Standard errors: Demystifying the CMIE numbers to explain 'curious' trends

The sample for CMIE's weekly estimates is of the order of 30,000 respondents and that for the monthly estimates is 130,000, with minor variations every week or month

A sharp rise in unemployment rate during the third week of April and then a sharper fall in the fourth week caught the attention of the stylish and sometimes combative TV anchor, Karan Thapar.
 
He quizzed me on his show, ‘UpFront with Karan Thapar’, on April 29 on the zig-zag nature of unemployment rates being put out by CMIE recently. The rate rose during December, January and February. But it dipped sharply in March 2019. It was shooting up during the first three weeks of April, but then it nose-dived last week.
 
I explained that, a few deviations apart, the trend was undeniably one of a rising unemployment rate.

 

The debate after this explanation turned out to be very engaging. I avoid participating in TV debates and Karan was kind to excuse me.
 
Dr Shamika Ravi, Director of Research at Brookings India and member of the Economic Advisory Council to the Prime Minister was dismissive of the CMIE data on unemployment. She stated that she did not take CMIE data seriously. That's her choice. But, I take her several remarks on Karan Thapar's show seriously.
 
A point she made repeatedly was that CMIE did not release the standard error of its estimates. Well, if she was seriously looking for those estimates and did not find them, all she had to do was to ask.

 
So far, only one economist, of Indian Statistical Institute, asked for standard errors of the estimates, and we provided these in March 2019.
 
Dr Ravi said that she would start putting faith in CMIE estimates the day we put out standard errors systematically. We make a beginning here.
 
In the first week of April, the unemployment rate was estimated at 7.91 per cent with a standard error of 0.006195. During the second, third and fourth weeks of April, the unemployment rate and the corresponding standard errors were 8.06 and 0.005309, 8.38 and 0.007408, and 6.65 and 0.005139, respectively. As is evident from these, the standard errors are so low that the estimates are likely to be correct with a probability of 95 per cent up to the first place after decimal. We, therefore, mostly report the unemployment rate up to the first place after the decimal point.

 
The average standard error of our weekly estimates of the unemployment rate is 0.0064. By comparison, standard error of the monthly estimates is much smaller at 0.0032. So, we consider the monthly estimates to be a lot more robust than the weekly estimates. This is no rocket science. It is easy for even an undergraduate statistics student to expect a low standard error if the sample is large. The sample for our weekly estimates is of the order of 30,000 respondents and that for the monthly estimates is 130,000 with minor variations every week or month.
 
Veteran psephologist, accomplished academic and ever-gentle politician, Yogendra Yadav, patiently explained to Dr Ravi on the show that the CMIE sample was large enough to dispel her doubts, but to little avail.

 
The second point Dr Ravi made was that we do not control for seasonality. There are two problems she needs to understand here. First, adjusting for seasonality requires a long time series of monthly estimates. CMIE is the only organisation that produces monthly estimates of unemployment for India and its effort is only 3 years and 4 months old so far. We require at least 5 years of monthly data before we start working on seasonality. Second, the available monthly series was disrupted by the shock of demonetisation.
 
The third criticism by Dr Ravi was that the definitions used by CMIE are 'curious'. She seemed to suggest that CMIE considers people who are not employed, not looking for jobs but desirous of jobs as unemployed. She is completely wrong. CMIE's definition of an unemployed person is one who is not employed, is actively looking for job but is not able to find one. This is what she said is the ILO definition. But, this is the CMIE definition.

 
The fourth point was that CMIE's data are proprietary, where all kinds of puzzling things happen. With the display of her limited knowledge of the CMIE dataset on Thapar's show, it is understandable that she is puzzled. But, is Dr Ravi suggesting that datasets produced by private agencies should not be used? Will she dismiss the surveys conducted by Brookings Institution, similarly?
 
It would have been better for Dr.Ravi to answer Karan Thapar's question to her — on whether the unemployment situation had got worse than it was during UPA or after 2017-18 — rather than displaying her disdain and ignorance of the CMIE datasets.