In a recent research note ("How reliable is the Adcorp Employment Index?", 1 March 2012), two University of Cape Town academics, Prof Martin Wittenberg and Dr Andrew Kerr, criticize Adcorp's method of estimating informal employment in South Africa (see here). They suggest that the method lacks a scientific foundation.

By way of background, Adcorp has been involved in a long-running debate with Statistics SA (Stats SA). Stats SA, using questionnaires completed by around 30,000 people every three months, estimates that there are 2.1 million people employed in the informal sector in South Africa.

Adcorp, using discrepancies between cash in circulation (as a measure of transactions occurring in the cash economy) and official tallies of economic activity, suggests that official records do not present the whole picture, with informal sector employment estimated at 6.2 million people. The upshot of the dispute is that Stats SA estimates South Africa's unemployment rate to be around 25%, whereas Adcorp estimates it to be 9%.

Before proceeding with our response to these criticisms, it is worth noting that Adcorp and Stats SA agree about almost everything. The only disagreement concerns informal sector employment. The correlation between Adcorp's and Stats SA's estimates of formal employment is 83%, not only across time, but also across sectors and occupations.

Adcorp, as South Africa's largest employment services company, is probably the most extensive user of Stats SA's employment data aside from the government and possibly universities. Stats SA's data on wages, benefits, sectoral employment trends, retrenchments and other variables are absolutely vital to Adcorp's research and analysis of the South African labour market.

In other words, discrepancies between Adcorp's and Stats SA's measurements of informal sector employment are important, but the debate should not be taken to mean that Adcorp wishes to discard the entire Stats SA enterprise. Presumably, Wittenberg and Kerr do not intend to imply that the debate concerns employment in all its aspects.

Also, it is worth stating at the outset that, to many people, it may seem that this disagreement about inscrutable statistics and arcane procedures has no practical significance. On the contrary, the very stability and continuity of the South African government hangs on this thread. If youth unemployment is as high as Stats SA says it is, then it is highly probable - as sovereign rating agencies have recently warned the Department of Finance - that the country may experience the kind of unmitigated political chaos that swept through Tunisia, Libya, Egypt, Yemen, Syria and other countries during the Jasmine Revolution over the past two years.

Precisely because the implications are so serious, the debate must be had. It is just not good enough to hope and pray that Stats SA has got it right. It is an entirely legitimate endeavour to sense-check the official figures with all the statistics and methods, direct and indirect, at our disposal. Presumably, Wittenberg and Kerr are not against any debate around Stats SA's employment estimates, simply the methods that Adcorp employs.

It is useful to list Wittenberg and Kerr's criticisms, for ease of reference and in order to structure our response. Unfortunately, several problems arise in our attempt to do so. Firstly, certain of their criticisms cannot be dealt with, even in principle; and not knowing how to respond, we have no alternative but to set them aside. Included, here, is the claim that Adcorp uses its "imagination" when stating that unrecorded economic activity involves unrecorded employment, or their claim that the supposed scale of unemployment in South Africa can be confirmed by simply "looking around", or their claim that Adcorp is "mixed up" about the technical definitions of various classes of employment, or their claim that Adcorp's conclusions are a "publicity seeking exercise", or their claim that Adcorp's conclusions are tantamount to "humbug", "quackery", alternative medicine and climate change denialism.

Ad hominem attacks against a person (and, in this case, a company) rather than a disinterested dispute concerning facts is a surprising tactic for serious academics. But in a sincere attempt to bring the debate to some sort of productive point, we shall overlook their graceless procedure.

Secondly, Wittenberg and Kerr's criticisms have all the attributes of academic respectability. They are senior academics, connected with the faculty of South Africa's leading university. They have published widely, and Wittenberg has published an article (on a particular application of regression) in Harvard's Review of Economics and Statistics, the sixth mostly highly ranked economics journal in the world (gauged by citation intensity).

Their research note broadly follows accepted manuals of academic style, including the citation of references. But the propriety is a veneer. For example, they fail to disclose that their research outfit, DataFirst, provides training and consulting services to users of Stats SA data, which gives them a material vested interest in the unassailability of Stats SA's employment estimates.

To give another example, they fail to cite two ground-breaking research contributions by South Africa's foremost monetary economist, Prof Brian Kantor, which studies are reference works for estimating unrecorded economic activity and its associated employment levels in South Africa, and which studies we provided to Wittenberg and Kerr and which they suppressed.

