic of Congo, where the latest World population prospects understates mid-century population by anything from half a million to 1½m following decades of downward revisions: Niger, Mauritania and Madagascar seem similarly undercounted, to the tune of perhaps half a million between them. But against such examples should be set those of Mali, Mozambique, Benin and Malawi where numbers are unlikely to have reached those reported, contributing probably well over a million to the contrary side of the account alongside 2-3m excess Nigerians. The net result is not an increment of twenty million but rather a reduction of two million against the present data.

That leaves us still with the original 228 million or so for 1950 - nearly a million more by the UN’s reckoning, a million fewer here. The corrections which would raise the total by twenty million have already been made, in the sensible adjustments that decades ago raised numbers from the clearly deficient 200m implied by raw census numbers and contemporary administrative estimates. Frankema and Jerven have erred spectacularly in directing valid criticisms of uncorrected returns against data incorporating (sometimes in excess) the very revisions they urge.

The authors are correct though in their critique of Manning’s flatline interpretation of pre-1920 African population. Here they echo comments made here about the inappropriateness of using Indian growth trends in an African context and then double-counting adverse factors already incorporated to begin with in the selection of a slow "default" rate. Their suggestion of looking instead to higher southeast Asian rates is attractive:

India’s demographic development constituted an aberration from the Asian pattern. Why would African demographic regimes, operating in a context of open land frontiers, better reflect that of densely populated India rather than land-abundant Indonesia?

Frankema & Jerven conclude by proposing an alternative set of default growth rates averaging the Indian trend with a minimum (1% over 1851-1920, 2% in 1920-50 and 2½% in 1950-60) for Indonesia and the Philippines. The results are a great improvement on Manning's, but again err on the side of low growth by incorporating India’s dire 1876-1920 experience along with inexplicably low rates even for that troubled subcontinent in the 1850s-60s, 1880s, 1900s and 1930s.

The authors have identified much of the problem with Manning's data, and their observation as to the usefulness of southeast Asian trends is a valuable insight. Unfortunately much of their time was spent pursuing a 1950s data revision that was accomplished decades ago and already underpins most of today’s estimates. Tellingly, the table illustrating the impact of their inflation of 1950 population doesn't extend as far as 1960 (unlike the unadjusted series), because then it would have to show an improbable total of 300-310m (against a generally accepted 282-285m) or incorporate implausibly low 1950s growth.

19th-century African population is likely to remain a bone of contention owing to the unknown timing of the rise in annual growth from pre-colonial levels. Independent estimates by Caldwell, Durand and McEvedy (as well as my own projections) have tended to centre around a continental total of 50 million 500 years ago, before the Atlantic slave trade made a significant impact, suggesting in turn roughly 30-35m at CE 1000 and annual growth of 0.1-0.2% over the millennium to 1800, rising to 0.3% by the end of the period.

Assuming some 228 million in 1950, all-Africa population in 1920 is likely to have numbered around 145m, in 1900 125m, beyond which projections are subject to assumptions about the onset of acceleration: 95-105m in 1850 (dismissing Frankema-Jerven's improbably low 0.3% in 1890-1920 along with their deficient 1930-50 rates), depending on whether underlying sub-Saharan growth rose gradually to 0.7% a year by the 1900s and 1.1% in the early 1920s, or stayed below half a per cent until a "big bang" c.1920 following the traumas of conquest, atrocity, war and epidemic; and 80-95m c.1800 reckoning the Maghreb and Egypt at 10-11 million and southern Africa at 2½m.

In concrete terms, given constant or moderately rising Tropical African fertility (the latter assuming a constant crude birth rate of 48 per thousand, the former implying a slow decline from around 52 as gradually improving longevity lowered the proportion of women of childbearing age), the issue concerns the timing of the fall in the crude death rate from 44-48 in the mid-19th century to 29 in 1950, representing an increase in expectation of life at birth from 21-23 years to nearly 35.

Even an "optimistic" scenario incorporating substantial late nineteenth-century gains from New World crop diffusion places the overwhelming majority of the underlying mortality improvement after 1900, here from 24-25 years of life in relatively peaceful areas at the turn of the century to 26-27 in the early 1920s and 30 in the 1930s. Against this must be set losses through armed conflict, disease and exploitation, culminating in nearly four million influenza deaths in 1918-20: Caldwell indeed suggests that population in 1920 was eleven million lower than it would have been had the trend of 1840-80 persisted.

Consensus on the long-term movement is likely to remain elusive, depending on calculations of the aggregate toll of slave removals (including those killed in the process of capture or movement to slaving entrepots), the impact of New World crops, deaths in colonial conquest and through European excesses, and the recuperative power of local populations. But even a cursory reconstruction of the overall pattern suggests lower numbers than those offered in recent revisions.

It is all the more surprising that Morten Jerven should come to such a pessimistic finding about the past after his recent criticisms of at best low-growth estimates of post-1990s African economic performance. It seems likelier that the capacities that enabled Africans to survive and sometimes thrive in the aftermath of 1980s structural adjustment were present in earlier generations and centuries amid the traumas of slave-raiding, conquest and associated disruption.

In submitting their piece, Frankema and Jerven hope to ignite – or rekindle – an exchange about African population history. It is a long-overdue discussion, and their contribution offers a useful reminder of the fragility of all our estimates. But their own overcount offers little better a guide than Manning’s hypothesis of near-stagnation.

Still puzzling after all these years?

Morgan Kelly and Cormac Ó Gráda are at risk of reinventing the Clark-Huberman-Lindert wheel with their attempt at disentangling the seeming anomalies of early modern trends in English food output and consumption. And their paper raises significant questions over how much the frantic and increasingly abstract number-crunching and -regurgitation of intervening years has really added to our net stock of knowledge.

The authors offer a useful reminder of just how widely scholarly estimates of output and intake continue to diverge after (or even because of?) all the painstaking work of the likes of Campbell-Overton, Turner, Allen and the wayward Clark. Even the rival estimates cited for grain yields in 1800 (probably one of the less contentious topics!) vary by 15-30% for individual crops, and the margin of disagreement widens still further for earlier periods or for livestock produce. It’s hardly surprising that the field is bedevilled by differences of interpretation and periodisation, but that’s down to our limited sources more than to those who’ve sought to make sense of them.

As in Clark et al’s classic 1995 paper, the authors point to the discrepancy between the demand increase expected from rising per capita GDP and sluggish growth in consumption of agricultural produce, though this time the conundrum is placed in a far longer timeframe:

According to Broadberry et al, English GDP per capita trebled between c.1260 and the mid-eighteenth century. What they dub 'modest but positive trend growth' is troubling for their estimates of agricultural output and average calorific intake, since even very low income elasticities are not easily reconciled with zero sustained improvement in calorie supplies over the same period.

