The results were quite interesting. It was found that the two sets of data obtained by questioning differed quite sharply, with the consumption estimates obtained on the basis of the one week reference period being much higher than those obtained on the basis of one month recall. It was also found that the one month reference period generated information that corresponded much more closely to the data on the basis of actual weighing of food items. This led Mahalanobis and others to conclude that the one month reference period was better suited to the purposes of the estimates especially of food consumption.
 
Since then, it has been standard for the National Sample Survey to use the one month reference period for food items, although both the one month and one year reference period have been used for some non-food items. In the five quinquennial surveys of household consumption expenditure between 1972-73 and 1993-94, information for clothing, footwear and durable goods was collected from each sample household for two reference periods - "last 30 days" and "last 365 days". In the 50th round, "educational" and "institutional medical" expenses were also added to the list of items for which data were collected by these two reference periods.
 
However, during the 1990s, the question of the most suitable reference period has resurfaced, and the NSS has in its thin samples experimented on the basis of alternative schedules (based on one week and one month recall) in independent samples during the course of the same survey. This was done for the 51st Round (1994-95), the 52nd Round (1995-96), the 53rd Round (1997) and the 54th Round (Jan-Jun 1998).
 
In all of these surveys, half the sample was canvassed using a Type 1 schedule which had a 30 day reference period for all items, but the other half was canvassed using a Type 2 schedule which had a one week reference period for food, pan, tobacco and intoxicants and a one year reference period for clothing, footwear, durable goods and educational and medical (institutional) services. However, since the Type 2 schedule was not comparable to earlier NSS surveys, the results by this schedule were not tabulated in the NSS Reports of the relevant rounds, so that all the available analysis of consumer expenditure and of poverty during the nineties are based on the Type 1 schedule.
 
It is now reported that for the large sample of 1999-2000, this experiment has been carried one step further, to the point where the two different schedules of Type 1 (one month) and Type 2 (one week) have been canvassed from every sample household. To the extent that the two schedules give varying results, incorporating them so that all the households respond to both schedules is obviously problematic since it would bias the results of both schedules. It also means that there are definite problems of comparability with data from past surveys, for which the reference period was essentially one month and one week was not used at all.
 
Fortunately, the NSS has now released the preliminary results on the effect of the choice of reference period for the 51st, 52nd, 53rd and 54th Rounds in its Report No. 447: Choice of Reference Period for Consumption Data. It is, therefore, possible to examine the effect of choosing one reference period over another. As mentioned, above, in all of these Rounds, one half of the sample (Type 1) had a reference period of one month (30 days) throughout; for the other half (Type 2) the reference periods were as follows : one week for all food, pan, tobacco and intoxicants; one month for fuel and light and miscellaneous goods and services; and one year for clothing, footwear and durable goods as well as education and institutional medical expenses.
 
The results of these samples based on alternative schedules are extremely interesting. It emerges that Type 2 samples give higher overall food consumption, exactly as Mahalanobis had predicted and as was also confirmed by the pilot investigation in the West Bengal villages in 1952. Thus, the Type 2 schedule-based samples also suggest that poverty is much lower.
 
In Charts 1A and 1B, the cumulative distribution of population below specified per capita total consumption levels as obtained from the Type 1 and Type 2 schedules are plotted for the rural and urban sectors, using data from the 52nd round. It may be seen that the proportion of population below any expenditure level is always higher by the Type 1 schedule than by Type 2, and the difference is extremely large. Thus, using the Planning Commission poverty line, about 39 per cent of the rural population would be below the poverty line in 1995-96 by the Type 1 schedule but this percentage would be only around 20 per cent by the Type 2 schedule. The corresponding percentages for urban areas are 30 and 15. Similarly large differences are obtained for the 51st, 53rd and 54th rounds.
Chart 1a >> Chart 1b >>
 
This of course raises the question about which of these two schedules gives a better measure of the actual incidence of poverty in India. But an even more important problem is the implication that the 55th round may end up giving a totally biased picture of poverty trends during the 1990s if the results from this are compared to earlier rounds which did not use the "one week" reference period.
 
Thus, suppose, for example, the 55th Round comes up with an estimate of 25 per cent of rural population below the poverty line in 1999-2000. This would be compared to the corresponding 37 percent rural poverty obtained from the 50th Round in 1993-94, and the implied large reduction in poverty would be greeted by the reformers as vindication of their policies. But, since this does not compare like with like, such a conclusion would obviously be erroneous and the debate about trends in poverty would enter a statistical minefield which the 55th Round results will not be able to resolve.