To give a further example, we provided Wittenberg and Kerr with the report of a large-scale survey conducted by The Business Trust in 2010 which estimated that there are 5.8 million more small business owners and employees than Stats SA estimates, yet they suppressed this research and later claimed that it is "not in the public domain". These research techniques leave a lot to be desired. But in the interests of having a fruitful debate, we shall overlook these doubtful practices.

That leaves us with the issues, of which there are five, some of which have been combined for convenience.

1. Survey-based employment estimates and Statistics SA's approach

Wittenberg and Kerr state that surveys (i.e. direct questionnaires completed by households or businesses, often on their behalf by field workers) are the "gold standard" for estimating employment around the world.

In an ideal world, this is true. We would ask everyone at regular intervals how they spend their time and how much they earn, and they would respond openly and honestly. But in practice, this is unfeasible, not only because population censuses are expensive (South Africa's 2011 census cost around R3 billion), but because a census of the total population is often less reliable than a good sample survey. For example, Stats SA calculates that the undercount during the 2001 census was around 17%, though some researchers put it higher at 26%. The potential discrepancy is enormous: between 7 and 11 million people.

But, for all the sound reasons to use them, sample surveys can be unreliable too. For example, Stats SA estimates that job creation was 365,000 in 2011 - but its data show that the estimate could lie anywhere between 112,000 and 619,000. For 5 out of 9 provinces - covering 45% of total employment in South Africa - the statistical significance of Stats SA's estimates is so poor that changes in the unemployment rate are essentially meaningless. Two surveys, both conducted by Stats SA, differ by about 4.9 million in their estimates of total employment - one, the Quarterly Employment Statistics, is a survey of formal business enterprises; the other, the Quarterly Labour Force Survey, is a survey of households.

These discrepancies are not restricted to employment surveys: the difference between expenditure- and income-based methods of calculating gross domestic product (GDP), also calculated by Stats SA, varies by between -4.2% and +4.3% in particular years, and going back to 1946 the discrepancies are as high as 15% in a single year.

The point, here, is that statistical estimation is invariably imprecise. In some cases the failing is inevitable, associated with very large statistical populations, and in other cases it is a failing of the particular research procedures employed. But there is simply no such thing as a "gold standard" for measuring employment, if by that is meant a statistical procedure that yields precise, accurate, reliable and unassailable results.

Given the imprecision inherent in large-scale economic surveys, it is important to sense-check estimates against other data, including non-survey data. Perhaps some employers do not report some of their economic activities. Perhaps they pay their employees in cash and do not pay income taxes. Perhaps they do not comply with labour laws and regulations. Surely it is naïve to believe that economic activity as reported to Stats SA is all there is? For one thing, we know that survey respondents lie, possibly on a large scale. For example, only 58% of employees report that their employers deduct UIF contributions on their behalf, when employers (with some exceptions) are required to do so.

Only 57% of employees say their employers deduct PAYE or SITE from their wages, when the pattern of income suggests that 70% should be doing so. Only 1.1 million employees report that they receive wage increments in accordance with bargaining council agreements, when the National Association of Bargaining Councils suggests that nearly 2 million workers should be. In 2011, 3.6 million people failed to declare to Stats SA field workers that they received government social grants, an undercount of 25%. Between 2003 and 2010, SARS figures show that the number of personal taxpayers increased by 2.5 million or 74%.

The amnesty programme for small businesses netted around 355,000 new business taxpayers. Isn't it highly likely that these taxpayers concealed (and some continue to conceal) their income? Clearly there is a whole lot more going on in South Africa than official surveys conducted by government field workers suggest.

In the midst of this contested terrain, Wittenberg and Kerr would have us believe that Stats SA's figures are accurate and reliable. In their view, the unreported economy does not exist, no-one is employed in it, and all survey respondents (both businesses and households) make full disclosures to Stats SA. However, the evidence against Stats SA is mounting from many quarters. Bank of America Merrill Lynch analysts, using careful analysis of aerial photographic records, suggest that the size of the economy is (possibly substantially) underestimated.

Actual VAT payments to SARS, when compared to hypothetical VAT payments (calculated by applying effective VAT rates to qualifying final purchases in the national accounts), suggest that the economy's size (notably household consumption, accounting for 64% of GDP) is substantially underestimated. FinScope, a survey conducted under the auspices of The Business Trust, estimates that there are substantially more small business owners/employees (around 5.8 million) than Stats SA. In their enthusiasm to discredit Adcorp's employment estimates, Wittenberg and Kerr failed to appreciate the variety and force of alternative estimates in this hotly contested terrain.