"Troubling" is putting it mildly, because with agriculture contributing a third to two-fifths of output in the mid-18th century, a tripling of per capita GDP in the preceding half-millennium would make real starvation the 13th-century norm well before before putative overpopulation and soil exhaustion took their toll. Happily Broadberry’s growth rates work out at "only" 2.3-fold income growth over 1270-1760 (though it’s annoying to have to reconstruct the missing aggregates). And even that’s an overestimate derived from an improbably low late-medieval starting-point entirely obscured by the omission of actual GDP estimates.

The problem is exacerbated by the Broadberry team’s overstatement of early and mid 18th- relative to mid 17th- or mid 19th- century income, the implication of an excessive growth rate in 1650-1700 and improbably low ones in 1760-80 and 1800-30 (population conversely appears understated in relative terms over 1650-1780, though to a lesser extent, though any error is difficult to pinpoint as even this number is absent).

If average income barely doubled over 1270-1760 (as a simple estimation of life’s necessities and the incomes of the non-agricultural population might have suggested had Broadberry & co stuck to quantities rather than churning out speculative growth rates), we’re still left with the apparent food consumption paradox raised by faster growth afterward, a problem masked by their high estimates for 1700 and to a lesser extent 1760 compared to 1830 – which brings us back to the very period of the original and still only partly resolved 1995 "food puzzle".

And unfortunately replacing the Broadberry output index doesn’t help towards raising food output growth at this time, because the 2011 paper offers astonishingly low "current-price" GDP shares for agriculture of 29.7% in 1759 and a frankly incredible (even just by comparison with 1759 and allowing for intervening price movements) 26.7% in 1700. Now it’s true that contemporary estimates leave much to be desired and that recent occupational studies have pointed to a far larger industrial workforce than previously assumed. But even amid generally low cereal prices a real agricultural GDP share significantly under 40% seems unlikely for the first half of the 18th century given the upturn in grain exports and the fact that it remained above a quarter even into the 1820s, combined with the small contribution of industry (only some £25m of English GDP even at mid-century, under £30m for Great Britain).

Kelly & Ó Gráda do identify two possible solutions to the paradox, one explored more fully in the 1995 study, the other potentially only deepening the problem. To take the latter first, they suggest that livestock output was higher in the late 18th century than widely thought. And they’re right: both milk and meat appear underestimated in the Broadberry paper, chiefly reflecting low cattle numbers. The problem here is though that while 18th-century numbers may be understated, the same goes still more for yields in earlier centuries. It’s likely on balance that livestock production grew if anything at a slower rate than these estimates indicate.

The other avenue is far more rewarding. If our evidence conflicts with assumed elasticity of demand for foodstuffs, then either the evidence is wrong or our elasticity is inappropriate. And sure enough, Clark &c pinpointed the solution two decades ago: “What people eat and drink differs from the food products supplied by agriculture and net imports.” And what we spend in addition to the cost of crude foodstuffs rises with income, urbanisation and specialisation and our associated move from farm to retail: this was indeed the case in the 18th and 19th centuries when the share of processing and distribution rose from roughly a fifth of final consumption to 45%.

The difference between the two elasticities is striking for the period under review: 0.55 for final consumption, 0.25 for crude farm produce, both falling over time as we should expect. Transport, processing and trade together contribute 0.30 of the figure for end goods, alone more than doubling the rate. Coupled with higher medieval income at about half of the 1730 average, this suggests growth of rather over two-fifths in per capita consumption of crude foodstuffs over 1270-1870, consistent with 19th-century agricultural output estimates and a late medieval population of four million or so. This result captures both Clark &c’s "food multiplier" and shifts in the balance of overseas foodstuffs trade.

The situation in the key period 1770-1830 is less clear-cut. It seems likely that large sections of the population were unable to enjoy the nutritional opportunity implied by per capita GDP growth. Wartime disruption & inflation and subsequent protection may well have constrained food supply until the more favourable 1830s. But agricultural output data are particularly lacking around the latter part of the period, and it may be that contemporary food expenditure returns have failed to capture some of the undoubted gain since even the grain-surplus years of the mid-18th century.

An implication of the wide elasticity margin is, however, slower agricultural growth than in Overton's estimates - here about 2.6-fold over 1700-1870, consistent with appreciably higher cattle and pig output c.1700 (partly offset by lower sheep numbers) and lower 19th-century milk yields than those assumed in the Broadberry project. Coupled with the Cambridge occupational history project's findings of a larger industrial workforce around 1710, this in turn raises questions about 17th-century labour productivity growth. The Agricultural Revolution is likely to prove a longer affair than aggregate output growth suggests.

Unbundling our ignorance

I’m none too impressed by the free-marketeering agenda of London’s Legatum Institute. Nor am I convinced that an inevitably ideologically-driven think-tank backed by billions in private investment capital is necessarily a suitable candidate for charitable status. But top marks to it anyway for posting its rewarding “History of Capitalism” lecture series for all to view. A few years ago it seemed that mainstream academia was poised to do away with intellectual exclusion through the medium of the online seminar presentation, but it’s since dozed off on the job.

And so to the engaging Nick Crafts on Industrialisation: Why Britain got there first (and mercifully not "Why someone else didn’t get somewhere they weren’t headed to begin with"). It’s an enjoyable talk replete with the latest findings of the scholar who with Knick Harley in 1982-83 gave us our first reliable (if even then slightly pessimistic) British industrialisation-era growth rates, for which we’d previously had only the estimates of Hoffmann and Deane & Cole – each similarly state-of-the-art in its day but relying on incomplete or inappropriate weighting or pricing.

Besides a welcome but guarded appreciation of Allen’s high-wage economy explanation (highly regarded hereabouts for all its lacunae) for Britain’s “take-off”, Crafts offers a likewise worthwhile though similarly sketchy grounding of British growth in long-run agricultural performance. Here it’s founded in enclosure (principally of the extra-parliamentary variety) – rather less satisfyingly than a more demand / output-sided approach – but he at least places the origins of Britain’s divergence well before the classical Industrial Revolution period.

There’s the obligatory nod to recent 1990-dollar estimates of past GDP, with 11th-century China this time at $1,244 (I’m not sure where that one came from, but it’s another to add to the mounting heap of similarly questionable extrapolations). But to his credit Crafts doesn’t make too much of the latest trendy figures beyond quite properly noting England’s early relative prosperity – even if it wasn’t quite of the magnitude of the estimates cited, so far as one can tell from any plausible guess at aggregate output.

As probably the leading practitioner of 18th-19th century British growth accounting, Crafts makes great play of total factor productivity growth, the Solow residual or otherwise the famed “measure of our ignorance” – the differential between real output growth and the summed increment of capital and labour inputs, which he more or less equates here with technological advance.

The familiar conservative Crafts take on Industrial Revolution-era growth rates is on show, notably in his finding of annual growth in 1800-30 of a mere 1.7% in aggregate GDP and 0.3% in labour profuctivity, the latter half the increase suggested by my own output estimates (though his TFP growth rate is far closer - of which more later).

