Shoprite’s latest results contained a number that should be worrying statisticians. For the second half of 2025 the supermarket chain reported internal food inflation of 0.7%. Statistics SA’s equivalent measure for food and nonalcoholic beverages was 4.7% for the same period. That is a sevenfold gap that demands an explanation.
It is worth stepping back before diving in. Philosophers of science will tell you that any measurement involves some error from reality — perfect knowledge is impossible. I am an instrumentalist about statistics: what matters is how well they enable good decisions. By that test, official statistics could do better.
The Shoprite discrepancy is the latest in a series of conflicting data claims. Last year private fibre-network operator Fibertime published analysis from its rollout data — drone footage, aerial imaging, AI modelling — arguing that South Africa’s population is closer to 95-million than the 65-million estimated by Stats SA.
GDP measurements have also been challenged, particularly the third quarter 2024 figure, which showed a dramatic fall in agricultural activity, much to the surprise of several agricultural organisations. Unemployment data has also been prominently challenged by Capitec.
These concerns are compounded by the underfunded status of Stats SA itself, which carries a vacancy rate of 20.7%. An agency that cannot fill a fifth of its posts cannot be expected to keep pace with a fast-changing economy.
Pick ’n Pay’s numbers tell a similar story to Shoprite’s: its inflation measure for the six months to end August 2025 was 2.1%, against Stats SA’s 4.6% for the equivalent period. Together, these two chains dominate food retail in South Africa. How is such a discrepancy possible?
Part of the answer lies in methodology. Stats SA tracks a basket of 65 food and nonalcoholic beverage line items, reweighted every five years — most recently in 2023, with a refresh of specific items in January last year. Each month field workers visit supermarkets and read prices off shelves.
As the basket is fixed, it is blind to substitution effects: if chicken prices rise, consumers may switch to another protein, but the basket keeps measuring chicken. By contrast, Shoprite’s measure spans 56,000 items and captures what consumers are actually spending — substitution and all.
There is also a representativeness problem. Stats SA’s basket is constructed from income and expenditure surveys of what people say they buy. Shoprite’s is built from what they actually buy. If I want to know what prices South African consumers are genuinely facing, the Shoprite data is closer to reality — even though it has its own limitations, including that higher-spending, wealthier consumers have a disproportionate weight in total expenditure figures.
But none of this fully explains why the Shoprite and Stats SA measures — which tracked each other closely for years, with Shoprite consistently running about a percentage point lower — diverged so sharply in the most recent period. That gap widened dramatically in the second half of 2025.
A plausible hypothesis is that the major retailers, under sustained consumer pressure after several years of elevated inflation, have been running unusually aggressive promotional activity and absorbing margin to hold or cut prices. If so, a fixed-basket price-reading methodology would not fully capture this, whereas actual transaction data would. But that remains to be tested. Stats SA and the retailers should be working together to find out.
This is not without precedent. When Capitec published an analysis suggesting that official unemployment data was significantly undercounting the size of the informal sector, the result was a productive engagement between the bank and Stats SA that led to changes in the Quarterly Labour Force questionnaire and meaningful improvements in how informal market activity is measured.
Capitec’s data had its own weaknesses — it cannot cleanly separate remittances from employment income — but the engagement produced a better system. That is exactly the model to follow here.
The deeper issue is whether we can build something structurally better. Modern technology has opened pathways to inflation measurement that were unavailable when current methodologies were designed. Shoprite, Pick ’n Pay and Spar could provide real-time scanner data to Stats SA via API — actual transaction volumes and prices across millions of purchases, updated continuously, rather than a fieldworker reading a price tag once a month.
The UK’s Office for National Statistics has been experimenting with supermarket scanner data precisely for this reason. The obstacle in South Africa is not primarily technical; it is a combination of data sharing agreements, institutional capacity at a resource-constrained Stats SA and, frankly, political will. But such approaches could also lower data collection costs and create financial capacity.
Food and nonalcoholic beverages constitutes 18.2% of the consumer price index basket. A near-four percentage point error in that component does not stay contained — it flows directly into the headline inflation number, which in turn influences Reserve Bank interest rate decisions, government expenditure allocations and wage negotiations across the economy. The cost of getting this wrong is not abstract.
We no longer live in a world that should tolerate five-yearly basket updates and monthly price readings from supermarket shelves. Stats SA deserves adequate funding and the institutional mandate to modernise. As the primary users of this data the Reserve Bank and National Treasury have every reason to push for it.
• Dr Theobald is founder and chair of research-led consultancy Krutham.
Would you like to comment on this article?
Sign up (it's quick and free) or sign in now.
Please read our Comment Policy before commenting.