In a pointed analysis on Scroll.in using data from the National Crime Records Bureau (NCRB), Shivam Vij shows that the rate of rioting in India (per 100,000 population) has fallen continuously since 1992, reaching a plateau around 2003 and remaining pretty much constant since then. Vij is correct that the Manmohan Singh years have been the most peaceful on average since 1963, but it is also true that this level was first reached in Atal Bihari Vajpayee’s final year in office.
The Scroll report excludes data for 2012 and 2013, but these are available in the NCRB’s annual Crime in India 2012 (PDF) and Crime in India 2013 (PDF) publications. Incorporating data from those years shows an uptick in riots in 2012, taking the number above the 6 riots per 100,000 population mark, before it drops back down below this level in 2013. This doesn’t amount to anything more than statistical noise, particularly in the context of the larger trends the report highlights.
A couple of points are worth mentioning. One is that riots here are defined as incidents that are in violation of sections 143-145, 147-151, 153, 153A, 153B, 157, 158, or 160 of the Indian Penal Code. This is different from the Home Ministry’s definition of “communal incidents” that I used in BJP-ruled states more communally violent. To take one example, the NCRB lists 72,126 riots in India in 2013, while the Home Ministry offers a figure of 823 communal incidents for the same year. It is possible that the smaller number is a subset of the larger one (not all riots have a “communal” angle), but we don’t know that for sure.
More importantly, it is also possible that the number of riots is being undercounted. As Rukmini Shrivinasan points out, NCRB data only record the “principal offence” reported in a police first information report i.e. the offence that attracts the maximum punishment. This means that a riot that resulted in a death could be counted as a murder or attempted murder instead. This is common international practice (as NCRB deputy director R Rajasekaran has written), and even in the US only a quarter of all incidents are disaggregated by types of offence (under the National Incident Based Reporting System). There is a programme underway to automate and network crime data collection co-ordinated by the NCRB but when this will happen is anybody’s guess.
Chunauti 
The point about undercounting is useful, but there is also, I believe, a tendency to overcount — if what we are trying to understand is communal riots. There is a definitional issue, and an operational issue. The definition of “riot” reflected — from the 19th century — in the IPC extends from what I would call a “brawl” (where “five or more people” beat someone up, for example, or two gangs fight) to the much larger — in terms of people involved and geography — “riots” over, say, food (in the 19th century). The operational issue is that when there is such a disturbance — and getting more than five people involved in India is pretty easy — and the police arrest and charge them, they are likely to prefer the most serious charge (and all charges) that might apply, with the understanding that at a later stage, the more serious charges that can’t be proved will be dropped from the charge sheet. It is a sort of bargaining situation, where a maximum charge is preferred in hopes of getting at least some charge sustained. Thus if the police break up a fight — all of these IPC sections are under “Offences against the public tranquility” — they are likely to use the “riot” parts of the IPC, which are vague enough to fit many cases involving violence, and destruction of property, etc. In my small experience of looking at Indian crime data at local and state levels as well, the number of “riots” with a “communal motive” were indeed a very small part of the whole. I would stick to the Home Ministry report.