Poor Walt Rostow’s hypothesis of a compressed "take-off" stage receives short shrift, with Crafts echoing Deane & Cole in seeing his proposed investment rate increase as a longer matter than the putative 1783-1802 doubling period. For Rostow’s 10% level we’d indeed probably have to look to the 1820s, with the period between the Peace of Paris and the Treaty of Amiens yielding instead a rise in reproducible capital formation only from around 5 to under 7% of net national income (or from 7% to 9% or so for all assets including land), barely half the surge proposed in Rostow’s 1956 speculation (curiously, he was close with his 5% and 10% at the two dates - though unfortunately for different rates).

Crafts is in excellent company – notably that of Deane & Cole – in pointing to 1830-60 as the standout period of elevated investment. But in rejecting their excessive 19th-century output growth estimates (the result of deflating deficient contemporary current-price GNP estimates by Rousseaux’s over-volatile general price index) he errs too far in the opposite direction, replacing their finding of marked mid-century deceleration with one of a failure to realise corresponding productivity gains from the unprecedented investment rates of earlier decades.

It could be, of course, that the underlying capital data are themselves at fault, shaky as they are even in comparison to our necessarily limited GDP reconstructions. Certainly half of King’s 1688 grand total – his capitalisation of human labour – is omitted from modern definitions, as is the £300m of national debt counted in Beeke’s agggregate more than a century later. And the subsequent estimates of Pebrer for 1832 (itself merely a re-scaling of Colquhoun's of two decades earlier) and still more so Giffen for 1885 themselves look on the high side (chiefly owing to the value given for land) given a contemporary overall price level only marginally in excess of that of the last pre-WW1 years.

If that were the problem, however, one might expect the resulting discrepancies to be borne out by the figures. But if the answer lay in Beeke’s unadjusted estimate (the likeliest candidate from the rates offered), we should see a far higher capital growth rate before 1800, otherwise the implied 1760 capital stock is far too large unless we accept an anachronistic mid 19th-century valuation of land for a time well before the arable expansion and near-tripling of urban population of 1800-50.

Instead, it seems that Crafts has accepted needlessly pessimistic rates of capital formation overall, particularly for the 1800-30 period, for which a gloomier perspective is indeed borne out to some extent by his GDP growth assumptions. Where contemporary reckonings indicate something approaching a fivefold rise in the value of domestic capital (here including land) over the century to 1860, Crafts’s rates (and those of Feinstein and Broadberry on whom he draws) suggest something nearer threefold.

It could of course still be the case that contemporaries from King to Giffen and beyond got it badly wrong. But the onus here is on their putative challengers, none of whom have yet offered persuasive grounds for rejection of the broad sweep of the 18th- and 19th-century calculations. On the contrary, intervening revisions of British GDP tend to support a gradual fall in the total capital / output ratio from well over to under six times national income over the two-and-a-quarter centuries preceding World War 1. That’s a different matter to the precipitous drop implied by Crafts’s data.

So we’re left with the rates. And reconstructing their practical implications for the nation’s aggregate capital stock suggests gross understatement of growth of both land and reproducible capital. Indeed, it appears that estimators from Feinstein to Broadberry and Crafts (though not Deane & Cole, who sought meticulously to differentiate the two elements) have erred in applying a rate more appropriate to all domestic assets to reproducible capital alone. Couple that with an unfeasibly low growth rate for the land component itself, and it’s hardly surprising that there’s a problem. Crafts's reproducible capital growth rates – at least for 1760-1830 – lie well below those indicated by contemporary estimates of the resulting stock, even correcting for inflation over 1740-1815 and deflation thereafter. For that part of the national wealth we need to be thinking in terms of a yearly increase exceeding 1¼% in the late eighteenth century and 2% in the early nineteenth, otherwise we face an improbably high total at the start of the period or even steeper growth afterward.

And that's not all. To suggest that the real value of Great Britain’s land appreciated by only 3% over the whole of 1800-30 – a period when arable acreage rose by a fifth, pastoral acreage by perhaps a seventh, the stock of buildings (not included, but clearly relevant to nearby areas) by upwards of £150m, urban population almost twofold from 3.8 million to nearly 7.4m, and mineral output probably faster still – is clearly insupportable without a truly heroic initial assessment for undistinguished tracts of intrinsically far poorer eventually realisable worth. The finding is all the more puzzling when the more leisurely transformation of 1760-1800 is reckoned to have had five times the annual impact.

Land valuation for growth-accounting purposes may well be an insuperable difficulty. The contemporary valuations between them give a clear impression that it was increasing (far more rapidly than the area in productive use) over the two centuries to Giffen's 1875 estimate, and falling thereafter until around 1930. In fact the total appears strongly correlated with aggregate agricultural output, as might indeed be inferred from the link between productive capacity, rental value and purchase price. The latter approach appears unavoidable if we are pricing capital according to its cost, and has the benefit of reflecting current-price capital / output ratios.

The upshot of the prevailing underestimation of capital growth is to greatly overstate TFP's 18th-century contribution: that Crafts is right in the 19th only arises from the happy combination of understatement of both capital and output growth coupled with an improbably high labour input rise in the last period. The apparent stagnation of labour productivity and TFP change in 1800-30 appears to reflect improbably low GDP growth and earlier understatement of capital input.

One finding that these reworkings support is the decline in the share of TFP's growth contribution to 1830, or rather c.1800. This is problematical for TFP as an indicator of technological progress, as neither the 18th century nor still less the period 1760-1830 stands out as a time of relative technological sluggishness. What we may be picking up here is an echo of the commercial disruptions occasioned by Britain’s long intermittent struggle with France: if so it’s a reminder that as a residual, the measure may incorporate a range of intangibles, not least the wider trading context.

Cutting China (and the past) down to size

Kent Deng and Patrick 0’Brien have joined the ranks of those unconvinced by the Maddison take on historical national accounting, or at least its uncritical acceptance among a broad swathe of scholars of retrospective GDP. And their observations shed welcome light on the widening mismatch between currently fashionable backcast projections and benchmark comparisons of past international output levels.

It’s a broad-ranging paper, covering everything from two millennia of wildly conflicting population estimates (a Deng pet peeve, which he then concludes isn’t particularly relevant here) to the reliability of Ming cultivated acreage returns (which the authors similarly determine can’t be relied upon to resolve the matter).

But the cornerstone of their critique concerns Maddison’s assumed present-day international poverty-line GDP level for the Han era, which informs much of the intervening presently accepted guesswork. Coupled with the World Bank’s controversial (and since partially reversed) 2005 downward revision of China’s (and much of the developing world’s) estimated PPP-adjusted income per capita, their unpicking of the long-run assumptions underlying Maddison’s time series suggests persistent overstatement of China’s GDP. And that raises questions about everyone else's past GDP too.

Indeed, the authors go further:

Sensitivity tests reveal that there is something amiss not simply with the evidence and conjectures for the rates of growth utilized for forward projections from the Han era to 1850 but more seriously with the conceptual foundations of any index expressed in constant 1990 International Dollars utilized as a base line level for the projection of long term trends in GDP per capita.

