I keep staring at the graphs of vote-share vis-a-vis seats won and trying to gain insights (there is obviously no universal relationship because of post-poll alliances, first-past-the-post system, overall turnout per constituency, etc.)
The data are from different sources. Vote share data for UPA, NDA, and cumulative Others are from 5forty3. Seat count data are from Wikipedia as 5forty3 does not offer alliance-wise seat breakdowns. Curiously the "vote share" data in Wikipedia are different for nearly all elections than what 5forty3 has... I wonder if it is some completely different method for calculating vote share. I believe 5forty3 uses % of all votes cast (not sure about Wikipedia, so I ignored their figures for vote share).
Few things that stand out:
1) 2014 was the first time that NDA actually got a higher vote share than both UPA and Others. NDA has been getting consistently more vote share than Others since 1998, except for a slight shortfall in 2009-- yet it has always had less vote share than UPA until 2014.
2) However, the vote shares seem to translate very differently into seat counts for NDA and UPA. This could be because UPA (INC) always had more cash to turn out more absolute numbers of voters in elections prior to 2014, but not necessarily to do so in more constituencies. On the other hand NDA has in general been able to translate vote share into seats more efficiently (better ground game?)
3) Of concern. Look at the 1998-99 and 1999-2004 deltas. In 98-99, UPA actually INCREASED its vote share but faced a modest decline of 9 seats (144-135). NDA meanwhile DECLINED in vote share (34.7-31.9%) but gained 16 seats, possibly due to alliance factors.
BUT then look at the 1999-2004 deltas. NDA actually declined LESS in vote share during 99-04, compared to 98-99! Yet this decline of 1.1% cost NDA 89 seats. Even more surprising, UPA actually lost (not gained) vote share from 99-04 (a delta of -1.3%) but ended up with 83 more seats. In terms of vote share the only gainer in 99-04 was the Others: they gained 2.4% vote share, though only 6 additional seats.
This tells a very different story than 1989-91 (National Front/Others govt kicked out)... where they lost 6.5% vote share and 82 seats, or the INC/UPA defeats in either 1996 (-4.5% delta vote share, -144 seats) or in 2014 (-12.4% vote share, -204 seats).
To sum up, the rejection of NDA1 by the electorate in 1996
did not look anything at all like the rejection of JD-National Front in 1989, INC in 1996, or UPA in 2014. In all those other cases, there was a considerable decline in both vote share and seats for the rejected incumbent. Compared to this there was little or no anti-incumbency in overall vote-share for the Vajpayee govt in 2004; but the loss of seats for the NDA was catastrophic.
The UPA will definitely try to repeat that this time around. The vote share for Modi is unlikely to change much in 2019 over 2014 (just as it did not change much for Vajpayee between 1999 and 2004). UPA's own vote share will not increase much from 2014 (Pappu has been singularly uninspiring). The UPA's desperate hope is that somehow, even a minor shift of votes to the Others in key constituencies could create room for UPA to helm a khichdi coalition keeping Modi out.
The GOOD news: Modi was able to pull off his 2014 victory with lower vote share than ANY UPA govt in the past... indicating that NDA is simply much more efficient at translating votes into seat count. I simply cannot believe that only 35-40% of voters want Modi back this time around, after all that he has done to win the favour of the masses over the past five years. I anticipate NDA's vote share this time will look much more like the range of UPA's vote shares between 1989 and 2009 (median ~44.1%) if not even higher. Coupled with the NDA's ability to more efficiently translate votes into seat count, this could very well lead to an even bigger victory for NDA in 2019 than 2014.