2. Currency use and unrecorded economic activity

In short, the currency demand method of estimating unrecorded economic activity measures discrepancies between income that is officially reported to the authorities (the "recorded" economy) and income that is derived in cash (the "unrecorded" economy). There is, quite naturally, some degree of overlap between the two - but not perfect coincidence.

It is possible to estimate a demand-for-money function (using the determinants of the demand to hold cash, such as household income, the cost of holding currency, regulations covering banks' cash reserves, the availability of electronic alternatives to cash, changes in fiscal policy including government transfers to households, and so on).

It is also possible to estimate a supply-of-money function (using the determinants of the supply of cash, primarily South African Reserve Bank policy captured in an econometric equation known as the central bank reaction function). Observed changes in the use of cash are, technically, the result of changes in both the demand for and supply of cash, and a fairly intricate econometric procedure (such as a simultaneous equations system) is needed to disentangle whether changes in cash use are connected with changes in demand or supply. Having done so, it is possible to estimate whether cash use is above, below or in equilibrium with the forces of demand and supply.

Adcorp has done so, using a procedure pioneered in South Africa by the country's foremost monetary economist, the University of Cape Town's Prof Brian Kantor, whose papers "Black Unemployment in South Africa" (1980) and "The Value of Unrecorded Economic Activity in South Africa" (1989) are reference works for estimating unrecorded economic activity and its associated employment levels.

It is worth noting that there is nothing innovative or controversial in Adcorp's use of this procedure. Prof Kantor's methods and conclusions have never been faintly criticized, let alone refuted, in a recognized peer-reviewed publication and, following universal academic practice, his methodology must stand.

Using his method, Adcorp calculates that the use of cash in South Africa over the past 15 years has come to substantially exceed what can be justified by official economic records alone, or to put it differently, growth in the use of cash has exceeded what can be justified by the demand for and supply of cash alone. Specifically, the unrecorded economy is now around 14.9% of the recorded economy or, to put it differently, official economic records understate the true extent of economic activity by around 15%.

As indicated earlier, this is not contested terrain: the National Treasury's VAT collections are substantially higher than official economic records suggest they should be, implying that the economic records underestimate the true scale of economic activity. The VAT collections discrepancy (around 12%) closely matches the cash-use discrepancy (15%) and the income-expenditure residual discrepancy (17%), and perhaps there are other methods of estimating unrecorded economic activity that can further enhance our understanding of the true economy, but we are not aware of them.

There is no question, in other words, that Stats SA underestimates the true size of the economy. But Wittenberg and Kerr are determined to undermine this now widely accepted conclusion (for the reason that it is the first part of Adcorp's argument that informal sector employment is underestimated), and to do so they use a combination of tangential arguments for which they offer no demonstrable evidence or refutable detail ("cash has other uses than in unrecorded activities"), technical assertions for which they offer no statistical support ("the velocity of circulation of money has changed over time"), and imprudent statements about the national income accounts ("unrecorded activity may well be recorded").

Wittenberg and Kerr would have been much more productive and effective in their criticisms of Adcorp's research if they had paid attention, not to whether Stats SA underestimates the true size of the economy (which is an indisputable fact), but rather to two criticisms which are, if not on a surer footing, then at least more plausible.

The labour intensity of the unrecorded economy

Wittenberg and Kerr claim that Adcorp "guessed" that the unrecorded economy is more labour intensive than the recorded economy. What they characterize as a "thumb suck" is, in reality, a statement of fact. The unrecorded economy is, well, unrecorded - there are no records on which to base any assertions regarding the use of labour in the unrecorded economy.

The unrecorded economy, due to its unofficial status, does not have the same level of access to banks, capital markets, information technology, managerial skill and capital equipment as the recorded economy. The high labour intensity of the unrecorded economy is not a guess - it is an indisputable fact.

We know that the unrecorded economy is dominated by retail and wholesale trade and professional and business services, and since this knowledge comes from Stats SA surveys we suppose that Wittenberg and Kerr will not dispute it. The labour intensity of these sectors is, not surprisingly, higher than the labour intensity of the other sectors of the economy.

Depending on one's assumptions regarding the proportion of various services in the unrecorded economy, the labour intensity of the unrecorded economy is between 12% and 26% higher than the recorded economy. Adcorp's estimate - 20% - is as reasonable as any in light of the estimation uncertainties involved. But Wittenberg and Kerr employ this as a diversion.