It’s possible indeed that Deng & O’Brien’s implied downscaling (“implied”, because at this point they sensibly conclude that the numbers game really isn’t worth the candle given the paucity of available data) doesn’t go far enough. The shortcoming here lies in their conversion of Maddison’s Han-dynasty baseline from anachronistic 1990-era dollars to rice-equivalent units, seemingly the first step in their analysis.

I must confess some bafflement at the authors’ argument here. Leaving aside for a moment the crucial conversion error whose correction would strengthen their case hugely, their equating of Maddison’s year 1 CE income to 60% more than the rice-equivalent associated with food security suggests an uncharacteristically conservative conjecture on the late professor’s part. Even very poor people don’t spend all of their scant money on food: there’s clothing and hopefully housing, fuel, tools and utensils to be acquired one way or another, whether by purchase or by making them, either of which means GDP on top of food production. An outlay of 62% of income on the most basic adequate diet seems if anything quite high, still more so when applied to a whole "national” population of poor and rich.

But it isn’t a daily per capita GDP equivalent to a precarious-looking 3,374 kcal of edible rice that’s the problem here, it’s a purported annual income two millennia ago equivalent to something around three tonnes of grain, a level no country (apart from perhaps the odd city-state) attained at current domestic prices before the modern era: even booming eighteenth-century England reached the threshold only once, in 1761, a year of unusually low cereal prices. Here Deng & O’Brien have underplayed their hand spectacularly in converting Maddison’s Geary-Khamis dollar figures to rice-equivalents on the basis of a 1:1.1538 G-K dollar:yuan exchange rate now entirely superseded by intervening ICP revisions and in any case the wrong tool for the purpose.

As noted here previously, a feature of Geary-Khamis modern-dollar GDP results (and a source of my growing annoyance with their incomprehensibly escalating ubiquity in HNA) is that they don’t readily relate to any sectoral or particular product value, thanks to the blanket impact of inflation of output of services and other internationally non-traded items. The consequence is that rebasing the G-K numbers according to any real output quantity just doesn’t work. In my opinion this very non-additivity renders them all but worthless and irrelevant for any meaningful purpose, but on that score I’m in a minority, albeit one that on this latest evidence seems to be growing.

So regardless of any underlying conceptual nonsensicalities, a G-K 1990-price Chinese GDP of $450 two millennia ago shouldn’t necessarily relate to any Chinese rice price of 1990. But while strictly incompatible with the PPP method, the exercise gives a striking illustration of the scale of the problem. The authors' domestic edible rice price (648 yuan per tonne) appears in fact to be the average state procurement price for paddy, and therefore needs raising to take account of farmers' market sales: the result is $135-150 per tonne of paddy, depending on whether we use official or market exchange rates – not far short of the reported global average producer price of $160. As Maddison's prices are international, I use the latter, which also disagrees least with his data.

Estimates for China in the first half of the twentieth century (in which I’ve rather more faith than have Deng & O’Brien) suggest a current-price GDP per capita rather below the value of a tonne of grain. And there’s no reason to believe things had ever been much better given a millennium of improved yields and new crops. On that basis, Han per capita income would be nearer $150 than $450 at 1990 prices - indeed somewhat less, since rice even as paddy tends to command a higher price than most grains.

To put it another way, agriculture (here including forestry and fishing) contributed some $80-90 per head of 1990-price GDP (roughly 550kg in grain-equivalent) in the relatively well-fed mid-1950s, and is most unlikely to have produced appreciably more over any prolonged period in the past. Estimates for the 1930s put its share at 60-65% of current-price GDP, and this time it's unlikely to have previously been much less: recent studies by Broadberry and Shi Zhihong et al indeed put it higher in earlier centuries. But Maddison's Han GDP estimate is at least five times any plausible agricultural value added reckoned at actual 1990 international prices. This might well be a proper consequence of the backcast PPP approach, but would a procedure implying such drastic upsizing and distortion of sector shares really have anything useful to tell us?

Limiting the exercise to Maddison's own calculations up to his death in 2010 indeed understates the scale of the problem. For subsequent work by Bolt & van Zanden finds still higher pre-modern G-K incomes for most regions, though not (yet) for China, which is thus left lagging behind (in itself a wholly plausible finding) even on the basis of a Qing figure that now appears improbably high. Shi & his colleagues go one better by proposing an early Qing per capita GDP of $807 - more than the 2005 ICP revision's implied level for 1990 itself. Broadberry & co go still further with a mindboggling $1,518 (this time exceeding also the perhaps more realistic 2011-derived result) in 1020, only decades into the prosperous Song era and before the impact of early-ripening rice in the south: even supposing China's people each to have munched or worn their way annually through a heroic tonne of grain-equivalent in agricultural output, the latter would barely exceed a tenth of GDP. At a time when we might hope for downward revisions across the board, the message from Groningen, Utrecht and elsewhere is on the contrary one of still further inflation of past incomes.

Deng & O'Brien have done us all a favour with a courageous paper that errs only in pulling its punches when they had the wherewithal to strike a mortal blow at the phantom of an unrealistically bloated economic past which seems to pervade ever more otherwise promising quantitative research. The initial Maddison project gave us an early tentative glimpse into the scale and timing of long-run economic growth, but its increasing tendency to ossification of and extrapolation from questionable and now seemingly superseded data threatens to reduce it to a purely academic exercise ludicrously divorced from historical reality.

Maddison's own shift of base-year from 1980 may itself have exacerbated the problem by widening the divergence between ancient and recent commodity and services prices, though his failings seem small potatoes compared to the reworkings of his heirs. The whole undertaking is in need of more drastic surgery than just more of the same doubtful backcasting of projections from projections. The cure must begin with the price base. Prices of the very recent past can’t be counted on to say much about ancient, medieval or early-modern economies, beyond how big they might have been if the prevailing prices had been those of centuries or millennia later. But they weren’t.

Tax is good for you

Liam Brunt and Cecilia García-Peñalosa offer a provocative take on State capacity, urbanization and the onset of modern economic growth:

We argue that the First Industrial Revolution occurred in England in the middle of the 18th century precisely because it was uniquely urbanized before 1750.

This is something of an oversimplification of the paper’s case: Britons, or more precisely English farmers, produced more because they had to pay out more than foreign growers in rent, tithe and taxes. This forced them to produce more food, thereby supporting larger urban populations whose increased volume and range of economic and social interactions favoured intellectual exchange and innovation. The proposition is attractive in that it draws together the phenomena of agricultural growth, urbanisation and demand-driven and/or technology-led industrialisation. But could agrarian rents really be a driver of urban industry? It’s a lovely thought amid today’s welter of tosh about voting and property rights, but is it a speculation too far?

Brunt & García-Peñalosa seem on solid ground so far as the productivity-urbanisation relationship is concerned. Adam Smith said something similar in 1776:

As subsistence is, in the nature of things, prior to conveniency and luxury, so the industry which procures the former must necessarily be prior to that which ministers to the latter. The cultivation and improvement of the country, therefore, which affords subsistence, must, necessarily, be prior to the increase of the town, which furnishes only the means of conveniency and luxury. It is the surplus produce of the country only, or what is over and above the maintenance of the cultivators, that constitutes the subsistence of the town, which can therefore increase only with the increase of this surplus produce.

