The Guardian’s data blog has run a piece on the ILO’s latest global unemployment statistics here, which gives me all the opportunity I need for a minor rant about the comparability of employment figures across the world and to issue a warning about statistics more generally.
The primary problem for employment statistics isn’t actually the methodology of data collection, or the fact that different definitions of employment, labour force or unemployment are used. The problem is rather that the *same* definitions are used, even where they may not capture how employment and unemployment actually work.
Basically, all statistics involves the creation of synthetic categories in order to generate measurable scales with defined boundaries. This has a lot of uses – it allows us to measure things over time and over space, which has obvious benefits, which are exploited in pretty much every intellectual discipline in the world. The trade-off is that they have to shoehorn complex, moving phenomena into clear categories. Depending on the specific statistic under question, this loss of complexity can be very damaging (in this category I would put most indices of ethnolinguistic fragmentation) or very minor (GDP statistics, which can be adjusted to retain virtually all of the original complexity). In labour statistics, the categories don’t lose much complexity in some kinds of economy and lose a lot in others.
More concretely, employment statistics basically measure the size of the labour force, the number of people in employment and the number of people who are unemployed. Each of these categories is defined according to a set of behavioural rules. Problems arise when these rules, kept constant in order to allow us to compare across space, cease to represent the world they are measuring. Take the definition of the labour force. In South Africa, the labour force is defined as everyone between 15 and 65, with exceptions including housewives, students, those unable to work due to illness or disability, those not seeking work, and those in early retirement. The exception of housewives from this category causes difficulty; it captures the wife of the relatively prosperous worker with secure wage employment, what might be called a ‘developed-world’ housewife. However, it also covers the wife of a rural or urban wage labourer with insecure employment, who may have to engage in a number of activities to supplement family income. The latter person is actually part of the labour force, albeit one who moves in and out of employment rather fluidly.
The ILO standard definition of the unemployed is also problematic for similar reasons. Two of the central planks of the definition are that the person must be characterised as ‘being without work’ and secondly that he or she must be ‘actively seeking work’. Both parts of this are especially problematic in Sub-Saharan Africa. Firstly, the condition of ‘being without work’ discriminates against women who are involved in economically productive activities. To be useful in making gender distinctions, labour statistics must cover all the activities carried out to produce goods and services. Women in Africa are very likely to combine economic activities with ‘non-economic’, household activities, and thus work intermittently over the year, or to work from home in either own-account or waged occupations – massively increasing the likelihood of being under-represented as employed labour. What’s more, in many countries, economically productive activities conducted by women (which may be crucial inputs to smallholder agriculture or home-based goods production) are classed as ‘housework’ and not an input into production. Men carrying out separate parts of the same production process are counted as working, self-employed.
Another, equally large problem with the idea of unemployment exists in much of Sub-Saharan Africa: the condition of being completely without work hardly ever exists for people of working age without disability. State safety nets rarely exist, and social safety nets are massively over-stretched by the elderly and very young. Able-bodied people who do not have regularised employment therefore spend much of their time doing whatever they can to get by: washing cars stopped at traffic lights, offering to show tourists around (or just pestering tourists, sometimes), selling whatever small items they can cook or make. When the time comes for the employment survey, few such people would described themselves as having been employed, but though marginal, they have been involved in economic activities. A much bigger problem than unemployment in Africa is, therefore, underemployment. For this reason, from an economic perspective, it makes far more sense to measure the labour supply (i.e. the man hours of fully committed work available) and to compare this to the labour demand, which will give us a much better sense of how far the economy is to utilising its labour capacity.
This brings us to the condition of ‘looking for work’. This again has difficulties when applied to Africa. Typically, ‘looking for work’ is defined as applying for advertised jobs, going to employment exchanges and the like. Yet the vast majority of jobsearchers in Africa have a completely different modus operandi. Particularly for workers in rural or peri-urban environments, job exchanges may be too far to travel too, and advertised jobs few and far between. For them, ‘actively seeking work’ means asking around among friends and acquaintances, visiting likely places or simply hoping for the good luck of meeting someone who can offer work, making their availability especially for piecework openly known. If they know that there are no jobs openly available, are they any less unemployed if they are not asking every day? It may not be the best use of their time.
I’m not saying that employment statistics are useless. I am saying that the behavioural norms that have been used to set the boundaries of definition fit the developed-world reality of employment, economy and society far better than they do African realities. What this means is that as snapshot data for employment in Africa is a very distorted reflection of the reality of what we are trying to measure. It has more value as a time series, once we acknowledge these issues; and as economies become more modern, these issues are reduced.
This is a general issue with statistics. Statistics are absolutely crucial for having any real understanding of the world except in the most specialist, detailed way. They are incredibly important to any attempt to understand how a number of places and times may be linked or how they may respond to similar stimuli. However, every form of statistics is based on a very specific way of dividing the world up so we can measure it. This does not make them all weak or useless, but it makes it necessary that we understand the conceptual basis of statistical indices before we begin applying them anywhere. I’ve given the example of employment statistics because these are a statistical index I’m familiar with. The lesson generalises. Anyone using, for example, an index of ethnolinguistic fragmentation must first know exactly what norms are used to divide different ethnicities and languages, how these are weighted, and how the same distinctions in different countries can mean different things. Otherwise they cannot really know what they are saying about the findings they come up with from the data.
So, maybe David Caruso is right. The evidence doesn’t lie. But we can mislead ourselves with it quite easily.