Some Criticism of CMIE’s Consumer Pyramids Household Survey(CPHS)

Recently we have seen lot of papers and discussions based on CMIE’s Consumer Pyramid Household Surveys. There is lot of debates going on about whether this survey is nationally representative and its divergence with Periodic Labour Force Survey (PLFS). I will try to analyse these issues in this article.

During period 2017–18 to 2021–22, according to CPHS estimates, there was negative employment growth (-0.30%) and -5.62 million in absolute terms, whereas according to PLFS, during the same period employment growth was 4.55% and 88.86 million in absolute term. This mismatch between CPHS and PLFS is sometime attributed to increase in unpaid family labour, that has grown quite considerably during above mentioned period. Now even after discarding this category the mismatch between CPHS and PLFS still exists. CPHS shows negative employment growth whereas PLFS shows employment growth was 3.58% and in absolute terms, it was 56.76 millions, which one can see in the table attached below.

Source: Mahesh Vyas’s Presentation

This divergence between CPHS and PLFS results can be attributed to following two reasons.

1. Sampling Issues in CPHS

In CPHS, multi-stage stratified sampling is used to draw the sample of household. The primary sampling units are villages and the town of 2011 Census. From these primary sampling units, ultimate sampling units i.e. households is drawn. 110 homogeneous regions (HR) are created for broader stratification out of 642 districts on the basis of agro-climatic conditions, urbanization level and female literacy. HRs are divided into rural and urban sub-strata. Each urban portion of HR is further stratified into four strata on the basis of number of households in the town as Very Large (VL), Large (L), Medium (M) and Small (S). From each rural stratum, 25–30 villages are selected on the basis of SRS. In urban stratum, one town from each population based sub-stratum are selected and from each sampled town minimum 21 CEB(Census Enumeration Block) is selected.

From each sampled village, 16 households are selected using systematic sampling. The survey team selects every nth household after entering the main street. Since start is not randomized, there is high chance of over-representation of main-street households in the sample and as a result affluent communities may get over-represented in the sample since it is known that generally affluent households reside in the vicinity of the main street.

Anmol Somanchi , in his paper, has shown that it is indeed the case. He has shown that CPHS estimates of age distribution, literacy level, sex ratios, social composition and ownership of household assets differs significantly from all-india representative survey like NFHS-4 and census-2011. For example he has shown that according to CPHS estimates, 95% of Bihar households had television compared to 35% estimated from NFHS-5. This clearly shows presence of bias towards richer household in CPHS samples.

Another associated problem that arises due to sampling problem in CPHS is over-representation of people with formal education in the sample. Anmol Somanchi has shown that according to CPHS in early 2019, share of adults with no formal education in 15–49 age group was 2% compared to 17% as estimated by PLFS 2018–19. This seems very unrealistic. Das and Roy have observed inverse relationship between the level of education and employment among young people. Because of this, over-representation of people with formal education may leads to overestimation of unemployment.

2. Employment Definition

CPHS considers a person to be employed if the person states that he/she was engaged in an economic activity for a better part of the day of the interview, whereas PLFS considers a person to be employed if the person is engaged in an economic activity for at least one hour in the last 7 days preceding the date of the interview. The CPHS insistence on ‘economic activity for a better part of the day’ excludes large part of economic activity from the definition of employment. For example: by the employment definition of CPHS, a person who is selling pani-puri for 3 hours in the evening, delivery worker who works for 2–3 hours in a day(or gig worker, in general), plumbers (Ex-CEA Dr. Krishnamurthy Subramanian has an upcoming paper on this topic) and different job originated from digitalisation may get excluded. That’s why there is high chance of underestimation of employment using CPHS methodology. Also a person may be unemployed at the day of the survey but there is high chance that he/she may get employment during a period of 7 days. These type of people will be counted in unemployed category in CPHS and that is why CPHS may overestimate unemployment compared to PLFS.

Now when we club these issues i.e. less representation of poor in CPHS samples, over-representation of people with formal education, and exclusion of part-time and gig jobs from the employment criteria, the explanation of divergence between CPHS & PLFS and overestimation of unemployment rate becomes much more clearer.

Hence it will be safe to say that CPHS suffers from above mentioned problems which is leading to divergence between CPHS and PLFS. PLFS estimates of age distribution, literacy level, sex ratios, social composition and ownership of household assets are more closer to truth and they are also in sync with census results. A Srija and Jitender Singh has shown that CPHS unemployment rate has positive correlation with GDP which is not consistent with Okun’s law whereas PLFS unemployment rate has negative correlation with GDP. They have pointed out few more inconsistencies. Their paper is great read. PLFS is nationally representative survey and hence its results are more reliable compared with CPHS.

All comments, criticism and suggestions are welcome and most important, Please point out the error.

References:

1. Anmol Somanchi. “Missing the Poor, Big Time: A Critical Assessment of the Consumer Pyramids Household Survey”. SocArxiv.
2. Panchanan Das, Swarup Roy. Chapter 5: “ Youth Unemployment, Education and Job Training: An Analysis of PLFS Data in India”. Published in: Reimagining Prosperity
3. Mahesh Vyas. “Employment Challenges to Growth”
4. CPHS Sampling Procedure : https://consumerpyramidsdx.cmie.com/kommon/bin/sr.php?kall=wkb
5. Details of Dr.Krishnamurthy Subramanian’s upcoming paper: https://twitter.com/SubramanianKri/status/1652984848481411072
6. A Srija, Jitender Singh. “How Reliable Is Labour Market Data in India? PLFS vs CMIE” EPW