As we might expect, higher food output per worker indeed tends to be associated with a larger proportion of townspeople, though an admittedly rough estimate for principal products – albeit one based largely on later data – yields a rather more urban Holland, Belgium & Italy and a more rural Wales, Hungary & Romania than might be suggested by agricultural productivity alone (presumably indicating sizeable food imports among the former group and exports of grain and livestock from the latter).

The authors’ analysis starts by overstating a point that isn’t entirely irrelevant to their analysis, and might usefully have been improved upon:

China, which is regarded by many historians as the most technologically advanced country in the Middle Ages, had much higher urbanization rates than Europe around the year 1000, with about 3 per cent of the population living in cities in the former and virtually zero in the latter. However, over the next 700 years Western Europe experienced a massive increase in urbanization rates, reaching almost 10 per cent by the year 1700, whilst no substantial change took place in China.

This is over-egging the pudding somewhat. The authors’ dismissal of extravagant claims for China’s urbanisation is welcome, and they are right to note that while Europe forged ahead in the following centuries, China struggled to add a single percentage point. But while Europe’s urban population around 1000 was still struggling to recover from its post-Roman low, it certainly wasn’t zero, even in the then relatively peripheral north-west. There are grounds to think that western Europe wasn’t so far behind China, especially if one adopts a lower urban population threshold than the authors’ preferred 10,000 (the cut-off point for de Vries’s city-by-city study): the proportion was lower in north-western Europe, but still not far short of the Continental average. China had more big cities with its sophisticated imperial administration and advanced internal communications, but Europe contained many smaller regional centres, trade emporia and former Roman capitals in varying states of decay or repair. At a more appropriate urban threshold of 2,000 the Europe-China differential would all but disappear.

So perhaps western Europe on the eve of its 11th-century revival wasn’t quite such a backwater in the urbanisation stakes, despite the continent’s relative economic frailty. Just as importantly, by 1300 or even 1200 Europe – however defined - may well have been already ahead, a crucial point lost in comparing only the beginning and end of the 700-year period. This doesn’t invalidate the authors’ case, and may even support it: exactions at the expense of Europe’s cultivators had doubtless increased in the 11th-12th centuries as emerging monarchies and feudal elites jockeyed for advantage. But while overall urbanisation remained low, more precise dating of Europe’s lead underlines the need to look at the medium as well as the long term, and to seek possible explanations of the “Great Divergence” also at times well before the 18th century.

Urbanisation is only one step in the analysis, though a key one in that it is seen as a proximate motor of Europe’s technological revolution. The cornerstone of the authors' argument is the link between agrarian extraction (effectively rent, since this is the main difference between the English and Chinese rates) and the agricultural productivity growth which made large urban populations sustainable. In trying to identify something approaching a “take-off”, however, Brunt & García-Peñalosa seem to have allowed their enthusiasm for English agricultural growth to get the better of them:

Clark finds an increase in output per worker of 44 per cent between 1700-09 and 1770-79. Moreover, Clark documents that output per worker already rose sharply in the 16th and 17th century, increasing fivefold between 1500 and 1650.

A fivefold increase in output per worker in 1500-1650 is of course out of the question, not least given an intervening increase of only around five percentage points in the urbanisation level. Fivefold pre-industrial productivity growth would in any case surely be anathema to Clark’s neo-malthusian heart, necessitating a compensating population increase to soak up the surplus output. In fact Clark finds a productivity drop of about a quarter, not entirely implausible against a background of rapid population growth and the resulting reversal of the land windfall occasioned by the excess mortality of the 14th and 15th centuries. My own estimates suggest a modest rise in labour productivity in the two centuries before 1650, and an increase of a third or so in 1700-70, the latter closer to Brunt (2000) than Clark. For the following century, though, Brunt & García-Peñalosa seem needlessly cautious in their estimates – a rise of a mere fifth when population and trade data suggest an improvement of three-fifths or more. That debate will doubtless run and run, sustained by the high 18th-century output estimates which Clark derives from his series for land rent, wages and capital inputs.

Again, none of this disproves the paper’s basic premise, but it raises once more the issue of timing. There seems to have been nothing exceptional about the level of rural exactions in the latter half of the 18th century when England outstripped its continental rivals in urbanisation (or rather most of them, a significant qualification) and industrialisation: the burdens faced by growers had been in place for at least a couple of centuries, for much of which time provincial urbanisation had been sluggish. Agricultural advance was, like urban growth, a gradual process. But the greatest increases in both agricultural productivity (though not aggregate output) and urbanisation would come in the following century.

The next element – leaving aside the matter of state capacity, not entirely convincingly dealt with (I remain unpersuaded that this correlates with high agrarian rents) – is the scale and impact of compulsory extraction from agriculture:

In table 4 we report the tax burden on agricultural workers in England and China in c.1775, which was almost 100 times greater in England. Since output per worker was around ten times higher in England than in the typical Chinese province, the tax rate was around ten times higher in England than China.

Here again, some of the data seem questionable (“tax” here includes rent and tithe, but that’s clearly indicated). Given that output per English agricultural worker was only perhaps £35 in the 1770s – equivalent to four or five tons of grain – it’s difficult to imagine a Chinese peasant producing a mere tenth as much, still less sharing it among family, landlord, tax-collector and buyers. Chinese productivity was certainly a good deal behind, but a more accurate ratio would seem to be around a third, even lowering Perkins’s generous per capita grain allowance (based as it is on the best years of the 1950s rather than the perhaps more representative 1930s). Nor is the authors’ estimate of 47 pence for China’s land tax per man in agriculture entirely convincing: this might have been the amount going to the Imperial coffers, but other estimates suggest a burden perhaps three times as great after allowing for provincial surcharges, payment in lieu of corvée, outright misappropriation and differences in prices. Again, that doesn’t invalidate the paper’s argument, but it moderates the implied differential, even without considering whether tenancy had already begun to replace peasant ownership on a significant scale, a process whose timing remains unclear.

A more serious shortcoming is the paper’s omission of comparisons between the burdens of English growers and their Continental counterparts. Limiting the exercise to England and China addresses one part of the Great Divergence, but what of the “Little Divergence” within Europe, or perhaps we should say the many little divergences between those countries that surged forward and those that lagged behind? What would estimates for France (where peasant burdens were notorious but urbanisation remained well behind England’s until the latter half of the 20th century) and the highly-urbanised Netherlands or less prosperous lands further east tell us? And what light might, for instance, be shed on the issue by a similar exercise involving British and French territories in North America or the Iberian colonies to the south?

