Beyond the obscurity of the allegations that Statistics SA made an error or delayed the consumer price index (CPI) re-weighting, which, by and large, has been put to bed, there is a more fundamental underlying issue: to what extent can an official statistical agency live with a bias in its measure of inflation?

The CPI measures the inflation rate based on the average household. This averaging out is what is often forgotten. No single household has the same inflation rate, nor do different income groups.

The 2005-06 income and expenditure survey, released in March this year, for example, showed that for the first income quintile (the poorest 20 percent of households), food constitutes close to 40 percent of total expenditure, while for the richest 20 percent of households, food constitutes less that 10 percent of total expenditure.

Assuming food inflation is higher than nonfood inflation, and keeping everything else constant, the inflation rate of lower-income households will be much higher than high-income households since the former spend more of their total expenditure on food. Naturally, the converse will hold if food prices go down. So it is not unreasonable to suggest that the current average inflation rate has a downward bias for low-income households and perhaps an upward bias for high-income households. The point is that the idea of bias should not been seen as an error, but intrinsic to the practical measurement of inflation.

Is our measure of inflation accurate?

Naturally, the one issue that has accentuated the timing of the release of the new weights is the concern with the high inflation environment, and whether there would have been less of an issue if we were in a lower inflation situation. A very reasonable question would be: is Stats SA being irresponsible by not fast tracking the new weights (and rebasing) in view of the new data it has at its disposal? Does it have an obligation to act immediately because current price movements show that the inflation level roughly could be lower?

A more complete question would be this: if the economic environment went in a different direction, in which the current CPI underestimated the inflation rate relative to the new weights, what should we do? Should we also act and fast-track so that all those who negotiate wage increases are not short changed?

The answer is no. One of the key principles of good statistical practice is to pre-announce and plan a schedule based on decisions about quality with a balance to timeliness, whatever the economic environment. This is our underlying philosophy and conforms with best international statistical compliance and practice.

The issue of bias in the CPI was elevated to a global debate when, in the mid-1990s, the US statistical agency came under scrutiny on the basis of various studies, prompted primarily by academics, that the inflation rate in that country was exaggerated. The Boskin commission in the US in 1995 concluded that the US CPI overstated inflation by about 1.1 percentage points per year in 1996 and about 1.3 percentage points before 1996. While this may be a perfectly reasonable finding, the practical organisation of managing a CPI in an official statistical agency implies that we cannot have a current-weighted index, and a bias will always exist.

Naturally, many of these findings have been taken on board to minimise bias, but they are done in a systematic fashion and often after a serious review of what is possible or not under the constraints of managing a large month-to-month CPI release.

Unfortunately, all CPIs are unavoidably affected by biases and whereas some are upward, there are just as many downward biases that offset and even overshadow the former. The responsibility of a government statistical office is to keep them from getting significantly out of balance.

There are several ways in which this is done. It is important not to single out one bias to the exclusion of others. For example, one could make the argument that by not introducing quality changes into the CPI, some bias could arise in either direction. In fact, some of our approach to motor cars and furniture, for example, has a downward bias. In the end, there is a contradictory bias. Similarly, as mentioned earlier, had price movements taken a different course, Stats SA would have been in the situation that inflation would have had a downward bias by delaying, so to speak, the re weighting for a year.

Other sources of bias are well known in the literature. For example, a common concern with a fixed-weighted CPI (weights remain static until a new re-weighting takes place) is substitution bias, where consumers substitute relatively less expensive goods for more expensive goods when relative prices (the price of one good relative to another) change.

Others include bias arising from which outlets are surveyed, assuming that there is a bias towards either lower- or higher priced outlets. Bias also occurs when new products are not introduced. In the new CPI, new products have a 25 percent weight and the prices for these new products have only been collected since January.

More frequent surveys

The 2005-06 survey was done in greater detail than in the past and conducted over a whole year to capture seasonality and a full profile of expenditure patterns throughout the year (unlike 2000 and previous ones that were simply conducted in a month). The scale of changes meant, on the basis of international peer review, that Stats SA has to at best conduct a parallel survey to have a year-to year comparison that is economically meaningful.

It is practically very difficult to have more than a five-year frequency of the current scale (currently close to 25000 households are surveyed). But there is another question. Could we live with a narrow CPI based on metropolitan SA requiring a smaller sample of households in the order of 2000 to 3000, which would lead to more frequent re-weighting, or should we continue our five-year cycle and maintain the current structure and form that attempts to capture expenditure patterns across the country, rural and urban as well as the provinces? Either will create a bias.

Rashad Cassim is deputy director-general, economic statistics, at Stats SA.

Business Day