Shifts in the Indian Labour Market (an ongoing series)

There has been much debate and discussion about the deep shifts that have happened in India’s labour market these last couple of decades. As workers leave agriculture, or newer and younger entrants into the workforce look for non-agricultural work, the key issue is the ability of sectors outside agriculture to absorb the influx.To the extent they cannot, workers are forced into the ‘informal’ sector – largely low-paying, small scale jobs with little security and no benefits, in industries such as retail trade (running or working in, shops) for instance.

It should be pointed out here that 2011-12 has turned out to be a historic year for the Indian economy. It’s the first time that workers in agriculture were less than 50% of the workforce.

It used to be that the biggest employer outside agriculture was the manufacturing sector*, and that is still the case. In addition, sectors such as trade and transport were a big absorber so to speak, of informal labour. But arguably the other big story in the last decade has been the emergence of one other sector as a huge repository of labour outside agriculture and that’s construction.

The chart below shows the extent to which construction is displacing other sectors as an employer outside agriculture – among the biggest employers, it was the only one between 2004-05 and 2011-2012 to increase its share of the non-agricultural workforce. It’s now second only to manufacturing, but has rapidly narrowed the gap.

The emergence of the construction industry as a major employer is not just one of degree but of kind.

The chart below is an interactive one which shows the breakdown of the workforce by gender, region (rural or urban), and conditions of work – whether a worker is salaried and whether he or she has any kind of social security (provident funds, pension, maternity benefits etc). The flowing vertical lines show how each of those ‘dimensions’ (gender, industry etc) are linked together. By hovering over each of the lines, you can figure out for instance, what number of rural males are employed in construction as casual labour. You can vertically reorder the dimensions to change the flow. For instance, hovering over ‘industry’ changes the mouse to a vertical arrow which you can drag up or down. You can open the chart in its own page by clicking here.

The striking characteristic about the construction industry here is its big reliance on casual labour as opposed to the other categories of employment (‘self employment’ and ‘salaried’). A larger chunk of those employed in construction also tend to be those with no social security, even more so than manufacturing.

To be sure, the overall NSS data shows an increase in the proportion of the salaried workforce as compared to other categories between 2005 and 2012. Whether that pattern continues in the future, will partly depend on whether (a) construction continues to expand at the pace it has done so in the last decade and (b) whether conditions of employment in that sector become more or less ‘informal’.


*However, if all the services industry groups (trade, transport etc) are considered as one group, like manufacturing, then services is the biggest employer

‘NR’ under the social security dimension indicates data not available or not reported. Data on social security is only compiled for salaried employees or casual labour, not for the self-employed

All data based on the National Sample Surveys  for 2011-12 and 2004-05. The numbers are estimates by NSS of total populations in each categories. The employment criteria used here is the ‘usual status (ps+ss)’ one

For a big picture look at employment trends these last couple of decades, this paper in the EPW is a good reference

First chart made using R/ggplot2. Second chart uses d3 with Jason Davies’ parsets plugin

Different Ways of Winning

In a previous post, we looked at voting patterns in one constituency in UP – Muzaffarnagar – which, in 2013, saw serious communal riots. The hypothesis was that the riots would polarise voting across the area. To see if this was true, we looked at the vote share of the biggest party (i.e. the one scoring the most votes) in each polling booth. What was apparent was that many,many booths across the constituency essentially turned into winner-take-all contests. In these booths, the largest party ended up with a very high vote share – often in excess of 90%. Whole villages or areas under a single booth chose to vote sharply one way or the other. This was in contrast to 2009, where the average booth saw a much more diverse pattern of voting behaviour.

The map below essentially extends that analysis to the whole of UP. The darker areas saw more polarised voting, and the lighter areas saw less. What we see is a sharp difference between Western UP (where Muzaffarnagar is located), and the rest of the state. Voting in both central and Eastern UP was far less polarised than it was in the West. The white spaces on the map are those for which the data could not be compiled.

This map needs some explanation of how it was constructed, so here goes.

There are over 1.3 lakh data points underlying this map – one for each booth (but with gaps). I could have plotted each point on the map, coloured by the vote share of the largest party. This would have been overkill simply because multiple booths are located at one point (thus overlapping each other on a map). The problem was to aggregate data for booths in the same locality but without aggregating so much as to lose some of the complexity of voting patterns, even within a single constituency.

The technique I used (explained in much greater detail here), was to overlay the map with a number of cells. These are the tiny hexagons you see on the map – just zoom in and they will be more visible. Each of these cells is exactly the same area. I look at the polling booths which fall within the bounds of each cell, and take the median of the vote shares of the largest parties in just those booths. Then we color the cell according to that median value – darker red for higher vote shares and lighter colors for lower vote shares. The grey lines are the actual boundaries of each of the 80 parliamentary constituencies in the state.