A full comparison of peasant burdens among Europe’s economies alone would be a mammoth undertaking, fraught with still greater difficulty than the relatively clear-cut England-China case. But the example of France suggests that the results may not be so supportive of the Brunt-García hypothesis. Against the authors’ finding of a 23% burden on English farms (possibly on the low side, it must be admitted), that of French peasants has been put at up to a third of output – one of the many grievances that fuelled the Revolution. Yet French agricultural labour productivity remained far below England’s – perhaps £16 of output per person engaged, well over China’s £11 but just half of the English level – and the gap was widening. France was also considerably less urbanised than England, and again falling behind: as late as 1801 under a fifth of the French population lived in towns of 2,000 or more, no greater a proportion than in the England of 1700, and in the meantime England’s urban share (using the same threshold) had topped 35%.

Against the case of France we may contrast that of the Netherlands, where urban population in Holland proper had shot up as early as the 14th century at a time when rural exactions were modest, widespread peasant ownership reflecting easy terms designed to draw cultivators to newly-reclaimed land. By 1514 more than 35% of Holland’s population lived in towns of 2,000-plus, a level not to be matched by England for another three centuries (though of course on a far greater scale), and the province was already a world leader in terms of income. Yet agrarian tenancy only became widespread during the 16th century, with townspeople prominent among the new rural landlord class. City growth continued apace, it’s true, with Amsterdam alone booming from a respectable provincial town of 12,000 in the 1510s to a commercial metropolis of 105,000 in 1622 and 150-200,000 on the eve of the disastrous French invasion half a century later.

But crucially, Holland’s initial urban explosion wasn’t matched by neighbouring provinces with a less unburdened peasantry. Nor was the rapid income growth of 1400-1650 sustained from the third quarter of the 17th century, despite the putative rent stimulus to the industry of an innovative and highly productive farm population. Having quintupled in 1514-1672 to half a million, Holland’s urban population stagnated for the next century and a half. Agrarian productivity remained high, at around £30 per worker, the highest on the Continent owing to a highly developed dairy sector, though by 1770 somewhat behind England. But a comparison of France and Holland – or indeed between 15th- and 18th-century Holland – would yield the very opposite finding to the one suggested by the starker but less like-for-like England-China contrast.

At the end of the day we are left with the familiar chicken-and-egg dilemma. To put it crudely, did London’s population quintuple in 1600-1800 because the bloated land rents of Middlesex’s dwindling rural portion forced the county’s remaining farmers to produce more to feed their urban neighbours? It’s an absurd question, but illustrative of the wider problem of causality. London drew in its supplies from a far wider area, both domestic and international, and while its growth clearly profited from adjacent resources, it in turn impacted on nearby land values: Clark’s early 19th-century rent benchmarks show clear crests around London, Manchester and Birmingham, as might be expected from their leading role in Britain’s commercial and industrial advance.

Brunt and García-Peñalosa’s approach has definite appeal: along with literacy, the rise in European urbanisation well before the classic industrial revolution period remains something of a mystery, if not a cause of the later economic upsurge then at least an expression of forces that would later underlie industrial expansion – whether agricultural growth, diversity and exchange, Allen’s trade-driven high-wage economy or Clark’s “survival of the richest”.

the authors have offered a stimulating avenue for future research, though their approach shares the limitations of all such attempts at monocausal explanation. In fact they could have gone one better by investigating the implications of their tax-and-grow scenario for Allen’s alluring (though itself not fully explained) hypothesis of an exceptionally commercialised economy driven to labour-saving innovation by the cost of its own workers, indeed by its very success. Then we might really be on to something, even if only as a partial answer to the hugely complex question (as Brunt titled his 2000 comparison with France), “Why England?”.

A darker continent

Patrick Manning has some interesting things to say about African population growth in the eighteenth and nineteenth centuries (in press; earlier paper with Scott Nickleach here): whether his conclusions will find widespread acceptance is another matter.

Manning’s latest addition to the growing body of scholarly speculation concerning Africa’s demographic past continues the recent trend toward ever higher estimates of pre-colonial – and here also colonial – population. In place of slow growth interrupted by the devastating impact of slave trading, we now face two-and-a-half centuries of near-stagnation giving way to sluggish recovery.

Where Frankema & Jerven proposed something just under a doubling of the continental total in the century to 1950 – itself a pessimistic growth assessment compared to the earlier guesstimates of McEvedy & Jones and (to a lesser extent) Caldwell – Manning offers a picture of overall decline in the century to 1880 followed by an increase of little over half over the next six decades.

The picture isn’t one of unrelieved gloom, though it comes perilously close: the northern and southern ends of the continent get to grow throughout the nineteenth century (though until 1920 at a pitifully slow rate), while the Guinea coast – the first region to feel the impact of European slave shipments – more-or-less holds its own after the post-1792 decline in removals to the New World.

All present-day estimates of Africa’s mid-twentieth century population seem to agree broadly that the continent contained around 280 million people in 1960 and 220m or so in 1950. There the agreement ends. For 1920, Manning’s estimate of 159m compares with the 147m of Frankema & Jerven, Caldwell’s 142m, and fewer than 135m implied by MacEvedy & Jones. Further back in time, the gap widens: in 1900, 145m (Manning), 137m (Frankema), 129m (Caldwell), 110m (McEvedy); for 1880, 142m, 128m, 120m and under 100m respectively. Manning’s total for 1800 is fully twice McEvedy’s, and 50% greater than the level indicated by Caldwell.

The sources of these large discrepancies are inevitably complex, but for the pre-colonial era they boil down chiefly to the degree to which slave removals and related deaths were offset by demographic recovery, ongoing growth through continued expansion of cultivation, and later the spread of new crops and (toward the end of the period) the beginnings of economic development. For the colonial period, the principal issue is the extent to which early census returns underreported total numbers.

Taking the latter period first (since all the estimates are essentially back-projections from the more reliable data for the 1950s and subsequent decades) there is little disagreement that colonial censuses – many of them barely worthy of the name – generally understated African populations. Contemporary estimates around 1950 indicated a continental total of only around 200 million, roughly a tenth below the true level. Nigeria’s population was widely believed to be only around 25m before the 1952-53 census returned a population of 30m, itself an undercount.

Manning is thus right to raise many of the official colonial population returns. But his procedure goes far beyond mere adjustment of deficient counts and government estimates: what he has done is effectively to discard all the pre-1950 data on the basis of a glib rejection of a few patently flawed guesstimates from the 1930s. This overlooks the wealth of intervening scholarly and indeed bureaucratic revision. European colonial officials were themselves aware by the 1930s of the shortcomings of their raw data, limited as they were by meagre administrative resources and legitimate native distrust of head-counts: the author of Nigeria’s 1931 census report for instance inflated the crude return by a tenth to compensate for under-enumeration and evasion. (Nor were earlier errors in population estimation all in the same direction, unless one is to accept figures of up to 4½m for Uganda in the 1900s.)