If one assumes the same labour intensity in the unrecorded economy as in the recorded economy, South Africa's true unemployment rate is 14% according to Adcorp's calculations. If one assumes that the unrecorded economy is, say, 20% more labour intensive, the true unemployment rate is 9%. In either case, the true unemployment rate is substantially lower than Stats SA's rate.

Unrecorded activity and informal sector employment

It is true that Adcorp's procedure for estimating unrecorded economic activity goes further than Prof Kantor's. Specifically, we make the (apparently reasonable) assumption that, if there is an unrecorded economy, at least some people are employed in it. It certainly isn't possible that the unrecorded economy, amounting to 12% (VAT method) or 15% (cash method) or 17% (GDP residual method) of the recorded economy's size, employs no-one at all.

Specifically, we take the total (recorded plus unrecorded) economy, estimate the labour intensity of the unrecorded economy, estimate the total employment connected with the economy, and deduct Stats SA's estimate of total employment to give us the "employment discrepancy", which numbers around 6.2 million people.

Wittenberg and Kerr have misunderstood the procedure, which leads them to claim that Adcorp does not trust Stats SA's informal sector employment figures, but then appears to use them in the estimate of the employment discrepancy. Adcorp does no such thing. We make our adjustment because Stats SA's definitions of employment have been heavily politicized, and what appears to be unemployment is, in fact, not.

Wittenberg and Kerr wrestle with this issue by going back to first principles, namely the definition of "informal sector employment", which they define (and they claim Stats SA defines) as "employment which takes place in firms employing less than 5 people who do not pay income tax or employment of ‘Employers, own account workers and persons helping unpaid in their household business who are not registered for either income tax or value-added tax'".

No doubt Wittenberg and Kerr can produce the document where they claim this passage exists, but we prefer to go directly to Stats SA's household questionnaire, which makes the following comment:

INFEMPL: This variable is intended to identify persons who are in precarious employment situations. Informal employment includes all persons age 15 years and older who are employed and work in private households and who are helping unpaid in a household business; or working for someone else for pay and are not entitled to basic benefits from their employer such as a pension or medical aid and have no written contract; or working in the informal sector. Formal employment includes all persons age 15 years and older who are employed and who do NOT meet the above criteria. Employers and own account workers age 15 years and older are included in the category 'Other'.

Apart from the contradictions between Wittenberg and Kerr's and Stats SA's definitions, it should be clear from the above passage that "informal sector employment" and "informal employment" are two completely different things. Informal sector employment is simply employment by unregistered employers. By contrast, informal employment is defined as precarious employment and, as such, it is a political, not an economic, concept arising from political lobbying by the International Labour Organization (ILO) to initiate national measurements of such concepts as employment insecurity and underemployment under the rubric of their "decent work" initiative. It is necessary to subtract the political elements embodied in the definition of "informal employment" to obtain a correct picture of informal sector employment.

A far more interesting criticism of Adcorp's research would be that economic activity and employment are measured independently, using completely separate surveys. In this case, economic activity could well be underestimated but employment might not be. This is what Wittenberg and Kerr, if they were thinking clearly about the problem, should be arguing.

The evidence that Stats SA underestimates economic activity is overwhelming (see VAT, cash and GDP discrepancies, above). The evidence that survey respondents underreport, not only their employment activities, but also their dependence on government grants and compliance with statutory contributions is similarly overwhelming (see Adcorp, Merrill Lynch, Business Trust and National Treasury discrepancies, above). Unfortunately, the only way to tell, one way or the other, is to rely on the 2011 population census to be released later this year, and we can only hope and pray that Stats SA avoided the shambles connected with the 2001 census.

3. Econometrics and the use of forecasts

Wittenberg and Kerr claim that Adcorp's calculations do not constitute econometrics, at least as the term is usually understood. In the broadest sense, namely that Adcorp's research is concerned with economic measurement, it is econometrics. But this claim is beside the point.