The white areas you see on the map are those for which either the locations of polling booths werent available or for which data did not exist. You’ll also notice a number of points which fall outside the boundaries of the state – these are obviously incorrectly located, but I left them in anyway. This post owes a huge debt to Raphael Susewind who actually put the polling booth location data together and cleaned it up.

A couple of broader points :

* There are a couple of outliers in this map. The darker area in central UP is Rae Bareli, the stronghold of the Gandhi family. The lighter coloured area in Western UP is Sambhal, which saw distinctively lower polarised voting than other constituencies in the region. Then there are constituencies like Ghaziabad (to the left of Sambhal on the UP border) which saw both types of voting behaviour – some areas saw a pattern of sharply polarised voting, but other parts of the constituency didn’t.

*The BJP won 71 of 80 seats in UP. It emerged strong across the state, irrespective of geographic location. What this map shows (and this is true of political parties in general I think), is that it had many different ways of winning. In context of an immediate election outcome this may not matter much (winning is winning, after all). But for journalists and researchers, exploring this idea of how a party won in different areas is, it seems to me, an idea very much worth exploring analytically and empirically.

For those already tired of UP election data, and election analysis in general, my apologies. But I may do just one more piece on this idea of ‘ways of winning’.

Here’s a version of the map which is also (a bit more) interactive. Caution : clicking on this, loads a 7 MB csv file, so you have been warned. It may take a minute or two to render, especially on a slow machine and/or internet connection. Mouse over a constituency to see details for that constituency. Unfortunately it’s also this part which is a bit buggy. Sometimes, the mouse over works, and sometimes it doesn’t, for some reason. I am trying to fix it.


Maps rendered using D3, with colors from colorbrewer.

Here’s What a Polarized Vote Looks Like

Before and during the election campaign, one constituency which got a lot of attention was Muzaffarnagar constituency in Uttar Pradesh, the scene of serious communal riots in 2013. Those riots were seen as a ploy by political parties to polarize the vote between Hindus and Muslims. In the event, the BJP won comfortably, by a substantial margin of over 400,000 votes, which translated into a winning margin of 36%. In contrast, in the 2009 election, the BSP candidate who won the seat, did so narrowly – by a margin of just 2.7%.

The chart below shows what that change means in a visual sense. It looks at the vote share of the biggest party in each of the 1500 or so polling stations in the constituency, and then plots those values according to how often they occur. So for instance, there were a relatively few polling stations where the largest party got 40% of the vote or below. There were many more with 60% or more of the vote in a booth. And so on. The number of polling stations where the largest party received any given vote share (from 0 to 1), can be read off the y-axis.

What’s the point here? What I’m trying to show is that in the aftermath of the riots, there were many booths (far more so than in 2009) which essentially seem to have turned into ‘winner-take-all’ contests. Whole villages or areas under a single booth chose to vote sharply one way or the other. Note for instance, that the ‘peak’ of the chart is hit around the high 90% mark i.e. there were close to 200 booths where pretty much every single voter decided to vote for just one party*. In contrast, the 2009 election was much more sharply contested. Here the peak is hit at the less than 50% vote share mark. In most booths in 2009 in the constituency, there was a fair amount of diversity in voting.

In contrast, in 2014, around 30% of booths in the constituency saw the top party receiving 90% or more of the vote.

Another point of comparison is the distribution of vote shares of the largest party for all booths in Uttar Pradesh in 2014. Here it is:

You’ll notice that this distribution is much more ‘conventional’, even given the remarkable scale of the BJP victory (71 seats out of 80) in Uttar Pradesh.

In another post, I may look at another constituency in UP, which had a similarly interesting pattern of voting even though it was not directly affected by the scale of communal violence seen in Muzaffarnagar.

One problem with the analysis above is that I only look at Muzaffarnagar constituency, whereas of course, the violence would have affected voting patterns in other parts of western UP.


*Incidentally, one of the Election Commission’s criteria for classifying a polling booth as ‘sensitive’ (i.e. a booth where extra security measures will be needed etc) is one which gives more than 75% of votes for a single candidate in the previous election. By this criterion alone, a large chunk of Muzaffarnagar’s booths will be classified as ‘sensitive’ in the next election.

In the chart, it seems as if there are some polling stations where the largest party has received a greater than 100% vote share, which is obviously incorrect. The problem is not with the data but with the way the labels are displayed on the x-axis.

Booth-wise data from the dataset compiled by Raphael Susewind from the official Uttar Pradesh data. Many thanks to him! Here’s his website.

Charts made with R/ggplot2.

« older posts