For the thus abandoned contemporary figures Manning substitutes assumed “default growth rates” which are themselves highly questionable. For these he draws on trends from the more complete census data for British-ruled India, which he then adopts as maxima for Africa given the continent’s disturbed condition prior to 1920. But this overlooks the very different character of the two regions: India was a long-settled society already boasting extensive dense occupation and highly developed state structures long before the Atlantic slave trade: Africa, by contrast was still in the throes of internal agricultural colonisation, as Caldwell concluded in his 1985 General history chapter (favourably cited by Manning), basing his projections on the assumption that:

the neolithic revolution has been slowly moving through sub-Saharan Africa for three thousand years bringing with it more intensive land use and denser settlement.

Furthermore Caldwell considered the resulting population growth to be accelerating in the nineteenth century through expansion into the forest belt and the spread of new crops (though the latter process was to be more marked during the colonial period). To these developments one might add the ongoing movement of Nguni and other Bantu-speaking peoples into today’s South Africa, nineteenth-century growth in Egypt and the beginnings of European settlement and agricultural transformation. Such an evolution has important implications for long-run estimates, because African population must in the past have been a good deal less than at the start of the colonial era, even allowing for intervening losses to slaving.

Manning’s default rates are not themselves obviously outlandish, the lower bound of his 0.2-0.3% pre-1920 annual range corresponding as it does to the increment needed to raise McEvedy’s “optimistic” 16½m Africans of AD1 to the billion of 2010: indeed given annual growth well in excess of 2% since the 1950s, rates must have been a good deal lower even than this for most of Africa’s past. But to assume such sluggish growth into the modern era reckons without the falling continental population share of the (until 1800) slower-growing north and the geographically ever more constrained hunting & gathering peoples to the south. It may be, of course, that sub-Saharan cultivators’ rate of increase fell markedly following their initial expansion through the southern half of the continent, but this does not appear to have been addressed, nor is there an especially strong case for it given that the global erosion of non-farming populations is likely to have been replicated on a smaller scale even in and around already settled areas.

While growth rates before the twentieth century remain a matter of conjecture, there is far less basis for the severity of Manning’s wholesale downward revision of those for the colonial period. While census-taking was at best patchy and in some areas subject to deterioration, the data for 1910-50 offer a consistent enough picture of quickening growth. That annual growth in the 1950s in most territories comfortably exceeded 2% a year itself suggests that there is nothing intrinsically implausible in rates of 1% or more in the century’s early decades or even of 2% in some areas from the 1930s. Manning’s default rates may be reasonable for the centuries to 1800, but appear at odds with the mounting demographic evidence as we move into the census period, conflicting even with the Indian evidence on which they are supposedly based.

The impact of Manning’s gloomy assumptions can be seen most clearly in the case of Egypt and South Africa, generally considered the continent’s fastest-growing nineteenth-century territories. Where French observers reckoned 2½m Egyptians following Bonaparte’s arrival in 1798 – since raised by twentieth-century scholarship to some 3-5m – Manning estimates 10½m, more than were counted in the census of 1897, usually considered the first reasonably accurate enumeration. For South Africa, his procedure gives 6.3m inhabitants in 1800 against only 5.2m counted in 1904 (the latter doubtless an underestimate, but coming after a century of colonisation both black and white): though probably an understatement like many of his numbers, McEvedy’s figure of a mere 1½m for the earlier year looks far more plausible given that much of the interior remained to be settled, particularly its western half.

There is a further problem with Manning’s projections: recall that these are his “default” rates, extended to the entire period to 1920 on the grounds of the continent’s experience of slave shipments and conquest by European powers. But what he then does is to apply a range of “situational modifications” representing (among others) these same factors, thereby effectively double-counting the losses and disorders which underlie his pessimistic global model. It is the compounding of these various elements which produces the pre-1890 population standstill, where a more plausible underlying growth rate subject to similar modifications would have shown a century of modest but quickening growth, preceded it is true by net losses in at least parts of the continent in the latter decades of the eighteenth century.

In sum, Manning’s estimates are an interesting but seriously flawed contribution to an area of historical demography in need of more work, on colonial demography and on the impact of slave removals & raiding on African populations. By selecting a set of growth rates and countervailing adjustment factors which effectively preclude dynamic development anywhere until well into the last century, he has constructed a model of built-in stagnation which inadvertently minimises Africans’ regenerative capacities and all but ignores widely divergent trends within the continent. There may well be grounds for raising all estimates of past African population, but approaches which offer so little scope for directly observed data and local variation are unlikely to offer a lasting way forward.

You guesstimate three tonnes, and what do you get?

Offline for a while, but back in time to catch up with Bolt & van Zanden’s “First Update of the Maddison Project”. And what an update it is: away with European per capita incomes of $450 or so (in those wonderfully descriptive 1990 dollars, of course!) – now it’s $600 or bust for any cutting-edge early-first millennium AD economy worthy of the name.

And it gets worse. For 1800, British per capita income is raised from around $1600 to $2100 – still below the Netherlands, now up still further at $2600, a level not reached according to Maddison until the return of rapid growth in the 1860s.

So just where would $600 in “1990 Geary-Khamis dollars” come from before the Roman Imperial economy had even necessarily peaked, let alone when Europe might be well past the imagined economic halcyon days of the early Principate? Remember we’re talking the equivalent here of four tonnes of grain net per head of population at 1990 international market prices, a level of real output not reached in current-price terms until the later Industrial Revolution in mid-nineteenth century Britain and the US.

Now it's true that G-K dollar values aren't necessarily meant to work in such terms: rather, when denominating real GDP they're intended to reflect economy-wide purchasing-power parities, but in such a way that the resulting aggregate can no longer be used to derive individual sectoral output values. This caveat applies doubly to Maddison-style extrapolation of modern-day PPP-adjusted GDP figures into the past on the basis of local GDP growth rates.

So far, so good. But the results for individual points and areas should resemble those from independent benchmark observations based on the same price level, in this case that of 1990. The problem is that they don't. Maddison’s series routinely throw up far higher past income levels and correspondingly lower growth rates than those suggested by individual point estimates based on identifiable output. Far from reducing this discrepancy, Van Zanden & Bolt’s revisions have widened it, from the first century through to the nineteenth.

So is there some realistic basis for such a gung-ho addition to Angus Maddison’s already highly generous allowance for pre-medieval Europe? Or is this just the product of some dubious approximation to commodity-price-basket-deflated putative “real wage” estimates dismissed in their earliest incarnation by Postan as ”not worth the paper they were written on”?

Well it certainly isn’t coming from four tonnes of grain-equivalent in per capita product - any grain, you name it, it just isn’t there unless the authors can identify some towering sphere of economic activity that no-one’s spotted to date - one moreover capable of tripling the estimated worth of real product relative to one calculated from identifiable outputs: it isn’t there in medieval or early-modern England or France, and there’s no indication of it anywhere else prior to modern industrialisation.

The stock answer to this is of course "services", which inevitably assume a greater value in past times when their GDP contribution is re-based on modern values: in pre-industrial societies service-sector labour is cheaper relative to other branches of the economy because wages and prices haven't been dragged up by better-paying productive mechanised industry, the growth of skilled occupations and resulting inflated living costs of labour.