Censuses and surveys are not "econometrics" either, and the criticism does not hinder our appreciation of Stats SA's estimates of formal employment for that fact. Also, there are no extant econometric procedures that can definitively establish Adcorp's estimates to be "true", any more than they can Wittenberg and Kerr's, if they had them. But there are several statistical procedures that Adcorp employs that are, without question, "econometrics":

• Simultaneous equations modelling (used to disentangle the demand for and supply of cash and determine the excess demand for cash in the economy)
• Time series modelling (used to produce efficient estimates of certain economic magnitudes, used in Adcorp's calculations, that are only known well after the fact)

This last procedure is particularly vexing to Wittenberg and Kerr, who state: "Since not all data is available every month to estimate [Adcorp's] index, Adcorp must forecast some of this, turning the index into a forecast of employment at time t, albeit one that is released after time t."

Wittenberg and Kerr have no concept of the difference between a forecast (i.e. a prediction ahead of time) and an estimate (i.e. an approximation after time). Adcorp's is an entirely acceptable procedure and used, in fact, by Stats SA itself.

Consider this: In Q4 2011, Stats SA estimated that there were 13,642,530 employed people in South Africa. What Wittenberg and Kerr do not tell us, lest they criticize Stats SA for the same thing they criticize Adcorp, is that this estimate derived from survey responses of 21,137 people. Stats SA applied a scaling factor (weighting) of 645 to get from 21,137 to 13,642,530. The scaling factor varies dramatically each quarter, based on retrospective forecasts of a range of factors.

Moreover, Stats SA revises its estimates after the fact, in some cases going back 70 years, and the revisions generally grow rather than shrink as time passes, when what would be expected is that, as time passes, we should grow more rather than less certain about what occurred in the past.

To criticize Adcorp for using appropriate and efficient forecasts of certain economic magnitudes is to misunderstand the overwhelming mode of statistical estimation. Wittenberg and Kerr would have us believe that they have a monopoly on econometric and statistical understanding. Their poor grasp of econometric procedures suggests otherwise.

4. Number of illegal migrants

Wittenberg and Kerr claim that Adcorp cannot "explain how this figure [2.5 million illegal migrants] was obtained [...]despite this being an important assumption and criterion for the public to judge the quality of the AEI [Adcorp Employment Index]."

In responding to this claim, two considerations merit attention.

Firstly, in answer to the question, the figure comes from the African Centre for Migration and Society, the continent's leading institution for research on migration. It was cited in a Financial Mail article by Claire Bisseker ("In the shadow", 10 November 2011) and is a widely accepted figure. Only the Statistician-General prefers to cite the number of Zimbabweans who successfully registered under the Department of Home Affairs' amnesty programme (126,000), rather than the more widely accepted 2.5 million.

Secondly, as a matter of correct argumentation procedure, it is dishonest that, in an extensive 4-month correspondence with Wittenberg and Kerr, they failed to ask where Adcorp obtained this publicly available and readily verifiable figure, yet in their research note they ask the public to judge the quality of Adcorp's research on the basis of an insinuation that Adcorp cannot provide support for its estimate. This is characteristic of Wittenberg and Kerr's modus operandi, and a disgrace if this is what passes for valid research procedure at the University of Cape Town.

5. Defence of Stats SA figures

An integral part of Wittenberg and Kerr's criticisms of Adcorp's employment estimates is the supposed superiority of Stats SA's estimates. We have addressed the issue of survey respondents' manipulation of field workers' recorded questionnaire responses above. And, indeed, we accept that inadequacies in Stats SA's figures do not necessarily make Adcorp's figures correct or even better. But it may nonetheless be useful to summarize the problems with Stats SA's employment estimates, in the constructive hope that Stats SA will improve its statistical offering to users of its employment data:

