Your complete guide to analysing National Sample Survey (NSSO) data

Welcome!

Without wasting any of your precious time, we will take you through the process in steps, utilising India's 68th round NSSO as an example. All your queries will be answered within 3 days!

STEP-1 : Understanding the sampling methodology and calculation of sampling weights

Here, we will illustrate the methodology for sub samplewise estimates in rural sector for the consumer expenditure survey of the 68th round of NSSO (2012-13). Let's say we want to see the data for Meghalaya as in Fig 1.

Meghalaya has several districts as can be seen in the map below in Fig-2.

NSSO  has a concept of  NSS State-Region:  An NSS state-region is a contiguous group of districts within a State having similar topography, agro-economic characteristics and population densities. For bigger States, the number of regions goes up to 7 while for smaller States/UTs, there is only one region. The regions have some distinctive geographical features and climatic conditions and this makes the regional estimates more meaningful and useful in some respects. The boundaries of a region generally do not cut across district boundaries or State boundaries.A region may comprise a single district for small States/UTs but for larger States, there may be 10-12 districts in a region.
In our example above, the entire State of Meghalaya has been assigned as only one region viz. NSS-SR= 171. Check out Fig-5 below,

Central Sample & State Samples: In most NSS rounds, survey for the State sample is done by the respective State Govts, while for the Central sample it is done by NSSO.  State samples & Central samples are drawn separately or allocated separately.

For rural India, 59,695 households (SSU, explained later) were selected from 7, 469 villages (PSU,explained later) selected from the country were sample surveyed under the Central sample. Subsequently we will discuss only about central samples.

Schedule Type 1 & Type 2 : Type 1 questionnaires  based upon consumption during last 30 days & last 365 days, while Type 2 questionnaries based upon last 365 days (infrequently purchased categories), last 7 days (some food categories) & last 30 days for (other food, fuel, etc)
Schedule Type 1 & Schedule Type 2 were canvassed in 2 independent samples of matching size drawn from each stratum/sub-stratum(same FSUs/Villages).Thus, under Schedule Type 1- 59,695 households were surveyed while under Schedule Type 2 - 59,683 households were  surveyed separately  from the same 7,469 villages under Central sample. We will discuss about Type-1 samples only subsequently.

Stratum: Districts within the NSS-SR are taken as the primary strata. However this is also broken into 2 sectors viz. Urban Sector & Rural Sector. We  will discuss only about the rural sector in this example. In our example, we have 7 strata under NSS-SR rural sector , each of them is a distinct district with their geographical boundary.

Sub-Stratum: If the number of  households of a district is large, then it is sub-divided into two or more sub-strata of nearly equal households by grouping together contiguous groups of villages having similar socio-economic characteristics.
In our example, the villages has been grouped in West Garo Hills district (Stratum=1, Distcd=1)  into 6 sub-strata, in such a  way that there are more  or less equal number of households in each such group. Check out Fig-5 table above and Fig-3 below as an conceptual illustration,

FSU(First Stage Units): These are  villages in the sub-stratum for the rural sector. Here, the Sampling Frame is the  list of villages under Census 2011.
USU/SSU(Ultimate/Second Stage Units): These are households available for survey within the village Here, Sampling Frame is all the households available for survey not all the households in the village (please note the distinction, it is important!)

Sub-Rounds: The period of survey is one year duration starting 1st July and ending 30 June (India's Agricultural Year). This survey period is  broken into 4 subrounds of 3 months duration as below,

Sub-round 1 : July-Sept

Sub-round 2 : Oct-Dec

Sub-round 3 : Jan-Mar

Sub-round 4 : Apr-Jun

Concept of Interpenetrating Sub-samples:  The method consists in drawing samples of FSUs/Villages in the form of 2 villages @8HH per  village,thus 16 HHs as ONE subsample and ANOTHER subsample drawn independently in a similar manner . Thus, we have  2 SUBSAMPLES IN EACH SUBSTRATUM. Checkout Fig-6 table below.
The main advantage is that the Relative Standard Error of Estimates (RSE) can be easily found out in this method, even when the sample design is complex.

Second Stage Stratum(SSS): Each household available for survey in the selected FSU/Village is categorised IN EITHER OF THE THREE categories, as the following,illustrated conceptually in Fig-4 below,

SSS 1 Relatively affluent
SSS 2 Of the remaining, having principal earning from  non-agricultural sources
SSS 3 All other remaining

Generally 2 households from SSS1 , 4 households from SSS2 & 2 households from SSS3 are selected from the household available for survey sample  frame  IN THE FSU/VILLAGE as  explained above(If some categories are not found  then it is made up as per priority rules ensuring ALWAYS that in total  8 households are selected from each FSU/Village)

Fig-7 illustrates the case for only 4 villages for your understanding,

Now we can show you how the sampling weights are calculated !

The meaning of the notations used in NSSO are as below,

Go back to Fig-6 

Can you try these now? Find out the relevant formula and check whether it matches the
calculated weight as given on left! If you can then you  have definitely  understood the 
sampling plan and  the concept of multipliers/weights in NSSO Unit Level data.

For the first row it is worked out for you as below for calculation of the multiplier for this household, See the above complete working in an Excel file here!  => ($V$2/4)*(1/$BG$15)*(BG29/COUNTIF(AR2:AR9,1))


Congratulations!

Thank you I hope you enjoyed understanding the apparently complex stuff  as much as I enjoyed in preparing its explanation!

Next, we will look at how to decipher and analyse NSSO data for a given problem.

So for now, good-bye and good luck!.