So valuing past or less-developed-country GDP in modern global or developed-country terms involves a big upward adjustment in the contribution of incomes from services. The problem is that the later the price base, the greater the potential distortion between countries and earlier point estimates relative to current-price estimates. Valuing output hundreds or even thousands of years ago at 1990 prices effectively compresses the range of past incomes by revaluing services with low productivity growth as if they were returning a late twentieth-century real income.

But that's not all that’s going on here. For one thing Maddison’s 0.3% annual increase in British per capita GDP over the bustling 1820s is here flattened further into 0.2% annual growth over the whole of 1800-25. The mind boggles: can it really be possible that the world’s leading industrialising economy performed so poorly during the middle decades of its key phase of early industrialisation when material output indices suggest a rate nearer 1%? Never mind Crafts & Harley’s comparatively modest revisions, even the dour Clark would surely be dumbfounded by a finding of such sluggishness.

And a discrepancy of such magnitude inevitably impacts on earlier estimates given the way these series are cobbled together. As we’ve seen, the would-be modifiers of Maddison’s admittedly rickety historical accounts assemblage are sometimes only too keen to splice together growth projections using procedures even shakier than the often cloudy data underlying the whole exercise. Nor is it at all clear just what valuation throughout in 1990 dollars can really be expected to accomplish.

Ultimately it has to be asked whether, for all their unquestioned scholarly worth, the academics of the Maddison Project are the best people for the task, given their differing approaches to past growth. Maddison’s own results to some extent involved an uneasy compromise between time-series projections and benchmark data. To deepen the confusion risks shoehorning disparate findings into a form for which they are ill suited.

Lost in Podolia

Lately catching up with Andrei Markevich and Mark Harrison on Russian/Soviet national income from 1913 to 1928, I’m impressed at the authors’ readiness to dirty their hands filling in the yawning chasm between our available benchmarks for a society transformed in the interim by war, revolution and chaos. I’m less than entirely convinced though by some of their choices in an exploration necessarily hedged to some extent by assumptions in the absence of comprehensive data.

It’s a huge topic, covering the economy of a nearly tenth of the world over a period of traumatic political & social upheaval and rebuilding, complicated by statistical uncerainties throughout and by the legacy of past manipulation and politicisation. Harrison and Markevich try to get to the truth by stripping the economy down to its nuts & bolts, with output series for key sectors weighted according to Falkus’s estimates and brought into line with Gregory’s 1913 national income figure by means of a further allowance for services.

For the most part the data are derived from existing production series, with interpolations for years for which firm figures are missing. Strictly the approach yields an index of gross output rather than value added, but it’s a tried & tested method that underlies most pre-20th-century estimates and for most countries remains the best we have. The resulting totals persuasively flesh out our hitherto sketchy impressions, broadly supporting the accepted picture: aggregate net national income for the territory of the interwar USSR fell by three-fifths in 1913-19, regaining its pre-war level in 1926/27, but income per head remained below the 1913 peak into the Plan era.

The calculation however involves some questionable assumptions which leave open the possibility that the drop may have been rather less precipitous and subsequent recovery more complete (though by no means complete in per capita terms).

First, in addressing the question of population, the authors rightly seek to correct the official population returns for what has long been known to be an ongoing cumulative overstatement arising principally from failure to deduct internal migrants from the estimated population of their former place of residence, while adding them to that of their new location. Deducting Finland and Poland from the population total for the "censused" part of the Empire, they then deduct a “compromise” 5½% to correct the official data, observing that this corresponds fairly closely to the findings of R I Sifman, though the resulting 152m for mid-1913 in fact falls short of Sifman's implied total by upwards of a million.

The authors’ next step is to arrive at a population for the interwar USSR by excising the areas lost after the Revolution: the later Baltic republics, western Belarus & Ukraine, Bessarabia and smaller areas of Transcaucasia. Here, Markevich & Harrison arrive at 19.7 million for the population of the ceded territories, concluding that the interwar USSR area contained 134.6 million people in 1913. But the figure for the lost provinces is inflated by the erroneous inclusion of Podolia, which instead remained a part of Ukraine throughout the USSR’s existence.

Deducting Poland and the corrected estimate for the lost territories from Sifman’s Empire total (which already omits Finland) and adding Khiva & Bukhara (not covered by the Imperial returns) yields a 1913 population of not 134 million but just under 140 million for the USSR area, very close in fact to Lorimer’s 1945 finding which put the ceded provinces at only 14.9 million in January 1914. The result underlines the unwisdom of applying a blanket correction factor across the Empire: while Szulc finds a 7% overcount in Poland, the interwar USSR included the principal receiving areas for internal migrants, so it requires a smaller adjustment than the Empire as a whole or even its non-Polish part. The error doesn’t affect the constant-area national income figure, but it threatens to distort the implied per capita trend.

One assumption that does affect the national income total relates to agriculture, where the authors accept Gosplan’s later upward revisions of pre-Revolutionary crop output. The modification has been widely questioned, as Markevich & Harrison acknowledge, but they propose its retention for 1913-19 in order to being the earlier data into line with the returns for the 1920s. They may be correct in so doing, but most sources retain the unadjusted returns. The Soviet authorities themselves seem later to have abandoned the correction, though this might be dismissed as an attempt to play down relative Soviet underachievement. Naum Jasny, among Soviet agricultural performance’s fiercest critics, seems to have found the original Imperial data satisfactory. The net impact could be to understate 1920s national income by 5% relative to 1913.

The paper’s assumptions don’t all point in the direction of higher pre-revolutionary per capita output. On the perennial question of product quality, the authors make no downward adjustment to Soviet output, observing that while some doubtless fell below pre-revolutionary standards, other sectors shared to some extent in wider technical advance - which might be read as a polite way of noting that the Russia of 1913 was itself hardly famed for cutting-edge finery.

An error that doesn't affect the constant-area national income trend but instead overstates the "Empire" series is the authors' application of the Falkus/Gregory aggregates for the Empire to their own "Empire excluding Poland and Finland". Falkus and Gregory indeed exclude Finland, as did the contemporary imperial returns - but they include Congress Poland, representing nearly a tenth of the total in the truncated form used by Markevich & Harrison (ie following Kholm gubernia's detachment in 1912). The smaller Empire on which the latter base their income figures may have reached the higher income total, but the estimates cited in the paper suggest something lower than the authors' findings. Happily the error doesn't affect the USSR series, which the authors base on Falkus's own Empire-to-Union conversion, itself a good deal more reliable than Gregory's.

All in all it’s a worthwhile effort, let down by some decidedly rickety & inconsistent population estimates and inattention to geographical coverage which could overstate the 1913 Soviet-area per capita income by as much as 4% relative to the 1920s, in addition to a possible far greater over-statement of pre-Revolutionary crop output and high figures for the old Empire. Markevich & Harrison are to be commended for tackling a too long neglected topic, but it remains one which would benefit from closer investigation.