Notes from the Fringes of Academia
On Censuses
Both regions that I operate out of professionally are in the middle of census sagas. In South Africa, the census 2022 data has not been received well. With an estimated undercount of 31 percent—the highest ever recorded for a national census outside of a war zone—it has gone beyond a simple margin-of-error problem, and as per usual, folks are not happy. In India, where Census 2021 was “skipped” due to the COVID-19 pandemic, things are warming up for a (almost-)mid-decade census, with notifacations etc. being gazetted for 2027. And in the classic neo-Indian model, folks are suddenly up-in-arms trying to get census to change what they feel are outdated definitions—e.g., for “urban” to reflect the “true extent of urbanisation in India” and “how we need to show the world that we are much more urban”.
In between all this noise, what often gets lost are the mechanisms of conducting a census, what can be done during different stages of the data pipeline, and by whom.
First, what is the general process followed during any given census?
[Paraphrasing from some older writing] The legal process for conducting a census is often based on specific legislative authority that establishes the procedures, scope, and protections for the estimation of national population counts. Governments notify censuses and appoint census commissioners to guarantee administrative oversight, legal compliance, data confidentiality, and effective public awareness.
The role of a census commissioner is typically time-bound and specifically tied to the cycle of a particular census. The commissioner is appointed for the duration of the census operation—from planning and execution to the completion of data collection and initial processing. Their powers and responsibilities are activated only for the period relevant to that census and generally cease once the census is officially concluded. While they could be housed within an independent agency, the commissioner position itself is specific to a census.
In most countries, this period spans several months, even years, depending on the scale and complexity of the census. However, the time-bound nature of the role ensures that the commissioner has focused authority and accountability for a particular census. Such a structure is supposed to maintain efficiency, transparency, and public trust in the census process. It is also supposed to allow for flexibility in appointing individuals with the necessary expertise and experience for each census, adapting to changing needs and contexts.
The role of the commissioner ends once the census is officially completed. This end is signalled when the final report is submitted to—and accepted by—the relevant government authority. After this, the commissioner’s powers are no longer active, and any further work (such as detailed data analysis or dissemination) is handled by the NSO or other permanent bodies.
Practically, what this implies is that once the government authority accepts the census report, the role of the census commissioner ends, and ‘raw’ data is archived. All analyses, thereafter, can happen only with the anonymised data. Issues with the ‘raw’ data need to be flagged before the report is accepted—there is enough precendence for this globally (e.g., India in the 1990 and 2000 cycles, US in the 2000 cycle, Canada in the 2010 cycle, Zambia in the 2020 cycle).
Why is the ‘raw’ data archived? Because it was within the legal role of the commissioner to ensure that data privacy laws were not violated. Once the commissioner hands in the report, the report is accepted, and the office of that commissioner is terminated, the ‘raw’ data is legally locked down. Even the NSO cannot technically access that data.
After the enumeration data is ready, reports accepted by the relevant government authorities, and anonymised data is handed over to NSOs or suchlike, processes for data dissemination are triggered. It is at this stage that “definitions” are applied. Tables are generated. Even imputations are made.
Now within this framework, where do the ongoing public discourse in SA and India fit in?
Let’s begin with SA, and let’s account for the complexities that played out during Census 2022.
To begin with, the decision to conduct a census in the middle of a raging pandemic. Detractors point out that one of the main reasons that the census went ahead was because the Treasury in all their wisdom decided that if the NSO didn’t use the budget allocated for the census, they would lose it. This points to the really problematic social state that SA finds itself in at the moment, where observable social sector spending competes with (often invisible) spending on the backend infrastructures that empower said spending. Leading to a chicken-and-egg situation, where in the current intercensal period, people are going cray-cray on how social sector spending will be calculated/budgeted since the backend infrastructure was weakened.
Anyhoo, the census had to happen. And it did. And as expected, this resulted in a hurried implementation, using untested digital solutions to speed up the enumeration process, without actually planning for possible eventualities associated with such solutions. And once the numbers starting getting tallied, suddenly the massive undercounts started emerging, and, as can be expected, the blamegames and mudslinging started.
But by then the government authority had already accepted the census report, and consequently, in line with established global practices, closed the chapter on the ‘raw’ data.