• Stats SA cannot provide a single, methodologically consistent estimate of either employment or the unemployment rate for the period 1994 to 2011.
• Stats SA continuously revises its employment estimates and statistical methodology, leading to inexplicable "jumps" in the employment data in the 2004, 2006 and 2008.
• The Statistician-General present Stats SA's employment estimates as reliable and definitive, when the organization's publications are (rightly) much more guarded. By its statisticians' own estimation, Stats SA's statistical procedure is theoretically only capable of stating that employment "increased", "decreased" or "didn't change" with any usable degree of confidence.
• The p-values of the sampling variabilities of Stats SA's employment estimates (measures of statistical significance) range from 0.00 to 0.99. For example, the p-values for changes in the unemployment rates in the Eastern Cape, Northern Cape, KwaZulu-Natal, Mpumalanga and Limpopo - together accounting for 45% of total employment in South Africa - are so high as to make the data meaningless.
• Stats SA continues to emphasize "formal" employment, technically comprising a written employment contract and employer contributions to medical aids and/or pension funds, when these are not the typical characteristics of employees in South Africa.
• Stats SA does not report separately on the increasingly important phenomenon of atypical employment (temporary and agency work), which now accounts for more than 30% of total employment in South Africa.
• Stats SA's estimate of informal sector employment declined from 3.6 million in 2000 (LFS, March 2000, Q4.19) to 2.1 million in 2011 (QLFS, December 2011, P0211), in large part due to the official undercount of the number of illegal immigrants and economic refugees (between 1.5 million and 3.5 million) from neighbouring countries.
• Stats SA forgot to collect data on worker incomes during 2008 and 2009, despite the vital importance of these figures for labour market analysis.
• Stats SA must avoid the challenges with the population census conducted in 2001, on which its population, labour force and employment figures for intervening years (including the hotly disputed informal employment estimates) are ultimately based.
• Stats SA count "underemployed" people in 2007, which dramatically increased and highly politicized the perception that even employed people are, in some sense, "unemployed".
• Stats SA's labour force survey continues to ask unemployed people economically nonsensical questions: whether they were "looking for work" (Q31), would have "liked to work" (Q34), and would "accept a job if offered" (Q39), with no corresponding questions about the conditions (specifically wages and benefits levels) under which they would be prepared to work. As a result, it is impossible to distinguish between the economically meaningful categories of "voluntary" and "involuntary" unemployment.
• Stats SA's estimates of the number of people supported by child support grants (2.8 million in Q4 2011) and welfare grants (0.09 million in Q4 2011) do not correspond to the government's own figures on distributions to households (15.2 million).
• Stats SA's estimates of the number of people receiving unemployment insurance (80,577 in Q4 2011) understate by a factor of 10 the distributions of the Unemployment Insurance Fund (787,000 for 2011/2012).
• Stats SA now records such highly irregular activities as "catching food" (Q59e) and "fetching water" (Q59b) as part of its labour market activities.
• Stats SA recently began to calculate "precarious" employment. The objective basis for this calculation, and the basis on which 1.1 million domestic workers and 105,000 people who work from home were classified as informal workers, was never clarified.
• Stats SA's employment databases are frequently unavailable and beset by availability issues spanning several weeks or months at a time.

I challenge Wittenberg and Kerr to publish their research note in an accredited, peer-reviewed economics journal. I doubt they will be able to do so, for three reasons:

Their selective bibliography ignores previous academic work dealing with this problem in South Africa. For academics, failure to acknowledge other academics' prior contributions is serious enough in itself. Worse, Wittenberg and Kerr deliberately suppressed pertinent and authoritative articles provided to them in the course of our correspondence.

Their research note possesses neither an analytical model nor a statistical procedure. For that reason, their claims are not testable or refutable, even in theory. It is just not good enough for senior academics to make vague hints, when they have no statistical proof and no alternative explanation.

Their research note has no relevance or application to South Africa. For reasons that are well known, South Africa has a unique history. Under apartheid, in order to boost white incomes, blacks were excluded from the most attractive employment opportunities. As the University of Cape Town's Prof Francis Wilson has shown, migration from neighbouring countries was carefully managed by the apartheid government to ensure that local blacks' real wages stagnated during the second half of the 20th century. To a great extent, employment opportunities for blacks were only available in the informal sector. No other country in the world has this history. Yet, absurdly, Wittenberg and Kerr cite OECD research, derived from the world's most highly developed economies, to cast aspersions on Adcorp's methods.

The informal sector is in evidence everywhere except the official employment figures. In the formal recorded economy, companies like Shoprite, Massmart, JD Group, African Bank and Capitec are among the Johannesburg Stock Exchange's highest performers, focused as they are on serving this segment of the population. In the informal unrecorded economy, domestic work, home-brewing, child-minding, beauty and hair care, spaza shops, vehicle repairs, and countless other personal and business services are booming.

Black incomes have been rising sharply - from 15% of the average white income in 2000, to 40% in 2011 - and the number of blacks now exceeds the number of whites in the broad middle class (LSM definition). Yet, according to Stats SA, the number of blacks employed has remained roughly constant over the past decade, around 9 million, and the number of people who are informally employed has fallen from 3.6 million in 2000 to 2.1 million in 2011. Something doesn't add up.

Adcorp stands by its estimates of informal employment, and we fully expect our estimates to be confirmed when the 2011 population census results are released later this year.

Loane Sharp is Adcorp labour economist

Click here to sign up to receive our free daily headline email newsletter