Yet the public discourse continued/continues. Certain groups of people are up in arms saying that the NSO effed up, adding (in a rather abrasive manner) that the data is “not fit for purpose”, and suggesting that the NSO give them access to the ‘raw’ data (that the NSO technically can’t do) so that they can “help” fix the imputation models that the NSO used. Obviously the NSO has closed ranks. No (public) institution likes to be at the receiving end of pointed fingers.
Yes, there is no denying that has been an institutional failure. But is this current equilibrium ideal? One side shouting, creating a ruckus, talking down to the very institutions to which they owe their bread-and-butter. Other side ignoring them and continuing with their use of data of sub-optimal quality. Who suffers in the end?
Could things have been done differently? Perhaps yes. Instead of forcing a field count in hostile conditions, SA could have paused the full enumeration for this particular census round and taking a leaf out of countries where the door-to-door enumeration approach have been limited, maybe relied on a strengthened mid-year population model, calibrated against social grant databases, population registers, and the electoral roll. But such solutions would have required planning, which was completely cut short by the institutional setup that prioritises “visible” spending.
The Indian story is another situation where those in the public space appear to be unaware of the census processes.
Let’s take the census definition of the “urban” as an illustrative example. First, let’s look at how settlements are defined as urban and rural in India.
There are two categories of urban classification: first, the statutory definition, where legislative bodies decide whether a settlement is urban or not. There is no statistical criteria for this decision. Usually this is determined by whether an existing settlement has achieved (an undefined) “urban-like” characteristics.
Second, there’s the statistical definition, which has three components: size, density, and labour characteristic.
Within the process of the census, do these definitions affect the enumeration process? Only the statutory definition does—at the planning stage, where the office of the commissioner develops the enumeration strategy.
Where does the statistical definition come in? Only after the census, once all enumerated data is tallied, those settlements that are not statutorily identified as urban, but satisfy the statistical definitional criteria get tagged as “census towns”.
Do statistically defined urban settlements get shortlisted by legislative bodies in the subsequent intercensal period and get notified as a statutory town? Short answer, no. There is some evidence that when statistical towns are located in the immediate periphery of a statutory town, they get notified as statutory towns, or get subsumed within the boundaries of the statutory town—but none of this is causal. At best, it could be argued that when a statistical town is in the path of the natural growth trajectory of the statutory town, it gets picked up. Often in clusters—few such statistical towns get combined to form a new statutory town. But still no causal linkage. And numerous studies also point to the “choice” factor, where settlements which have become statistically urban actively trying to stay out of statutory recognition, because they would lose their freedom to access government funds as well as having a say on their own development strategy by means of dilution of democratic power.
…and all this means that all the noise about “true level of urbanisation” is just an academic exercise. Knowing how many settlements have statistically gained urban characteristics does not move the needle of the statutory process. It’s just good for a few papers.
Additionally, another new trend emerging in India is “petitioning” the census to make their data dissemination more modern. Recent voices have been arguing that the census “[…] could adopt a spatial approach through an open-access geo-statistical portal [for data dissemination].”
Great, yes, should be done, modernising data dissemination and increasing access is always good. But the census enumeration process is already highly “spatial”. Right from defining enumeration areas, to reporting, everything is spatial. (And this is not a recent phenomenon—the census and national archives have the hand-drawn maps all the way back to the censuses from early 1900s!!!)
The problem lies in how the data is processed by end-users.
[Aside: another rant is due for how researchers don’t/can’t clean data anymore, everyone expects someone “lower” than them in the pecking order to clean data for them.]
Consider the administrative atlas series that are published every census. If properly digitised, they are a readymade skeleton to which all census data tables can be mapped. Even the digitisation has already been done for the censuses post-1990 by certain government agencies. No new portal is actually needed, at best collaborations to validate said data. The current noise for the census to build a completely new portal is just another ploy for the academic/research community to do the so-called intellectual work without putting in the grind.
So what was the point of this long post? Essentially that we as the so-called research community are either unaware, or perhaps more importantly unappreciative, of the complexities that the state-run institutions face when generating data for us to use. To borrow a phrase, we seem to be very keen to find straw-men (usually the NSOs) to fuel our need to be critical of the state apparatus—without actually coming up with workable solutions. All we think of are papers. The currency of our being.
Maybe that’s why all those (big/private) consultants thrive in the state apparatus. They provide half-baked solutions, that will need continuous refinement, thus guaranteeing a revenue generating pipeline for themselves.